Tag Archive | fact-checking

Building a Sustainable Workflow

By Cheval John | Vallano Media
This article was created with the assistance of artificial intelligence.
The final version was reviewed, edited and fact-checked by the author


Graphics created with AI-assisted design tools.
Visual concepts, editorial direction, composition, and final creative decisions were developed by Vallano Media.

A workflow can look great on paper.

Research.

Draft.

Verify.

Red-team.

Edit.

Quality control.

Publish.

Clean.

Logical.

Complete.

But there is another question that matters just as much:

“Can I keep doing it?”

A good workflow is not simply one that works once.

It is one that still works when I am tired.

Busy.

Under deadline.

Covering multiple projects.

Learning a new tool.

Fixing a mistake.

Or dealing with something I did not plan for.

That is where sustainability begins.

A Workflow Has to Survive Real Life

The easiest workflow to build is the ideal workflow.

The harder one is the workflow that works when the day is messy.

A sustainable system has to survive interruptions.

Unexpected reporting.

Technical problems.

A slow source response.

A graphic that needs to be redone.

A deadline that moves.

A story that becomes more complicated than expected.

If the workflow only works under perfect conditions, it is not really a workflow.

It is a demonstration.

I need something I can use again tomorrow.

Repeatability Matters More Than Complexity

There is a temptation to make workflows more complicated as the tools become more capable.

Add another AI role.

Add another check.

Add another prompt.

Add another automation.

Add another layer.

Sometimes that helps.

Sometimes it just creates more work.

A sustainable workflow should be detailed enough to protect quality, but simple enough that I can actually use it consistently.

That balance matters.

If a system requires too much effort to maintain, eventually I will stop using it.

Then all of the sophistication becomes irrelevant.

Every new step should have to justify the time and complexity it adds.

The Workflow Has to Earn Its Place

Every stage should answer a question.

Research: What do I need to know?

Draft: What is the story?

Verification: What can I prove?

Red-team: What might be wrong?

Editing: Is this clear and still mine?

Quality control: Is the complete package ready?

Publish: Am I willing to stand behind it?

If a step does not improve the work, protect the work, or make the process easier to repeat, I need to ask why it exists.

That is one of the ways I keep the workflow from becoming bloated.

Not every possible check belongs in every story.

Not every tool belongs in every workflow.

Not every new AI feature needs to be added just because it exists.

Complexity has to earn its place.

Build Around the Work, Not the Tool

This has become one of my most important rules.

Do not build the workflow around the AI tool.

Build it around the work.

Tools change.

Features disappear.

Models improve.

Interfaces change.

Subscriptions change.

Companies change direction.

If my entire workflow depends on one exact feature working forever, then the system is fragile.

The real structure should be independent of the tool whenever possible.

I need research.

I need verification.

I need challenge.

I need editing.

I need quality control.

I need final human approval.

Which tool assists me at each stage can change.

The principles should survive the software.

But complete tool independence is not always realistic.

A workflow may become more efficient because of a particular platform or feature.

That is fine.

The important thing is knowing where that dependency exists.

If one tool disappears tomorrow, what breaks?

That is a question worth being able to answer.

Start With the Problem

Many of the workflows I now use did not begin because I sat down and decided to create a workflow.

They began because something kept happening.

I needed a better way to verify claims.

That became part of the truth process.

I needed something to challenge the story instead of agreeing with it.

That became red-teaming.

I needed a better way to check whether AI-generated writing still sounded like me.

That became part of the voice and editing process.

I needed to prevent graphics from slipping through with obvious errors.

That became visual quality control.

The workflow grew from problems.

That is important.

I was not collecting workflows.

I was solving recurring problems.

A Workflow Can Reduce What I Have to Hold in My Head

One of the biggest benefits of a repeatable system is that I do not have to remember everything every time.

Without a workflow, I have to keep asking myself:

Did I verify that?

Did I challenge the headline?

Did I check the graphic?

Did I review the source?

Did I make sure the final version is the one being published?

That creates mental load.

A good workflow can move some of that burden out of my head and into the process.

The process reminds me.

That does not remove judgment.

It protects my attention so I can use judgment where it matters.

Checklists Help, but They Are Not the Workflow

Checklists are useful.

I use them.

But a checklist is not the entire system.

Checking a box does not prove the work was done well.

“Verify” is meaningless if I do not actually reopen the source.

“Red-team” means nothing if I ignore every criticism.

“QC complete” means nothing if I rushed through it.

The checklist provides structure.

The quality still depends on the thinking inside each step.

That is why I keep returning to the same principle:

Process supports judgment.

It does not replace it.

Sustainability Requires Different Levels of Review

Not every piece of work needs the same amount of scrutiny.

A short social post does not need the same process as a deeply reported article.

A routine game recap may not require the same review as a story involving an accusation.

A graphic caption does not require the same level of verification as a major factual claim.

A sustainable workflow has to recognize that.

Otherwise, the process becomes too heavy.

I need a baseline standard.

Then I need the ability to increase scrutiny when the stakes increase.

That makes the system both responsible and usable.

Consistency should not become uniformity.

Some stories need a different process.

The Baseline Cannot Disappear

Risk-based review does not mean no review.

There still has to be a minimum.

Basic facts.

Names.

Dates.

Numbers.

Attribution.

Headline accuracy.

Graphic accuracy.

Disclosure where appropriate.

Final human review.

Those are not optional simply because the story seems easy.

Simple work can still contain mistakes.

A lightweight workflow is still a workflow.

Build Loops, Not Just Steps

Workflows are often drawn as straight lines.

Research → Draft → Verify → Edit → Publish.

Real work does not behave that neatly.

Verification can send me back to research.

Red-teaming can expose a missing source.

Editing can reveal a factual problem.

Quality control can uncover the wrong version.

A graphic can force me to reconsider the framing.

So I think less about a straight line now and more about loops.

The system moves forward.

But it also needs a way to move backward.

That flexibility is part of what makes it sustainable.

The Workflow Should Make Failure Visible

A bad workflow hides problems.

A good one exposes them.

I want to know when:

the source is weak,

the claim is unsupported,

the story is not ready,

the graphic is wrong,

the headline is overstated,

the AI recommendation is unreliable,

or the workflow itself is failing.

A workflow also has to assume that polished AI output can still be wrong.

That matters because a system that always reports success is not useful.

Sometimes the correct output is:

Not ready.

Needs more reporting.

Needs another source.

Needs another edit.

Needs to be stopped.

That is not the workflow failing.

That is the workflow doing its job.

Repeatability can create confidence.

Confidence should not become complacency.

The eleventh use still deserves judgment, even if the first ten went smoothly.

I Need to Know Where the Bottlenecks Are

A sustainable system also requires observation.

Where am I losing time?

Where do mistakes keep happening?

What stage causes the most rework?

What part feels unnecessarily complicated?

What keeps getting skipped?

Those are signals.

If graphics repeatedly require five revisions, I may need a better design specification.

If source verification keeps slowing the workflow, I may need better source organization.

If I keep rewriting AI-generated copy because it loses my voice, I may need a better editing instruction.

The workflow itself needs editing.

A Useful Workflow Is Rarely Permanently Finished

This is something I did not fully understand when I first started building systems.

I thought the goal was to create the workflow.

Now I think the goal is to maintain it.

The workflow changes as I learn.

It changes when the technology changes.

It changes when I make mistakes.

It changes when I find a faster approach that does not weaken the standard.

It changes when a new risk becomes obvious.

That does not mean the system has to be constantly rebuilt.

It means I should be willing to improve it when the evidence says I should.

Maintenance Is Part of the Cost

A workflow is not free once it exists.

Prompts need updating.

Automations can break.

Documentation gets old.

Tools change.

Processes that once saved time can become unnecessary.

That maintenance cost counts.

A system that saves ten minutes every day but requires hours of constant repair may not be as efficient as it appears.

A workflow should reduce friction overall.

It should not simply move the friction into maintenance.

Stable Principles, Flexible Methods

That is the balance I want.

Stable principles.

Flexible methods.

The principles stay:

Verify.

Challenge.

Maintain accountability.

Protect the meaning.

Keep human authority.

Correct mistakes.

Do not publish what I cannot defend.

The methods can change.

One AI tool may become another.

A manual step may become partially automated.

A checklist may become a dashboard.

A prompt may become an agent.

A graphic workflow may improve.

The methods evolve.

The standard remains.

Automation Should Remove Repetition, Not Oversight

Automation becomes useful when it removes repetitive work.

Monitoring.

Sorting.

Formatting.

Organizing.

Flagging.

Scheduling.

Preparing.

But I do not want automation to quietly remove the checkpoints that protect the journalism.

That distinction matters.

The goal is not:

Automate everything.

The goal is:

Automate what can be handled safely without removing the points where my judgment still matters.

Sometimes that will mean automation.

Sometimes the simplest reliable step will still be manual.

Not every repeatable task deserves an agent.

Automation Creates Dependencies

There is another side to automation.

The more connected the workflow becomes, the more a single failure can travel through the system.

A bad input can affect a summary.

The summary can affect the draft.

The draft can affect the headline.

The headline can affect the social copy.

One early problem can move downstream quickly.

That does not mean automation is a bad idea.

It means dependencies need to be visible.

I need to know which steps rely on which other steps.

And where I can interrupt the chain.

The More Automated the System, the More Visible the Controls Should Be

As workflows become more automated, it becomes more important that their control points remain visible and usable.

Where does verification happen?

Where can I reject the draft?

Where can I stop publication?

Where do I approve the graphic?

Where do I see which version is final?

Where do I know what AI contributed?

If I cannot answer those questions, the system may be efficient but difficult to control.

That is not sustainable.

A Sustainable Workflow Needs a Fallback

The happy path is easy.

Everything works.

The source is available.

The AI tool responds.

The automation runs.

The graphic generates correctly.

The publishing system behaves.

But systems fail.

What happens when the AI tool is unavailable?

What happens when an automation breaks?

What happens when the source disappears?

What happens when the wrong version is scheduled?

What happens when the graphic system fails on deadline?

A sustainable workflow needs a fallback.

Sometimes that fallback is another tool.

Sometimes it is a manual process.

Sometimes it means slowing down.

Sometimes it means not publishing yet.

The important thing is that one failure does not automatically destroy the entire process.

Documentation Matters

A workflow becomes easier to maintain when I can explain it.

What happens first?

What happens next?

What requires human approval?

What sources are acceptable?

What does “locked” mean?

What happens when something fails?

What gets recorded?

The process should not exist only in my memory.

Documentation creates consistency.

It also makes the system easier to teach.

But documentation only helps if it stays current.

An outdated instruction can create as much confusion as no instruction at all.

A Workflow Should Be Teachable

This is another test I have started to think about.

Could I explain the workflow to someone else?

Could another person understand:

what each stage does,

why it exists,

where AI helps,

where the human decides,

and what happens when something goes wrong?

If the answer is no, the workflow may be too dependent on intuition.

Intuition will always matter.

But a sustainable system needs enough structure that someone else can understand how it operates.

That matters even more if only one person currently knows how everything works.

If nobody else can understand or repair the system, that is a hidden weakness.

Sustainability Also Means Not Burning Myself Out

This part matters.

A workflow is not sustainable if it produces more work than I can responsibly carry.

AI can increase capacity.

That can create a new temptation:

Fill all the capacity.

More articles.

More projects.

More posts.

More experiments.

More workflows.

More everything.

That can defeat the purpose.

Efficiency should not automatically become workload expansion.

A sustainable system should make the work easier to manage.

Not simply make it possible to overload myself more efficiently.

Capacity Needs Limits

This is where measurement becomes important.

How much can I actually manage well?

Not how much can I technically produce.

How much can I verify?

How much can I review?

How much can I publish without lowering the standard?

How much can I keep track of without losing control?

Those questions define capacity better than raw output.

A sustainable workflow respects those limits.

The System Should Protect the Important Work

The purpose of efficiency is not to do everything.

It is to create room for what matters.

More reporting.

Better verification.

Stronger editing.

Deeper thinking.

Strategic planning.

Learning.

Rest.

Sometimes the best use of a faster workflow is not another article.

Sometimes it is having enough space to do the existing work better.

Build for the Bad Day

I think this may be one of the best tests of a workflow.

Does it still work on the bad day?

Not the perfect morning when I have energy, time and no interruptions.

The bad day.

The deadline day.

The tired day.

The day when three things happen at once.

The day when the first graphic is wrong.

The day when the source responds late.

The day when the AI gives me a bad answer.

That is when structure matters most.

But there is another test too.

What happens when the workflow itself is causing the bad day?

If maintaining the system becomes a constant burden, something needs to change.

The workflow should support the work.

The work should not exist to support the workflow.

Simplicity Is a Form of Strength

A sustainable workflow does not need to look impressive.

It needs to work.

The best systems are often boring.

Clear stages.

Clear standards.

Clear responsibility.

Clear stopping points.

Clear ways to recover when something goes wrong.

That is enough.

Complexity should earn its place.

If it does not make the work better, safer, clearer or easier to manage, I do not need it.

The Workflow Has to Belong to Me

There is no universal perfect AI workflow.

Different journalists.

Different newsrooms.

Different beats.

Different risks.

Different resources.

Different standards.

What works for me may not fit someone else exactly.

That is fine.

The point is not to copy my workflow line by line.

The point is to understand the principles well enough to build one that fits the work.

The workflow should reflect:

what I produce,

what risks I face,

what standards I follow,

what tools I use,

and where I need human judgment.

That is what makes it usable.

The Journalism Still Has to Be Good

This is the ultimate test.

A workflow can become faster.

Cleaner.

More automated.

More repeatable.

And still produce worse journalism.

That is possible.

A smooth system is not automatically a good system.

The workflow should be judged by whether it protects the quality of the journalism, not by how smoothly the machinery runs.

Does the reporting remain strong?

Does the evidence survive?

Does the writing remain specific?

Does the process catch mistakes?

Does the system leave room for stories that do not fit the template?

If not, the workflow needs to change.

Sustainable Means I Can Trust the Process

Not blindly.

Never blindly.

But I can trust that the process will ask the right questions.

Did I verify this?

Did I challenge it?

Did I preserve the meaning?

Did I inspect the final product?

Did I keep control of the decision?

Is this ready?

That is what I want from a workflow.

Not a machine that guarantees correctness.

A system that repeatedly puts me in the position to make better decisions.

Building a Workflow That Lasts

A sustainable AI-assisted workflow is not built by adding every new tool.

It is built by knowing what the work requires.

Protecting the important checkpoints.

Reducing unnecessary friction.

Learning from mistakes.

Maintaining the system.

Knowing when to automate.

Knowing when not to.

And remembering that efficiency is useful only if the work remains worth publishing.

The tools will change.

The workflow will evolve.

The methods will improve.

But the principles underneath the system need to survive all of that.

Because the best workflow is not the fastest.

Or the most automated.

Or the most complicated.

It is the one I can keep using, adapting and understanding without losing the standards that made the work worth doing in the first place.


Graphics created with AI-assisted design tools.
Visual concepts, editorial direction, composition, and final creative decisions were developed by Vallano Media.

The Human in the Workflow

By Cheval John | Vallano Media
This article was created with the assistance of artificial intelligence.
The final version was reviewed, edited and fact-checked by the author


Graphics created with AI-assisted design tools.
Visual concepts, editorial direction, composition, and final creative decisions were developed by Vallano Media.

AI can do a lot.

It can assist with research.

Summarize.

Organize.

Draft.

Edit.

Compare.

Challenge.

Generate ideas.

Suggest headlines.

Help build graphics.

Help prepare material for publication.

But there is still a question underneath all of that:

“Who is making the decisions?”

That question matters more to me than how advanced the technology becomes.

Because the more capable AI gets, the easier it becomes to confuse assistance with authority.

I do not want that confusion inside my workflow.

AI can help me move faster.

It can help me see more.

It can help me test the work.

But the decisions that determine what I publish still belong to me.

This Is Bigger Than Being “In the Loop”

Earlier in this book, I wrote about the importance of keeping a human in the loop.

That still matters.

But simply being present is not enough.

I can technically be part of the process while allowing AI to make most of the meaningful decisions.

I can click approve without really reviewing.

I can accept a rewrite because it sounds better.

I can use a suggested headline without checking whether it matches the evidence.

I can publish a graphic because it looks polished.

I can follow a recommendation because the AI sounds confident.

That is human presence.

It is not necessarily human control.

The question I care about now is:

“Where does the authority sit?”

And there is another question behind it:

“Is that authority real?”

Assistance and Authority Are Different

This distinction has become one of the most important parts of my workflow.

AI can assist with a decision.

That does not mean AI owns the decision.

For example, I can ask AI to compare three possible headlines.

It can tell me which one appears clearer.

It can identify which one may overstate the story.

It can suggest alternatives.

That is assistance.

But I still decide which headline gets published.

I can ask AI whether a paragraph seems weak.

It can point out missing context.

It can challenge the reasoning.

It can suggest a rewrite.

That is assistance.

But I still decide whether the paragraph survives.

The same principle applies throughout the workflow.

AI can contribute.

I retain authority.

But keeping authority human only matters if I am still exercising independent judgment.

If I approve whatever the system recommends, the authority exists mostly on paper.

I Have to Know What I Am Delegating

The more AI becomes integrated into my work, the more important it is to know what I am actually handing over.

Some delegation is obvious.

“Summarize these notes.”

“Create three headline options.”

“Find the weak point in this argument.”

“Clean up the grammar.”

Other delegation is less obvious.

“Which fact matters most?”

“What should lead the story?”

“Which source is more credible?”

“Is this ready to publish?”

Those questions carry more editorial judgment.

They may look like ordinary prompts.

But they can quietly transfer decision-making influence.

That does not mean I can never ask them.

It means I need to understand what I am asking AI to do.

I can use its response as input.

I should not confuse that input with the final decision.

A Tool Can Shape the Decision Before I Make It

This is one of the subtler problems.

AI does not need to make the final decision to influence it.

If a system chooses which sources I see first, that matters.

If it ranks one piece of evidence above another, that matters.

If it proposes the structure of the story, that matters.

If it gives me three headline choices, it has already influenced the range of options I am considering.

The final click may still belong to me.

But the tool may have shaped the decision long before I reach that point.

That means human control cannot exist only at the end.

I need to pay attention to how the workflow is shaping what I see, what I consider and what I ignore.

Not Every Decision Has the Same Weight

Some decisions are low risk.

Should this sentence be shorter?

Should this paragraph move higher?

Does this headline need fewer words?

Other decisions carry much more weight.

Is this source trustworthy?

Does the evidence support the accusation?

Does this wording imply causation?

Is this person being treated fairly?

Does this story need another source before publication?

Should this information be published at all?

The higher the consequence, the less comfortable I am with automatic delegation.

That is one reason I do not think a single rule works for every use of AI.

The level of human involvement should match the importance of the decision.

The stakes matter.

Human Judgment Does Not Mean Human Perfection

There is another point I have to acknowledge.

Humans make mistakes too.

I can misread a source.

I can become attached to an argument.

I can overlook context.

I can write a bad headline.

I can approve the wrong graphic.

I have biases.

I can be inconsistent.

Keeping the decision human does not magically make the decision correct.

Human authority does not make the judgment neutral.

It makes the judgment accountable.

Someone still has to be able to explain:

Why did you publish this?

Why did you trust this source?

Why did you use this wording?

Why did you leave this out?

Why did you believe the evidence was strong enough?

AI cannot carry that responsibility for me.

Authority May Be Shared

My workflow is built around the decisions I make.

But that does not mean editorial authority always belongs to one person.

In a larger newsroom, responsibility may be distributed.

A reporter may make one decision.

An editor another.

A producer may control part of the presentation.

A standards team may become involved.

A lawyer may review a particularly sensitive story.

The exact structure can change.

The principle does not.

Somebody needs meaningful authority over the consequential decisions.

And the people exercising that authority need enough understanding to evaluate what the system is producing.

The Most Important Decisions Often Happen Between the Steps

A workflow makes the process visible.

Research.

Draft.

Verify.

Red-team.

Edit.

Quality control.

Publish.

But some of the most important decisions happen between those labels.

Do I need more reporting?

Do I trust this explanation?

Is this technically true but misleading?

Am I forcing the story into a cleaner conclusion than the evidence supports?

Is the headline becoming stronger than the article?

Does this quote change the meaning?

Should I stop?

Those questions are harder to automate reliably because they depend heavily on context, judgment and consequences.

Sometimes they depend on experience.

Sometimes ethics.

Sometimes simply noticing that something does not feel settled yet.

That is where the human in the workflow matters most.

AI Can Recommend. I Can Reject.

One of the most important things I have learned is that I do not need to accept a useful-looking AI recommendation.

That sounds obvious.

In practice, it matters.

AI can give me a polished answer quickly.

That creates momentum.

The easiest thing is to keep moving.

Accept.

Insert.

Publish.

But I have found that some of the best decisions happen when I say:

No.

That headline is too strong.

That sentence does not sound like me.

That criticism is irrelevant.

That graphic is wrong.

That source is not good enough.

That conclusion goes beyond the evidence.

That draft needs more work.

The ability to reject AI is just as important as the ability to use it.

Familiarity Can Reduce Skepticism

There is another risk.

The more often a system gives me useful answers, the easier it becomes to trust the next answer.

That is understandable.

It is also dangerous.

Reliability can create habit.

Habit can reduce scrutiny.

The more often a system is right, the easier it becomes to stop questioning it.

That is exactly when skepticism still matters.

I do not need to assume every answer is wrong.

But I should not allow a good track record to turn into automatic approval.

Sometimes the Best Decision Is to Stop the Workflow

Efficiency can create pressure to keep going.

Once research becomes a draft, the draft becomes an edit, the edit becomes a graphic and the graphic becomes a scheduled post, the process starts building momentum.

That momentum can be useful.

It can also be dangerous.

There needs to be a point where I can stop everything.

Not delay publication by five minutes.

Stop it.

Go back.

Report more.

Change the premise.

Abandon the story if necessary.

A workflow is only genuinely under human control if the human can interrupt it.

Otherwise, the workflow is just moving toward publication.

But even that is not enough.

Authority is meaningful only when the surrounding system allows the human to use it.

If deadlines, incentives, workplace pressure or culture make “stop” impossible in practice, then formal authority may not mean much.

AI Can Make the Wrong Decision Look Reasonable

This is one of the reasons I remain careful.

A weak human idea often looks weak.

A half-developed thought may be obviously unfinished.

AI can make that same idea look polished.

It can organize it.

Strengthen the wording.

Add transitions.

Make the logic feel smooth.

That creates a dangerous possibility:

The presentation improves while the underlying decision remains bad.

A polished recommendation is still a recommendation.

A confident explanation is still something that needs evaluation.

Good language does not turn weak judgment into strong judgment.

The Human Decides What Matters

AI can find patterns.

It can rank information.

It can summarize what appears frequently.

It can identify what looks unusual.

But journalism still requires someone to decide what matters.

That is not always the same as what is most dramatic.

Or most popular.

Or most repeated.

Or easiest to summarize.

Something can be statistically unusual and editorially unimportant.

Something can receive little attention and still matter enormously.

Something can be interesting without being relevant.

Something can be true without belonging in the story.

Those are editorial decisions.

The technology can inform them.

It should not silently make them.

And human authority does not guarantee that I will make the right choice.

It means I am responsible for making and defending that choice.

The Human Decides What the Evidence Means

Evidence does not always explain itself.

Two sources may disagree.

A statistic may look important until the denominator changes the picture.

A quote may sound powerful until the surrounding context changes its meaning.

A trend may exist without showing causation.

AI can help surface those problems.

But the final interpretation still requires judgment.

Human judgment does not mean inventing meaning.

It means deciding what conclusions the evidence can actually support.

What does the evidence show?

What remains uncertain?

What can I say confidently?

What needs qualification?

Those decisions determine whether the journalism is careful or careless.

The Human Decides What Does Not Get Published

This may be even more important than deciding what gets included.

Every story leaves something out.

Some information is irrelevant.

Some is weakly supported.

Some is repetitive.

Some is potentially harmful.

Some cannot be verified.

Some may be interesting but unfair to include.

AI can suggest what belongs in the story.

But restraint is part of journalism too.

The ability to leave something out can be as important as the ability to find it.

There is a balance here.

Caution should not become an excuse to avoid publishing well-supported reporting.

Restraint matters.

So does having enough confidence to publish when the evidence is strong.

The Human Owns the Exceptions

Workflows are built around patterns.

Most articles move through familiar stages.

Most fact checks follow familiar logic.

Most edits involve familiar questions.

But unusual situations happen.

A source contradicts the official record.

A new document arrives after the draft is finished.

A quote changes the central premise.

A graphic creates a meaning I did not intend.

A legal or ethical question appears late.

The workflow cannot predict every exception.

That is where judgment becomes more important, not less.

A good system helps me handle routine work.

It should also leave enough flexibility for me to recognize when the routine no longer applies.

This Changes How I Think About Automation

I am interested in automation.

I am interested in AI agents.

I am interested in systems that can monitor information, organize work and reduce repetitive tasks.

But the more I think about automation, the more I separate two things:

“execution” and “authority.”

I am comfortable automating parts of execution.

Monitoring.

Sorting.

Formatting.

Collecting.

Comparing.

Preparing.

Flagging.

I am much more careful about automating authority.

What is true?

What is fair?

What is strong enough to publish?

What deserves emphasis?

What should be excluded?

When should the process stop?

Those are different kinds of questions.

The Better the AI Gets, the More Important This Becomes

It might seem that better AI would reduce the need for human involvement.

In some tasks, it probably will.

But greater capability also means greater temptation to delegate.

If AI becomes better at research, drafting, editing, graphics and production, the question will not simply be:

Can it do this?

The question will become:

Should I let it decide this?

Those are not the same question.

Capability does not automatically create authority.

A system can be very good at producing an answer while still being the wrong place to locate responsibility.

My Job Is Becoming More About Decisions

This has been one of the biggest changes in my own work.

The more AI assists with execution, the more valuable my judgment becomes.

I am spending less of my attention on:

How do I physically produce every part of this?

And more on:

Is this right?

Is this useful?

Is this supported?

Is this fair?

Does this belong?

What needs another look?

What should happen next?

That is not less work.

It is different work.

And in many ways, it is closer to what I should have been concentrating on all along.

I Still Have to Know the Craft

There is a catch.

I cannot reliably direct work I do not understand.

If I do not understand reporting, I cannot reliably judge the research.

If I do not understand sourcing, I cannot evaluate the evidence.

If I do not understand writing, I cannot recognize a weak draft.

If I do not understand editing, I cannot know when meaning has changed.

If I do not understand visual communication, I may approve a misleading graphic.

AI can help people learn the craft.

But the more consequential the decision, the more dangerous it becomes to rely on a tool I cannot evaluate.

AI does not eliminate the need to know the craft.

It may make that knowledge even more important.

Because now I am not only doing the work.

I am evaluating work produced with assistance.

Direction Requires Standards

To direct AI well, I need standards.

What counts as a strong source?

What level of verification is enough?

What tone fits the publication?

What kinds of claims require more scrutiny?

What makes a graphic acceptable?

What does publication-ready mean?

Without standards, AI can produce endless options.

But options are not direction.

The human brings the standard.

The standard makes the decisions possible.

But standards need review too.

A consistent process can still produce consistently bad decisions if the standard itself is weak.

The workflow should not only test the work.

Over time, it should also force me to test the standards behind the work.

The Human in the Workflow Is Not a Symbol

I do not want the human role to become ceremonial.

Not a name at the end of the process.

Not a final click.

Not someone who approves whatever the system produces.

The human role has to exist throughout the workflow.

Setting the goal.

Testing the claims.

Rejecting bad recommendations.

Recognizing uncertainty.

Stopping the process when necessary.

Owning what gets published.

That is meaningful control.

Real authority means having decision space throughout the process.

Not just possession of the publish button.

AI Can Do a Lot

And I expect it will be able to do even more.

That does not bother me.

The goal is not to protect tasks simply because humans used to perform them.

If AI can handle something safely, accurately and efficiently, I want to understand how to use that capability.

But I also want to know where I draw the line.

Not because humans must perform every task.

Because somebody still has to make the decisions that carry responsibility.

For my work, that somebody is me.

AI can assist with research.

Draft.

Assist with verification.

Challenge.

Edit.

Organize.

Prepare.

Recommend.

But contribution is not authority.

And formal authority is not enough if I cannot exercise it independently.

The most important decisions still belong to me:

evidence,

fairness,

meaning,

uncertainty,

restraint,

publication,

responsibility.

And the more capable the technology becomes, the more deliberate I have to be about keeping those decisions genuinely human.


Graphics created with AI-assisted design tools.
Visual concepts, editorial direction, composition, and final creative decisions were developed by Vallano Media.

Measuring the Impact

By Cheval John | Vallano Media
This article was created with the assistance of artificial intelligence.
The final version was reviewed, edited and fact-checked by the author


Graphics created with AI-assisted design tools.
Visual concepts, editorial direction, composition, and final creative decisions were developed by Vallano Media.

One of the easiest claims to make about artificial intelligence is that it saves time.

That may be true.

But I wanted something better than a feeling.

I wanted evidence.

Not perfect evidence.

Not a controlled experiment.

Not a scientific study.

Just enough documentation to answer a practical question:

“What actually changed after I integrated AI into my work?”

That meant looking at time.

Output.

Workflow.

And the way I was spending my attention.

The answer turned out to be more complicated than simply saying:

“AI made me faster.”

It did make some parts faster.

But the bigger change was that I started working differently.

Before AI, Time Was Concentrated in Production

Before I built an AI-assisted workflow, producing an article meant I handled nearly every stage manually.

Research.

Organization.

Drafting.

Editing.

Rewriting.

Headline development.

Formatting.

Publishing.

For an article of more than 800 words, I could easily spend three to four hours moving from the idea to publication.

That number is not a controlled benchmark.

I was not timing every stage in those years.

Different articles required different amounts of research.

Some were easier than others.

Some involved interviews.

Some involved more complicated subjects.

So I cannot take a four-hour article from the past, compare it to a forty-minute article today and declare that AI made me six times more productive.

That would be cleaner than the evidence allows.

But the comparison is still useful.

My process used to require significantly more manual effort across nearly every stage.

Today, much of that effort has been redistributed.

I Started Measuring the Workflow

As I became more serious about AI integration, I started paying attention to production time.

Not because speed was the goal.

Because I wanted to know what was changing.

One of the earliest useful benchmarks came from this book.

The first day’s chapter took about “43 minutes” from the first draft through the publication-ready stage, excluding research that had already been done.

Later in the challenge, another chapter moved through the editorial process, locked manuscript, main graphic, tags and teaser in roughly “10 minutes.”

Those numbers are interesting.

They are also easy to misuse.

The chapters were different.

The research requirements were different.

I was becoming more familiar with the workflow.

The AI systems were operating within a structure I had already built.

And some of the improvement may have come from repetition and experience, not AI alone.

That matters.

I was not running a controlled experiment where every variable remained the same.

So I do not look at those numbers as proof of a specific productivity percentage.

I look at them as evidence of something more practical:

“The workflow was becoming easier to execute.”

The Indiana Article Gave Me Another Benchmark

A separate article gave me another useful data point.

The drafting-to-final stage took about “36 minutes and 54 seconds.”

Scheduling and production added roughly another “31 minutes.”

That matters because it shows why “writing time” alone can be misleading.

A finished article is not just a draft.

There is still production.

Formatting.

Graphics.

Scheduling.

Final checks.

Publishing infrastructure.

If I only measured the drafting stage, I could tell myself a much more impressive story.

But it would not represent the full work.

That became one of the lessons of tracking time:

“Measure the process you actually perform, not the piece that makes the best statistic.”

Faster Drafting Was Only Part of the Change

AI definitely reduced friction in drafting.

I no longer had to stare at a blank screen for as long.

I could start with research notes, identify the structure, build an outline and move into the draft faster.

But that was not the most important change.

The most important change was what I could do with the time that opened up.

I could challenge the argument.

I could check the claims.

I could go back to the sources.

I could run a truth pass.

I could red-team the story.

I could inspect the graphic.

I could compare versions.

I could spend more time deciding whether something should survive rather than spending all my time trying to produce the first version.

That is a different kind of productivity.

Output Increased Because Friction Decreased

Another change was output.

Once the workflow became repeatable, I could produce more without feeling as if every article had to begin from zero.

Research had a process.

Drafting had a process.

Verification had a process.

Editing had a process.

Graphics had a process.

Publishing had a process.

The result was not simply that AI could generate text faster.

The result was that I was building a publishing system.

That distinction matters.

A faster writer can produce more words.

A better system can produce more finished work.

Those are not the same thing.

More Output Does Not Automatically Mean Better Work

There is a danger here.

Once production becomes easier, it is tempting to produce simply because you can.

Another article.

Another post.

Another graphic.

Another idea.

AI can make quantity feel almost effortless.

But volume is not the same as value.

If the additional output is poorly researched, repetitive or unnecessary, then more productivity may just mean more noise.

And more capacity is only useful if I choose what deserves that capacity.

That is why I do not want to measure success by article count alone.

The better question is whether the additional capacity allows me to produce work that is:

more consistent,

better verified,

more useful,

more complete,

or more sustainable.

If the answer is no, then the increased output is not much of an achievement.

AI Changed My Role

There was another change that was harder to measure with a stopwatch.

My role changed.

Earlier, I was doing nearly every task myself.

Now I increasingly direct the tasks.

I decide what needs research.

I decide what needs verification.

I decide when AI should challenge the work.

I decide what version survives.

I decide when the manuscript is locked.

I decide when the graphic is ready.

I decide what gets published.

That does not mean I stopped working.

It means the work moved upward.

More of my effort went into coordination, judgment and approval.

I became less focused on executing every individual task manually and more focused on managing the system that produced the work.

That is one of the biggest changes AI created for me.

And it is difficult to express with a simple productivity number.

The Work Became More Repeatable

Before the workflow existed, every article depended more heavily on memory.

Did I check that number?

Did I verify that quote?

Did I review the headline?

Did I challenge the argument?

The more tasks I handled manually, the easier it was for the process to vary from article to article.

Now the workflow gives me checkpoints.

That does not guarantee quality.

But it creates consistency.

I know what happens after the draft.

I know what happens after verification.

I know when the red-team pass begins.

I know when the manuscript becomes locked.

That repeatability matters because publishing regularly requires more than creativity.

It requires a system that can work again tomorrow.

Some Work Actually Increased

AI did not make every part of journalism faster.

In some areas, it created additional work.

AI output has to be checked.

Sources have to be verified.

Rewritten sentences have to be compared against the evidence.

Graphics have to be inspected.

Generated suggestions have to be evaluated.

A polished AI answer can deserve extra scrutiny because fluency can make a weak or false claim look settled.

So there are moments when AI saves time in one part of the process and adds work somewhere else.

Sometimes AI does not remove time from the process.

It moves the time somewhere else.

That is not necessarily a problem.

The important question is whether those additional minutes improve the final work.

If they do, then the effort may be valuable.

The Fast Workflow Was Not Built Overnight

There is another part of the productivity story that is easy to miss.

The workflow I use today did not appear instantly.

It sits on top of months of learning.

Experimentation.

Failed prompts.

Revisions.

New workflows.

Abandoned ideas.

Understanding what AI does well.

And understanding where I do not trust it.

The current efficiency exists partly because of that earlier investment.

Someone starting from zero should not expect to reproduce my current workflow speed immediately.

The faster process came after building the system.

That setup time counts too.

Failed Attempts Count

There is another temptation when measuring productivity:

Remember the successes.

Forget the failures.

A graphic that works on the first generation is easy to count.

The five bad versions before the final one are easy to forget.

A clean draft looks efficient.

The abandoned structure that came before it can disappear from the measurement.

If I want an honest record, I have to include rework.

Failed experiments.

Bad outputs.

Wrong directions.

The things that did not survive.

A fair measurement system cannot count only the examples that went smoothly.

The failures are part of the cost of the workflow too.

Efficiency Is Not the Same as Cutting Corners

This distinction has become increasingly important to me.

Efficiency means removing unnecessary friction.

Cutting corners means removing necessary safeguards.

Those two things can look similar from the outside.

Both may reduce time.

But they produce very different results.

If AI helps me organize notes faster, that is efficiency.

If I stop reading the source because AI summarized it, that is cutting a corner.

If AI helps identify weak claims before publication, that is efficiency.

If I accept its factual corrections without checking them, that is cutting a corner.

The goal is not to make every stage shorter.

The goal is to make the whole process better.

Time Saved Needs Somewhere to Go

This may be the most important measurement question of all.

What happens to the time AI saves?

If I save thirty minutes and immediately fill those thirty minutes with more low-value work, I have increased activity.

Not necessarily productivity.

If I use the time to verify a source, improve the argument, learn a new skill, work on another important project or simply reduce unnecessary workload, then the savings become meaningful.

AI does not decide where the time goes.

I do.

That makes time savings a management question as much as a technology question.

The Numbers Need Context

I have started collecting timing benchmarks because I want a record.

But the numbers need context.

A ten-minute workflow does not mean every chapter should take ten minutes.

A forty-minute article is not automatically better than a three-hour article.

A fast result may reflect:

better preparation,

a familiar subject,

stronger systems,

less research,

greater experience,

or a simpler assignment.

A slower result may reflect deeper reporting or more complicated evidence.

And the examples I track are not automatically representative of every kind of journalism I produce.

An interview-heavy story may require a completely different amount of time.

So might an unfamiliar subject.

A breaking story.

Or a piece that requires deeper original reporting.

That is why I do not want to turn speed into the standard.

The numbers are signals.

They help me see how the system is changing.

They do not tell the whole story.

Measuring Speed Can Change the Work

There is another complication.

The moment I start measuring speed, I also have to watch whether the measurement itself is changing how I work.

If I know the clock is running, do I rush?

Do I become reluctant to go back to research because it will hurt the number?

Do I treat a ten-minute workflow as something I need to beat?

That would defeat the purpose.

The measurement is supposed to help me understand the workflow.

The workflow should not start serving the measurement.

If the story needs more time, it gets more time.

Quality Is Harder to Measure

Time is easy to measure.

Quality is not.

I can tell you that a process took forty minutes.

It is much harder to reduce the quality of the finished journalism to a single number.

Did I use stronger sources?

Did I catch an error before publication?

Did important context survive the editing process?

Did the headline accurately represent the reporting?

Did the workflow make my work more consistent?

Did I reduce avoidable corrections?

Those questions may ultimately matter more than the stopwatch.

That means any serious measurement of AI-assisted journalism needs more than productivity statistics.

It needs evidence about the quality of the work too.

The Better Measurement May Be Capacity

The more I have worked this way, the more I think the real benefit may be capacity.

How much high-quality work can I responsibly manage?

How many different projects can I keep moving?

How quickly can I move from research to publication without losing oversight?

How much attention remains available for the decisions that actually require me?

That is different from measuring words per hour.

Capacity includes time.

But it also includes cognitive load.

Consistency.

Repeatability.

Quality control.

And the ability to keep operating without every project becoming exhausting.

Capacity may be the most important benefit I see right now.

Future measurements may show that consistency, quality control or reduced cognitive load matter even more.

That is why I want to keep measuring.

Measuring the Wrong Thing Can Create the Wrong Incentive

Metrics influence behavior.

If I only measure production time, I will naturally try to make everything faster.

If I only measure output, I will try to publish more.

If I only measure traffic, I may start choosing stories based on clicks.

That is why measurement needs to reflect the kind of journalism I actually want to produce.

Some useful questions might be:

Did I verify the central claims?

Did the story survive the red-team pass?

Did I catch problems before publication?

Did the workflow reduce unnecessary repetition?

Did the finished work meet the standard?

Did the process become easier to repeat?

Those questions are less exciting than saying AI made me five times faster.

They are also more useful.

Efficiency Has a Cost Side

AI-assisted work is not free simply because it saves time.

There are tools.

Subscriptions.

Training.

Experimentation.

Failed generations.

And the time required to build and manage the system.

Those costs matter too.

A productivity measurement that counts only minutes saved and ignores what it took to create those savings is incomplete.

The goal is not to prove that the technology has no cost.

The goal is to understand whether the total system gives me enough value to justify it.

I Am Still Building the Dataset

This is not the final measurement.

I am still early in the process.

The timing records from this writing challenge are part of a larger dataset I am building around my own work.

More articles will give me better comparisons.

More mistakes will show which safeguards need improvement.

More repetition will show whether the gains are durable or just the result of novelty.

And a better record will help me distinguish what came from AI from what came from experience, repetition and improved systems.

That is how I want to approach AI productivity.

Not with one impressive example.

With a record.

What Changed?

So what changed after integrating AI?

I became faster in parts of the workflow.

I increased the amount of work I could manage.

The process became more structured.

The work became more repeatable.

More of my attention shifted from producing the first version to reviewing, challenging and directing the work.

Some tasks became easier.

Some verification work increased.

Some time disappeared.

Some time simply moved.

And my role changed from doing every task manually to managing a system of tasks.

That is more difficult to fit into a marketing slogan.

But it is closer to what actually happened.

AI did not simply make me faster.

It changed how I work.

And I am still measuring exactly what that change means.

Because the real question is not:

“How much faster can I publish?”

It is:

“What can I do better with the time, attention and capacity AI gives back?”


Graphics created with AI-assisted design tools.
Visual concepts, editorial direction, composition, and final creative decisions were developed by Vallano Media.

From Draft to Publish — The Full Workflow

By Cheval John | Vallano Media
This article was created with the assistance of artificial intelligence.
The final version was reviewed, edited and fact-checked by the author


Graphics created with AI-assisted design tools.
Visual concepts, editorial direction, composition, and final creative decisions were developed by Vallano Media.

A draft can look finished long before it is ready to publish.

That is one of the biggest lessons I have learned from working with AI.

The writing may be clean.

The structure may work.

The grammar may be correct.

The headline may sound strong.

The article may even feel complete.

But publication requires more than a good draft.

It requires a process.

A draft has to survive verification.

Challenge.

Editing.

Quality control.

And one final decision:

Is this actually ready to go out into the world?

That is what the full workflow is designed to help me answer.

But there is an important distinction.

“Completing the workflow is not the same as completing it well.”

A checklist cannot guarantee good journalism.

The value comes from the judgment exercised inside every stage.

The Draft Is a Starting Point

By the time I reach the first complete draft, a lot of work has already happened.

The research has been gathered.

The evidence has been organized.

The story has taken shape.

The first version exists.

That feels like progress.

It is.

But the draft is not the finish line.

It is the point where the next stage begins.

Now the question changes from:

“Can I write this?”

to:

“Can I defend this?”

That is where the editorial workflow becomes essential.

Step One: Check the Claims

Before I think about style, I look at the substance.

What factual claims am I making?

Which ones are central?

Which ones are current?

Which ones are sensitive?

Which ones depend on a source I need to revisit?

Which ones sound more certain in the draft than they did in the research?

The goal is not to verify only the obvious numbers and dates.

It is to inspect the statements the reader is being asked to believe.

That includes interpretation.

Causation.

Context.

Attribution.

Comparison.

The stronger the claim, the stronger the evidence needs to be.

Step Two: Go Back to the Sources

The next step is source verification.

I open the documents again.

I check the original quote.

I confirm the number.

I reread the section of the report.

I look at the official record.

I make sure the source actually supports the sentence I wrote.

This can feel repetitive.

That is part of the point.

The draft is a transformation of the research.

Something can change during that transformation.

A qualification can disappear.

A number can lose context.

A sentence can become stronger.

A quote can be shortened in a way that changes its meaning.

Going back to the source reconnects the prose to the evidence.

Step Three: Run the Truth Pass

The truth pass asks a broader question:

What in this draft might be wrong, misleading, incomplete or overstated?

I look at the article differently during this stage.

Not as the writer.

As the skeptic.

What would I challenge if someone else wrote this?

Which sentence makes me hesitate?

Which claim depends on an assumption?

Which section needs another source?

Which statement might be technically accurate but missing important context?

Sometimes the answer is:

Nothing significant.

Sometimes the truth pass sends me back into research.

That is not a failure.

That is the workflow working.

Step Four: Red-Team the Story

Once the factual foundation is solid, I try to break the article.

This is where the red-team stage comes in.

I ask:

What is the strongest challenge to the central argument?

What evidence could weaken the conclusion?

What might a knowledgeable critic question?

What context is missing?

What is the weakest part of the story?

Does the headline promise more than the article proves?

AI can help generate those challenges.

But I do not accept them automatically.

Some critiques are useful.

Some are irrelevant.

Some are simply wrong.

The goal is not to obey the red team.

The goal is to expose the story to resistance.

Step Five: Decide What Survives

The red-team stage creates another decision point.

What actually needs to change?

A weak claim may need another source.

A sentence may need softer wording.

A section may need more context.

A dramatic headline may need to come down.

A paragraph may need to disappear completely.

And sometimes the story survives the challenge exactly as written.

That matters too.

The goal is not endless revision.

It is identifying material weaknesses.

Then fixing the ones that matter.

Step Six: Edit for Voice

Once the reporting survives the challenge, I focus on the writing.

Does it sound like me?

Does the rhythm work?

Is the article direct?

Is it repetitive?

Did AI make the language too formal?

Too generic?

Too cautious?

Too dramatic?

Are the strongest lines still there?

Has anything become polished at the expense of specificity?

This is where I try to make the article sound finished without making it sound manufactured.

The writing should become clearer.

Sharper.

More readable.

But still mine.

Step Seven: Read for Meaning, Not Just Grammar

Grammar matters.

So does punctuation.

But one of the most dangerous editing mistakes is concentrating so much on the sentence that I stop checking what the sentence actually means.

A small wording change can alter:

causation,

certainty,

timing,

attribution,

or scale.

“Led” is not the same as “won.”

“Reported” is not the same as “confirmed.”

“After” is not the same as “because of.”

A cleaner sentence is not automatically a more accurate sentence.

So after the voice pass, I read again for meaning.

Step Eight: Editorial Quality Control

Then comes the final QC pass.

This is where I stop developing the story and inspect the complete product.

Names.

Dates.

Numbers.

Quotes.

Attribution.

Headline.

Graphics.

Disclosure language.

Formatting.

Anything that might create a problem after publication.

This is also where production begins to overlap with editorial work.

The article itself can be correct while something around it is wrong.

A graphic may contain an error.

A caption may overstate the story.

A headline may be stronger than the evidence.

A link may go to the wrong source.

The entire publication package can affect how the journalism is understood.

Step Nine: Inspect the Graphic

AI-assisted graphics deserve their own check.

The visual may look excellent.

That does not mean it is finished.

I look for:

incorrect text,

strange anatomy,

misleading symbols,

factual mistakes,

unlicensed or inappropriate third-party visual elements,

and anything that changes the meaning of the story.

But visual quality control is bigger than catching an extra hand or a misspelled word.

A visual can be technically clean and still imply something the reporting does not support.

The same principle applies to the graphic as the article:

Do not approve it just because it looks polished.

Look at what it actually communicates.

Step Ten: Lock the Manuscript

Eventually, the editing has to end.

That is when I lock the manuscript.

Locked means the article has completed the required editorial stages.

The reporting has been checked.

The claims have been tested.

The writing has been edited.

The QC pass is complete.

It is ready for production.

Locked does not mean perfect.

It does not mean the article can never change.

New evidence may emerge.

A late source response may change the story.

An error may become visible only after publication.

If that happens, I should correct or update the work.

A strong workflow reduces risk.

It does not eliminate the possibility of correction.

Locked simply means I have reached the standard required to publish responsibly with the information available to me.

Step Eleven: Build the Publication Package

The article is only one part of what the audience sees.

The full publication package may include:

the headline,

main graphic,

tags,

disclosure,

social copy,

preview text,

links,

and sometimes a teaser for what comes next.

Each element needs review.

If the article is careful but the social post exaggerates it, the publication can still mislead.

If the headline is accurate but the graphic creates a false impression, there is still a problem.

And every derivative asset should be checked against the verified story itself.

Not against another derivative asset.

A caption based on a social post based on a summary of the article creates opportunities for meaning to drift farther from the original reporting.

The evidence remains the reference point.

Version Control Is Editorial Control

AI-assisted workflows can produce a lot of versions.

Several drafts.

Several headlines.

Several graphics.

Several social posts.

Several rewritten paragraphs.

That creates another risk.

Publishing the wrong one.

The strongest version is useless if an outdated draft reaches the website.

The corrected graphic does not help if the earlier version is uploaded.

The accurate headline does not matter if an abandoned option gets scheduled.

“Version control is part of editorial control.”

I need to know which manuscript is locked.

Which graphic is final.

Which headline is approved.

Which copy is actually being published.

That may sound administrative.

It is also editorial.

Step Twelve: Final Human Approval

At the end of the workflow, one decision still belongs to me.

Publish?

That sounds simple.

But it carries everything that came before it.

Do I trust the sourcing?

Do I understand the evidence?

Did I challenge the story?

Did I review the AI assistance?

Does the writing sound right?

Does the headline fit the reporting?

Does the graphic work?

Is there anything I would hesitate to defend after publication?

Final approval matters only if I am genuinely willing to stop the process.

If “Publish” is inevitable no matter what I find, then the final review is ceremonial.

Meaningful human oversight requires the authority to say:

Not yet.

If something is wrong, I go back.

If the work is ready, I publish.

That final approval matters because of the decisions that led to it.

The Workflow Can Move Backward

The full process is not perfectly linear.

A red-team pass can send me back to research.

Editing can reveal a factual question.

QC can uncover a problem in the headline.

A graphic can expose that the concept itself is unclear.

New evidence or a late source response can change the story near the end of the process.

The arrows sometimes move backward.

That is normal.

A workflow should not trap the work in a rigid sequence.

It should give the work checkpoints.

Places where I know what question I am supposed to ask next.

Deadlines Change the Pace, Not the Responsibility

Real journalism does not always happen under ideal conditions.

Breaking news moves quickly.

Live sports move quickly.

Deadlines compress the available time.

Staffing may be thin.

The workflow has to function in that environment too.

That may mean some stages happen quickly.

Several stages may overlap.

The level of review may depend on the risk of the story.

But deadlines should not erase the steps that protect accuracy.

Speed can change how I execute the workflow.

It should not change what I am responsible for.

The Workflow Is Not the Journalism

This is important.

The workflow is not the thing that makes the journalism good.

The journalism still depends on:

good reporting,

strong evidence,

sound judgment,

fairness,

context,

clarity,

and accountability.

A checklist cannot replace those things.

A workflow simply creates a structure around them.

It makes it harder to accidentally skip an important step.

That is useful.

But process is not proof.

A completed checklist does not guarantee a correct article.

Completing the workflow is not the same as completing it well.

The thinking still matters.

The Workflow Becomes Faster With Practice

There is another benefit.

The more often I use the workflow, the faster it becomes.

I know what to look for during the truth pass.

I know what kinds of questions work during red-teaming.

I know where graphics often fail.

I know which parts deserve more scrutiny.

That familiarity reduces friction.

But it should not reduce skepticism.

The process can become faster.

The judgment should not become automatic.

The More Automation, the More Important the Checkpoints

As AI-assisted workflows become more capable, some steps may become more automated.

That can be useful.

But automation creates another question:

Where does human judgment still have real authority?

If the system automatically summarizes the research, drafts the article, proposes the headline, generates the graphic and prepares the social copy, the workflow may look efficient.

It also creates more places where an early error can travel downstream.

The more automated the workflow becomes, the more important it is to know where a human can inspect, challenge and stop it.

Automation should not make responsibility harder to locate.

Speed Is Not the Victory

One of the easiest things to measure is time.

How fast did the draft happen?

How quickly did the article get published?

How many minutes did the workflow take?

Those numbers can be useful.

But speed is not the victory.

The real question is:

Did the workflow help produce something I can stand behind?

Sometimes that takes ten minutes.

Sometimes forty minutes.

Sometimes several hours.

Sometimes the correct answer is:

Not ready yet.

That is still a successful outcome.

AI Changes Where the Work Can Happen

AI does not remove the work.

It changes where some of the work can happen.

Less time may go into producing the first draft.

AI can assist with the more repeatable and support-oriented parts of the process.

That creates the possibility of reallocating more attention toward:

verification,

challenge,

comparison,

editing,

and judgment.

But that redistribution is not automatic.

Saving twenty minutes does not guarantee I will spend those twenty minutes verifying a source.

I still have to make that choice.

AI creates the opportunity.

The journalist decides how to use it.

The Full Workflow Is Really a Series of Decisions

Research.

Draft.

Truth pass.

Verification.

Red-team.

Voice edit.

Quality control.

Production.

Publish.

Those stages look like tasks.

They are really decisions.

What is true?

What matters?

What needs evidence?

What deserves doubt?

What should stay?

What should go?

What sounds right?

What is ready?

AI can help at almost every stage.

But it cannot take ownership of those decisions.

That remains human.

Before I Hit Publish

The final button is the easiest part.

Everything before it is the journalism.

The research.

The verification.

The challenge.

The editing.

The judgment.

The refusal to publish something just because it looks complete.

That is what the workflow protects.

Not perfection.

Responsibility.

So my final question is not simply:

“Does this look finished?”

It is:

“Have I done enough to stand behind it?”

And one more question follows:

“If this is challenged after publication, can I explain how I know what I published?”

When the answer to both is yes, then I hit publish.


Graphics created with AI-assisted design tools.
Visual concepts, editorial direction, composition, and final creative decisions were developed by Vallano Media.

Keep the Writing Yours

By Cheval John | Vallano Media
This article was created with the assistance of artificial intelligence.
The final version was reviewed, edited and fact-checked by the author


Graphics created with AI-assisted design tools.
Visual concepts, editorial direction, composition, and final creative decisions were developed by Vallano Media.

AI can make writing cleaner.

That does not always make it better.

One of the easiest things to lose in AI-assisted writing is also one of the hardest things to rebuild.

Voice.

The phrasing that sounds like you.

The rhythm that feels natural.

The choices you make because of experience.

The way you emphasize one detail and move quickly past another.

The language you would actually use.

AI can help improve a draft.

It can tighten sentences.

Remove repetition.

Fix awkward phrasing.

Clarify confusing sections.

Suggest transitions.

Improve structure.

But if I am not careful, it can also sand down everything distinctive about the writing until the article sounds polished, correct and completely interchangeable with thousands of other articles.

That is not what I want.

I want the writing to become clearer.

Not anonymous.

Editing Is Not Rewriting the Writer

There is a difference between editing a piece of writing and replacing the person who wrote it.

Good editing should help the work communicate better.

It should strengthen what is already there.

Clarify what is confusing.

Remove what is unnecessary.

Fix what is wrong.

But the finished article should still sound like it came from the same person who began it.

That is one of the standards I use when I edit with AI.

If the sentence becomes grammatically cleaner but no longer sounds natural to me, that is not automatically an improvement.

If the paragraph becomes more formal but loses its energy, that may not be an improvement.

If the article becomes smoother but less specific, that is definitely not an improvement.

Editing should refine the voice.

Not erase it.

Preserving Voice Does Not Mean Preserving Every Habit

There is an important distinction here.

Not everything I naturally write deserves to survive editing.

I can repeat myself.

I can overwrite.

I can use an awkward phrase.

I can fall into habits that make the article harder to read.

Those things are not automatically part of my voice.

They may simply need editing.

Preserving voice does not mean protecting every sentence because it came from me.

The goal is to preserve what is distinctive without preserving what is unclear, repetitive or weak.

And voice is never a defense against weak reporting.

Accuracy still outranks style.

AI Often Falls Into Recognizable Patterns

Generative AI is very good at producing competent prose.

That is useful.

It also tends to fall into recognizable stylistic patterns.

Certain transitions.

Certain sentence structures.

Certain words.

Certain ways of summarizing an argument.

Different systems and prompts produce different results.

But if I accept enough suggestions without thinking, my writing can slowly begin sounding less like me and more like whatever patterns the system favors.

The problem is not one sentence.

It is accumulation.

One generic transition.

One overly polished paragraph.

One unnecessary summary.

One sentence that sounds impressive but not natural.

Eventually the article begins to drift.

That is why voice deserves its own editing stage.

Voice and Tone Are Not the Same Thing

Voice is broader than tone.

My voice is the pattern of judgment behind the writing.

The rhythm.

The emphasis.

The kind of details I notice.

The questions I ask.

The way I frame an argument.

Tone changes depending on the story.

A breaking news article should not sound like a reflective essay.

A game recap should not sound like an investigative feature.

A serious story may demand restraint.

Another piece may allow more energy.

Those articles can have different tones while still sounding like the same writer.

Preserving voice does not mean forcing every story into the same mood.

Ask What the Sentence Is Supposed to Do

When I edit, I do not only ask:

“Can this sentence sound better?”

I ask:

What is this sentence supposed to do?

Is it explaining?

Transitioning?

Establishing a fact?

Adding context?

Making an argument?

Creating emphasis?

If I understand the purpose of the sentence, I can judge whether an AI suggestion actually improves it.

Sometimes the original sentence is less elegant but more direct.

Sometimes the AI version sounds smoother but weakens the point.

Sometimes the best edit is not changing the sentence at all.

That is still editing.

Clearer Is Better Than Fancier

AI can make simple language sound more sophisticated.

That can be tempting.

A plain sentence becomes a polished sentence.

A direct statement becomes a more elaborate one.

The vocabulary gets bigger.

The rhythm becomes more formal.

But journalism does not need to sound complicated to sound intelligent.

If a simple sentence says exactly what I mean, I usually prefer the simple sentence.

The goal is not to make the reader admire the writing.

The goal is to make the reader understand the story.

Sometimes AI helps me get there.

Sometimes I have to reject the improvement.

And if voice gets in the way of understanding, clarity wins.

Read It Out Loud — But Do Not Treat That as the Only Test

One of the easiest ways I know to check voice is to read the article aloud.

A sentence may look fine on the screen and still sound wrong when spoken.

Too long.

Too formal.

Too stiff.

Too repetitive.

Or simply unnatural.

Reading aloud exposes rhythm.

It can also expose language AI made technically correct but awkward.

If a phrase feels unnatural when I read it aloud, I ask whether it belongs in the article.

But reading aloud is one test.

Not the test.

Written language does not have to sound exactly like conversation.

Some sentences work better on the page than they do in casual speech.

The larger question is whether the language fits the story, the audience and the writer.

Watch for Generic Transitions

AI loves transitions.

“In addition.”

“Moreover.”

“Ultimately.”

“On the other hand.”

“Taken together.”

None of those phrases are inherently wrong.

But repeated too often, they make the writing sound manufactured.

Sometimes the best transition is no transition at all.

The next paragraph begins because the next idea belongs there.

Sometimes a short sentence works better.

Sometimes a specific connection works better than a generic bridge.

I do not want the article to announce every transition.

I want the structure to carry the reader naturally.

Remove the Unnecessary Summary

AI also likes to summarize.

A section makes a point.

Then the final sentence explains the point again.

The next section opens by restating it.

Then the conclusion summarizes it once more.

The writing feels organized.

It also becomes repetitive.

One of the most common edits I make is removing sentences that explain something the reader already understands.

Trust the reader.

If the paragraph did its job, it does not always need a sentence telling the reader what the paragraph just did.

Preserve the Sharp Lines

Some of the strongest lines in my writing are short.

They may not sound elaborate.

They may not even sound especially polished.

But they carry the idea.

“AI output is not evidence.”

“The evidence decides.”

“Agreement does not test the work.”

Those lines work because they are direct.

An AI editor might expand them.

Explain them.

Qualify them.

Make them sound more sophisticated.

That can weaken them.

Part of editing is knowing when a sentence has already done enough.

Voice Comes From Judgment

Voice is not just vocabulary.

It is not slang.

It is not sentence length.

Voice comes from decisions.

What I notice.

What I question.

What I emphasize.

What I refuse to overstate.

What I think deserves another sentence.

What I think needs to be cut.

Two journalists can use the same facts and produce very different stories because they make different editorial decisions.

That is why preserving voice cannot be reduced to telling AI:

“Make this sound like me.”

The system does not fully know what that means.

The better approach is to preserve the decisions that make the writing mine.

AI Can Flatten Voice — or Help Reveal It

AI does not automatically erase voice.

Used carefully, it can help reveal it.

It can show me where I am repetitive.

Where the rhythm slows down.

Where a sentence is weaker than the surrounding writing.

Where generic wording replaced a specific idea.

Where I keep falling into the same phrasing.

That gives me information I can use.

The danger comes when I allow the system to make all the decisions about how those problems should be fixed.

AI can help expose the weakness.

I still decide what the replacement should sound like.

Do Not Let AI Make Everything Neutral

There is another risk.

AI can make language so balanced and cautious that the writing loses force.

Journalism should be fair.

That does not mean every sentence needs to sound emotionally flat.

If the evidence clearly establishes something, the writing can say it clearly.

If something failed, I do not need to write around the failure.

If a performance was poor, the article can describe it honestly.

If the evidence is strong, excessive hedging can become its own distortion.

The goal is not maximum caution.

The goal is accurate confidence.

But Do Not Let Style Outrun the Evidence

The opposite problem exists too.

AI can make prose more dramatic.

“Historic.”

“Stunning.”

“Devastating.”

“Unprecedented.”

“Game-changing.”

Those words may be justified.

They may also be empty.

A strong adjective makes a claim.

If I cannot support that claim, the word does not belong.

Editing for voice does not mean making everything more colorful.

It means making the language match the evidence and the writer.

Keep the Specific Details

Generic writing often comes from losing specific detail.

A sentence says a team “showed resilience.”

What did the team actually do?

A company “faced challenges.”

Which challenges?

An organization “made significant changes.”

What changed?

Specific detail gives writing credibility and personality.

AI can sometimes replace specifics with broad language because broad language flows more easily.

I try to do the opposite.

If the detail matters, keep it.

Specific beats generic.

Editing Can Change Facts

Editing feels like a language task.

It can become a factual task very quickly.

Changing a word can change the meaning.

“Led” is not the same as “won.”

“First” is not the same as “one of the first.”

“Caused” is not the same as “followed.”

“Confirmed” is not the same as “reported.”

A cleaner sentence can become a less accurate sentence.

That is why editing and verification cannot be completely separated.

Any edit that changes the substance of a claim deserves another look at the evidence.

And the more consequential the claim, the less casually I accept a rewrite.

Track What AI Changed

When AI makes substantial edits, I want to know what changed.

Did it cut context?

Change attribution?

Strengthen a verb?

Remove uncertainty?

Combine two sentences in a way that changes meaning?

Add language that was not in the original draft?

I do not need to obsess over every comma.

But I do need to pay attention when the edit changes the journalism.

The bigger the rewrite, the more carefully I review it.

Sometimes the Best Prompt Is “Do Less”

AI often assumes I want more.

More polish.

More explanation.

More structure.

More detail.

Sometimes I want the opposite.

Tighten this without changing the tone.

Remove repetition but keep the phrasing.

Fix grammar only.

Do not rewrite the voice.

Preserve the short sentences.

Do not add transitions unless necessary.

Those instructions matter.

The more specific I am about what I want preserved, the less likely the edit is to become a rewrite.

Compare Before and After

One useful habit is comparing the edited version with the original.

Not because the original is automatically better.

But because comparison makes the tradeoffs visible.

What improved?

What disappeared?

What became clearer?

What became generic?

Did the rhythm improve?

Did the argument become weaker?

Did the language become more accurate?

Did the writing still sound recognizably mine?

Editing should produce a net improvement.

Not simply a different version.

But comparison does not remove bias.

I can still prefer a version simply because I wrote it.

So the final question cannot be only:

“Which version do I like?”

It also has to be:

“Which version does the evidence support?”

Personal Voice Still Lives Inside Editorial Standards

My voice matters.

It is not the only standard.

Journalists often write within publication rules, editorial policies, legal requirements and house style.

A personal preference does not override those standards.

The goal is to preserve individuality inside the responsibilities of the publication.

That balance matters.

A strong personal voice should make the journalism more distinctive.

It should not make the journalism less disciplined.

The Final Read Is Mine

At some point, I have to stop asking the AI what it thinks.

The article belongs to me.

So I do one final read as the writer.

Not as the researcher.

Not as the fact-checker.

Not as the person managing the workflow.

As the person whose name is attached to the work.

Would I say this?

Does this sound natural?

Does this represent the reporting fairly?

Is anything here technically correct but still wrong in tone?

Is there anything I would be uncomfortable defending?

That final read matters because voice and responsibility meet at the same place.

The byline.

The Goal Is Clarity Without Losing Myself

My goal is for AI-assisted editing to make the writing clearer without making it less recognizably mine.

Sometimes the AI suggestion stays.

Sometimes I rewrite it.

Sometimes I restore the original.

Sometimes I create a third version.

The value comes from the comparison.

The challenge.

The options.

Not the automatic replacement.

AI can flatten voice.

It can also help reveal it.

The difference is whether I remain the person making the final decisions.

Do Not Outsource the Final Voice

AI can help me write.

It can help me edit.

It can help me identify problems.

It can help me find a better way to say something.

But the final voice should still belong to the person responsible for the journalism.

Because voice is more than style.

It is judgment made visible in language.

And if the journalist gives that away completely, something important disappears.

The article may still be clean.

It may still be polished.

It may even be technically correct.

But it may no longer sound like anyone.

That is not the kind of journalism I want to publish.

I want the article to sound like me while still serving the reader, the evidence and the standards of the publication.

Clearer.

Sharper.

More precise.

More readable.

But still mine.


Graphics created with AI-assisted design tools.
Visual concepts, editorial direction, composition, and final creative decisions were developed by Vallano Media.