You open AI to improve a sentence. Ten minutes later, it has changed the argument.

That can be useful. It can also mean that the most important decision in AI-assisted writing, what AI is allowed to shape, was never really made.

I kept catching myself doing exactly that. I knew what I wanted from the piece, but not always whether I wanted AI to improve the thinking, the wording, or both. The process could be productive, but it could also become a blur of better ideas, better sentences and unexamined hand-offs.

So I went looking for a simple way to make that division of labour visible before I write, notice when it shifts while I work, and learn from it afterwards. That search became the Thinking × Wording Lens.

If this sounds familiar, open the Lens now and keep it beside this article. Pick a position that feels plausible for something you are working on. Then read on. The interesting part is not where you start, but what the map reveals as the work moves.

What AI has pulled apart: Writing = Thinking × Wording

Writing has always involved at least two jobs: deciding what to say and finding the words to say it. They often happened so closely together that we treated them as one activity. Generative AI makes it possible to separate them much more explicitly.

You can bring the observation and judgement while AI helps formulate the prose. You can use AI to challenge your thinking and still write every sentence yourself. You can develop both together, or set a task and let AI carry most of both. All are truthfully “using AI”, which is why that phrase tells us very little.

“The useful distinction is not one scale of AI-ness. It is two variables: where the thinking comes from and where the wording comes from.”

Even a perfect AI scanner could not tell you whether AI polished sentences, challenged an argument, opened a research path or supplied most of the thesis.

That raises a related question I have been circling in earlier Perspectives: perhaps authorship is becoming a more useful way to talk about the value of a piece than the old shorthand of who “wrote” it. Authorship is not a percentage. It is the presence of an essential contribution. The Lens does not measure authorship, but it can help make the provenance of that contribution visible: in the thinking, the wording, the direction, the judgement, or across the process.

Why × and not +

The multiplication sign is deliberate. In one important sense it is mathematical: if either thinking or wording is zero, the outcome is zero. No substantive thought means there is nothing worth carrying; no wording means nothing reaches the reader.

I am not proposing a numerical quality formula. The point is interdependence. Polished wording cannot rescue empty thinking, and strong thinking cannot travel without usable words.

The mapping

I chose three positions on each axis rather than a binary human-or-AI split because the middle matters. Thinking and Wording can each be Mainly human, Developed together, or Mainly AI. Put those together and nine positions appear, with Thinking on the horizontal axis and Wording on the vertical. “Together” means substantive contribution from both. It does not mean a 50/50 split.

The codes are coordinates, not grades. T1W3 means mainly human thinking with mainly AI wording; T3W1 means the reverse. The question is not which square sounds most virtuous. It is which combination fits this particular work, and why.

The map describes provenance, not quality. Mainly human work can be poor; Mainly delegated work can be entirely fit for purpose. AI-worded content can help expertise travel faster or flatten distinctive thinking. Human-edited AI thinking can be useful for synthesis or make weak reasoning look more convincing. The map makes those choices discussable. It does not make them for you.

The position belongs to the work, not the writer

You are not a T1W2 or T2W2 “type”. A position describes a piece of work, or even a phase of it. A draft may start with mainly human thinking, move into developed-together thinking when AI surfaces a useful angle, then return to mainly human wording for the final voice. Movement is normal.

Before, during and after

Let’s say you are writing a CEO point of view. Before you start, you decide the judgement and conclusion should remain mainly human, but you want AI to challenge assumptions, so you aim for Think Collab: thinking developed together, wording mainly human. During the work, AI surfaces an angle that changes the argument and then starts helping substantially with the phrasing, so the piece moves towards Interplay. Before publishing, you map where it actually ended up and ask whether that still fits the purpose. If not, you change the process or rewrite. The value is not staying in one square. It is noticing, choosing and being able to explain the choice.

“The value is not staying in one square. It is noticing, choosing and being able to explain the choice.”

For teams, clients and agencies

The Lens becomes more useful when work involves more than one person. Before arguing about tools, a team can discuss the division of labour: which content deserves mainly human thinking, where developing together helps, where AI-led wording is efficient, and what guardrails follow. Different content types can legitimately lead to different answers.

“AI makes us more efficient” is incomplete until we agree what is becoming more efficient: routine wording, research, or judgement and strategy.

The Lens does not decide fees, scope or responsibilities. It can make the underlying bargain explicit first: what is the client buying, which human contribution remains part of the value, what can AI accelerate, and what still deserves human review?

A boundary: you need access to the process

This is not a forensic tool for judging a stranger’s finished prose. I can map my own process, or one I share with a colleague, client, agency or partner, because the people involved know what happened. Looking only at somebody else’s final text, I usually cannot know where either the thinking or wording substantially came from.

That is one reason I prefer this to scanner-policing. It asks participants to make the process visible rather than pretending provenance can be reverse-engineered from fluent prose. Without access to the process, the honest answer is: we do not know.

A lens, not a verdict

The Thinking × Wording Lens maps provenance, not quality. It is not a detector, rating, certification, compliance checklist or disclosure standard. A T1W1 piece can be poor; a T3W3 piece can be entirely fit for purpose. Accuracy, ethics, confidentiality, legal obligations and disclosure still require their own judgement.

This Lens isn’t finished. I’d like your hand in it.

I built it for myself first, and I am publishing it before pretending it is finished. A small international Feedback Circle has already sharpened it. Their recurring challenge: prove the practical value, avoid turning it into a lower-is-better AI percentage, and allow real processes to move.

One question remains open: is this useful beyond five minutes of self-reflection? I think the answer lies in the decisions it enables for specific content, teams and client relationships. I would rather test that in public than declare it solved.

So I would like fellow content creators to push against it. Where does the map break? Which labels mislead? Which cases would make it more useful? I will fold what I learn into successive versions, with reasoned decisions and credit where contributors want it. The framework is CC BY 4.0.

Explore the live Lens at bertvanloon.com/txwx/.

The Open Lab will follow as a public place to challenge, improve and add real cases. Open to contributions. Curated by Bert. Improved through iteration.

I have said that line to a lot of rooms full of content people: “We are content creators first, second and third. Marketers somewhere after that.” It is deliberately overstated. What bothers me is how few fellow marketers look offended.

Most of us came into this profession because we love ideas, stories, writing, video, editorial thinking, formats, craft. So that is what we got good at.

We built editorial calendars, production processes, content operations, increasingly sophisticated workflows. And then, usually somewhere near the end, somebody would ask what we were going to do with the thing now that it was finished. Which channels? Any paid budget? Who is posting it on LinkedIn?

A few years ago, Content Marketing Institute asked for my input for its annual trends update on where content marketing budgets were being wasted. I said over-creation and under-distribution, and reduced it to four words: less content, more marketing. I sometimes joke that we should have called the discipline marketing content instead. It is only a joke because there is enough truth in it to sting.

Since then, I have realised my diagnosis was too narrow. Distribution is only the visible half of the problem.

I have been drawing the whole thing on whiteboards for years as a shelf. Content is the row of books in the middle. Marketing is the two bookends holding it upright, one at each end. But bookends as most people picture them are flat and closed. Marketing turns up, props the work in place, and leaves.

So, I draw mine with a porch. An open, covered space on the inside face of each bookend, where the two disciplines are meant to sit together for a while.

The front porch is where marketing comes in before anything is made, and it brings two things. One is the business objective the content exists to support: what needs to change in the market, in someone’s mind, in behaviour, preference, demand or retention. The other is what we know about the audience, arriving early enough to shape the content rather than to approve it. Between them they answer the question we skip most reliably: why should this particular thing exist at all?

The back porch is where marketing supports the finished work on its way out: how the idea finds those people, how it behaves in the environments where they meet it, and what should happen once they have consumed it.

Most content operations have the bookends. Almost nobody builds the porches. Marketing arrives at the last minute, holds the work upright long enough for it to be published, and goes home.

“Marketing should not begin when the content is finished. It should decide what deserves to be created.”

The bottleneck moves

Then AI arrived, and at first it looked like it would make the imbalance permanent. Research moves faster. Drafts get challenged and improved. One idea becomes five formats. A team of three produces what used to take twelve.

All of which matters less than it sounds, because the part of the job we were already best at is the part that just got cheaper. When your strongest muscle gets a machine attached to it, the relative value of your weakest one goes up. We can now produce more content, but audiences did not acquire more hours in the day. We can publish in more places without understanding any of them better. We can generate an impressive volume of perfectly competent material without getting any better at deciding what deserves to exist.

“The bottleneck moved, and most of us are still standing at the old one.”

Make AI your difficult client

Here is the part I did not expect.

I spent two years worrying that AI would become another excuse to make more content. It still might. But the same technology is also the first practical way I have found to force marketing thinking into the process before anything gets made. Not by asking AI to be the marketer, which it cannot be, but by giving it the one role content teams systematically avoid.

Make it the difficult client. The one who refuses to nod politely, does not care how much work you have already put in, and keeps asking why.

Before you ask AI to research anything or draft anything, give it the actual brief: the business problem, the audience, what you know about them, what you believe needs to change, what evidence you have, where you expect resistance, which channels matter, and what outcome would make the work worth doing. Then tell it to interrogate you, not to help you.

My own prompt is three words: roast my idea. The feedback stings, but considerably less than publishing something nobody finds, reads or remembers. What comes back is the set of questions we are best at skipping:

• Are we solving a real audience problem, or filling the calendar?
• What assumption are we making about these people that we have not tested?
• Why would they care about this now?
• Is content even the right response?
• What would have to be true for this to work?

Keep the client in the room

The brief does not have to disappear once creation starts. Most of the damage happens after it: the piece drifts towards what interests us rather than what matters to the reader, the useful part ends up halfway down the page, we start assuming knowledge the audience does not have. Feed the marketing intent back in as the work evolves and ask whether it still serves the problem you identified. That is a different use of AI from asking it to make a paragraph punchier.

It changes the other end too.

A LinkedIn post is not a shortened article. A newsletter is not a dumping ground for whatever appeared on the website that week. A conference talk is not a blog post with slides. Yet the single most common use of AI in content teams remains mechanical: turn this article into five LinkedIn posts. That is not distribution. That is photocopying.

“That is not distribution. That is photocopying.”

Give AI the original marketing objective and the context of the channel instead, and ask which part of the idea deserves to travel there, what would have to change for it to work in that environment, whether the audience you want is even present, and what somebody should be able to do next if the idea earns their attention. Those are marketing decisions. Some belong on the front porch before the work is made; others belong on the back porch before it goes out into the world. None belongs in a last-minute export menu.

And once something is published, the same difficult client is useful again: which framing attracted the audience you actually wanted, which questions kept recurring, what produced a spike and nothing else, what deserves another iteration, and what should quietly be allowed to die. That is not reporting. It is the next brief, arriving early.

The muscle we keep training

Which brings me to what bothers me most about how our profession is currently learning AI. We gravitate towards using it on the part of the job we already know. Writers learn AI writing. Video people learn AI video. Content teams learn to produce, adapt and repurpose faster. Of course we should learn those things.

But content creators should spend less of their AI-learning time becoming faster at content creation, and more of it becoming better marketers.

I have argued for years that this industry needed less over-creation and more marketing. The fix, it turns out, may involve the greatest content-production accelerator anyone has ever handed us — pointed at the brief as often as the draft. It can sit on either porch with us, keep asking why, and make it much harder to hide the marketing questions we failed to answer.

Maybe AI will not just change how we make content. Maybe, finally, it can help us put the marketing back into content marketing.

For the first few years of generative AI I talked happily about AI-empowered professionals. Give capable people AI and they become more capable: faster researchers, better writers, more productive consultants, sharper marketers.

I’ve changed my mind about which way that sentence should run.

Not because it’s wrong. AI really does make people more capable. But the sentence starts in the wrong place. It starts with the technology arriving and asks what it does to us. The more useful question is what we hand it before it does anything at all.

Take the demo everyone has already run. Five friends each open the same AI assistant — Claude or ChatGPT, say — and ask it to plan a reunion trip to Rome. They end up with five different plans, ranging from genuinely good to unusable. Much of that difference will sit on the human side of the exchange: what they asked, what they told it, what they challenged, what they accepted.

The phrase human-empowered AI is decoration unless I can say what a person contributes that the system cannot simply supply for itself. At least four things come to mind.

The first is the problem. Not the request, the problem. The model receives what you typed, which is rarely the thing you actually need solved. Someone has to make that conversion, and it isn’t a prompting technique.

The second is context that the model cannot know unless you bring it: which constraints are real and which are negotiable, what somebody said in the room, why the obvious approach failed here two years ago.

The third is a standard. A model will produce a plausible answer to almost anything. Deciding whether a plausible answer is a good one requires a picture of quality you brought with you. If you don’t have that picture, you can keep accepting fluent work indefinitely.

And the fourth is accountability. When the output goes out under your name, you own it. That cannot be delegated to the thing that produced it.

AI can help us articulate the problem, surface relevant context and challenge our standards. But none of those can simply be assumed to come from the model. They are part of what makes its contribution useful in the first place.

“The power doesn’t start with the AI. It starts with what we bring to it.”

There is a complication I have to take seriously, because it cuts against me. The 2025 Quarterly Journal of Economics study Generative AI at Work followed 5,172 customer-support agents and found the strongest gains among the less experienced and lower-skilled. AI helped them adopt some of what the stronger performers already did.

Nobody in that study had to define the problem from scratch. The work was defined, the workflow was known, and the tool was dropped into it. It worked. So the honest version of my argument is narrower than the slogan: where the problem is already well framed and quality is already well defined, you can add AI to an existing process and create real value. That is not a minor exception. It is one of the clearest ways organisations can benefit from AI.

I see organisations making much the same mistake at scale. The seductive route is familiar: buy the tools, roll out Copilot, give everyone access, find a hundred use cases, run prompt training, add an agent here and a chatbot there, total up the hours saved and announce that the organisation is becoming AI-enabled.

None of that is wrong. But it can leave one rather important question unasked: should the process we’re accelerating still exist in this form at all?

Adding AI to a process can improve it. It can also help an outdated process produce its outdated result faster.

“AI is a terrible reason to put a turbocharger on a process that no longer deserves an engine.”

McKinsey’s State of AI in 2026 found that organisations reporting the greatest value from AI were far more likely to be fundamentally redesigning workflows rather than simply adding AI to existing ones. That doesn’t establish causality, but it is consistent with the distinction between AI adoption and AI transformation.

There is, however, a stronger objection to my argument. Sometimes you cannot know what genuinely better looks like until you’ve played with the technology. Nobody sat down and specified all the things a spreadsheet should eventually be used for before building one. Search, version control and countless other technologies found important uses because people first got access to a new capability and started messing around with it. AI is no different. People discover possibilities by using it.

So start with the outcome can become bad advice if it means don’t explore until you already know what you’re looking for. I don’t think those positions actually conflict. Exploration should be capability-led. Give people the tools, let them experiment, try things whose value you cannot yet predict and see what becomes possible.

But exploration and commitment are different activities. The mistake comes when nobody goes back and asks the outcome question before turning an experiment into a workflow, an agent, a platform, a headcount decision or a budget line. A promising experiment gets hardened into a permanent process before anyone checks whether that process deserves to exist.

“Explore from the capability. Commit from the outcome.”

Even exploration depends on something human. Somebody still has to recognise that what they have stumbled onto is interesting, useful or dangerous. Judgement has not disappeared. It has simply entered earlier in the process.

Once we move from exploration to actually changing how work gets done, I would stop beginning with Where can we use AI? I would begin with What are we trying to accomplish? For whom? Why does it matter? What would genuinely better look like?

Then comes the question of what the work requires. Where is human judgement essential? Where do context, empathy, expertise or accountability decide the outcome? Where can AI contribute speed, scale, synthesis or pattern recognition? What combination produces something better than either could produce alone?

Only then does it make sense to ask what the workflow should become — and what technology belongs in it.

It sounds obvious. Yet a great deal of AI adoption still runs the other way: technology, use cases, workflow and eventually, sometimes, purpose. That is how an organisation can become extremely proficient at using AI without becoming noticeably better at creating value with it.

And this gets more important as the technology improves, not less. Access will stop being what distinguishes anyone. The models will get better. The tools will spread. Today’s remarkable capability becomes tomorrow’s ordinary infrastructure.

What stays scarce is everything around it: the problems you choose, the questions you ask, the context you hold, the standards you set, the judgement you apply and the responsibility you take for the result.

I still believe in AI-empowered professionals. AI will make good professionals dramatically more capable, and it will let less experienced people do things they couldn’t do before. That’s the effect. Human-empowered AI is where the story starts.

We didn’t get the destination wrong. We got the arrow backwards.

Imagine briefing Michelangelo for the Sistine Chapel.

At one extreme, you could say: Please paint the ceiling. Plenty of freedom, not much help. At the other, you could specify the colours, scenes, figures and positions, perhaps ask for three sketches by Thursday. Plenty of instruction, not much Michelangelo.

Many years ago I came across a presentation built around exactly that thought. Damian O’Malley imagined what the brief for Michelangelo’s Sistine Chapel ceiling might have looked like, and Steven Stark turned the idea into a slide deck.

The cleverness was in what happened between those two extremes. A good brief could be clear about the ambition and meaning of the work while leaving the artist enough room to do something the commissioner could not have designed himself.

I have thought about that presentation surprisingly often over the years, partly because marketing has spent roughly the same period complaining about briefs.

Agency people complained that clients could not write them. Clients complained that agencies did not follow them. Creatives complained they were either vague enough to mean anything or detailed enough to make creativity unnecessary. And almost everyone has at some point received some version of: Make it innovative, but on brand. Be bold, but don’t frighten anyone. We want something disruptive. By Friday.

The remarkable thing is not that so many briefs were bad. The remarkable thing is how much good work still came out of them.

There was a reason for that: for years, great debriefing compensated for bad briefing.

A good strategist, creative, consultant or agency team rarely treated the written brief as holy scripture. They asked questions. They listened for the thing the client kept circling around without quite saying. They knew the organisation, the politics, the category and sometimes the client well enough to recognise that “we need a campaign” actually meant “we have no compelling reason for anyone to care about this product yet.”

They read between the lines. Occasionally they read between lines that weren’t even there.

In other words, capable humans became remarkably good at rescuing those who gave them mediocre instructions.

And then along came AI.

One of the more educational features of working with generative AI is its slightly irritating tendency to expose what you actually asked for rather than what you vaguely remember meaning. Ask a generic question and there is a decent chance you will receive a generic answer. Forget important context and the machine will not necessarily infer the particular piece of context that seemed blindingly obvious inside your own head. Give five conflicting ambitions and it may cheerfully attempt all five.

Then we look at the result and declare: AI is producing rubbish.

Sometimes it is. But sometimes a disappointing AI answer is less a failure of the model than an X-ray of the brief.

“Sometimes a disappointing AI answer is less a failure of the model than an X-ray of the brief.”

That is one reason I am increasingly sceptical about treating “prompt engineering” as an exotic new discipline built around secret incantations. Of course there are techniques worth learning, and the technology has its peculiarities. But much of what makes an AI instruction better is remarkably familiar.

What are we actually trying to achieve? For whom? In what context? What is already known? What matters most? What constraints are real? What should not happen? What does good look like? And where do I want the recipient to follow my instructions versus use judgement of its own?

Prompting may be new. Briefing isn’t.

Perhaps the most important part is the last one, because a good brief is not an instruction manual.

If I already know precisely what the answer should be, I don’t need much intelligence on the receiving end. I need execution. The reason we brief talented humans, and increasingly capable machines, is precisely because we expect something from them that we did not completely specify in advance.

That creates the tension at the heart of good briefing.

Too little direction and we create drift. Too much and we create obedience.

Freedom without direction creates drift. Direction without freedom creates obedience. Good briefing creates room for judgement.

“Freedom without direction creates drift. Direction without freedom creates obedience. Good briefing creates room for judgement.”

Somewhere while getting better at giving instructions to AI, I realised that this was no longer just a marketing lesson. Briefing and managing share a surprisingly difficult task: defining intent and boundaries without defining the answer.

Good managers do not simply throw an objective over the wall and call that empowerment. Nor do they specify every move and then wonder why nobody shows initiative. They establish what matters, what is fixed, where the risks lie, what success means and where another person is expected to think.

In other words: clear intent, scoped instructions and defined freedom.

The point is not that managing humans is the same as prompting machines. It very obviously isn’t. Humans bring relationships, emotions, tacit knowledge, ambitions, insecurities, institutional memory, dissent and the occasional desire to spend the afternoon doing something entirely different. Long may they continue to do so.

Good people have always been remarkably good at protecting managers from their own bad instructions. AI is less discreet.

That may turn out to be useful. AI gives us something close to a high-speed practice environment for delegation. You give direction. You see the result seconds later. You notice what was missing. You clarify the objective. You remove an unnecessary constraint. You explain what you meant by “more strategic,” because apparently that phrase was not quite the masterclass in precision you had imagined. Then you try again.

A feedback loop that might take days in an organisation can happen repeatedly in minutes.

We thought AI would mainly accelerate production. It may also be accelerating our education in giving direction.

“We thought AI would mainly accelerate production. It may also be accelerating our education in giving direction.”

And that lesson is likely to outlast today’s prompt techniques. Models will change. Interfaces will change. Many clever prompting tricks will disappear into the products themselves.

Being able to formulate intent, establish context, make choices, articulate constraints and leave useful room for judgement will not.

If anything, that becomes more important as the intelligence available to us increases. The better the person on the receiving end of a brief, the less sense it makes to prescribe everything they should do. As AI becomes more capable, the same principle starts to apply there too.

Capability raises the potential value of freedom, but only when freedom has meaningful direction.

Which takes me back to Michelangelo.

The point of that old Sistine Chapel briefing story was never that the commissioner should simply have said: You’re Michelangelo. Surprise me. That would not have been a great brief either.

The magic lies somewhere between control and abdication: enough clarity about purpose to point the work somewhere meaningful, enough boundaries to make the problem real, and enough freedom for somebody else’s intelligence to contribute something you could not have specified yourself.

For decades we relied on good people to repair bad briefs downstream. AI is pushing more of that responsibility back upstream, towards the person giving the instruction. That feels like progress.

And it suggests a simple test before you brief a colleague, agency or AI.

Can you answer four questions clearly: What are we trying to achieve, and for whom? What is fixed, and what is genuinely open? What does good look like? And where do I want execution versus judgement?

If you cannot answer those, another clever prompt probably will not save you.

If you can, resist the temptation to specify the rest.

The better the intelligence on the receiving end, the less useful it is to specify the answer. Our job is to specify the intent.