AI can give you ten headlines, a campaign structure or a decent first draft in minutes.

That is useful. It is not the same as good strategy.

Someone still has to decide whether the brief is right, the claim is true, the idea fits the audience, the context changes the answer and the work is distinctive enough to publish.

AI has made output cheaper. The valuable part is increasingly knowing what deserves to move forward.

Subscribe now

I use AI every day.

It helps me get through a blank page faster. It can organise messy notes, challenge an idea, generate alternative routes and turn an unstructured thought into something I can react to.

All of that is useful.

I have used it to generate ten headline routes in a few minutes. The time saving is real.

But the work after that is still human.

Which route fits the audience?

Which claim can be supported?

Which line sounds like the organisation?

What could be misunderstood?

And was the brief asking the right question in the first place?

AI has made producing a first version much easier.

It has not made those decisions easier.

Subscribe now

AI has lowered the cost of plausible work

AI use is no longer experimental for most organisations.

Stanford’s 2026 AI Index reports that 88% of surveyed organisations used AI in 2025, while 70% used generative AI in at least one business function.

The significance is not only that more teams are using the tools.

It is that producing a credible-looking first version of almost anything is becoming easier.

A basic audience profile.

A landing-page structure.

A content plan.

A campaign concept.

A research summary.

A slide deck.

Ten variations of the same headline.

The ability to produce these things used to create a certain amount of professional advantage.

It does not create the same advantage anymore.

When everyone can generate acceptable-looking work quickly, the valuable part shifts.

It moves from producing the answer to deciding:

• whether the question was right

• whether the answer is true

• whether it reflects the actual audience

• whether it fits the context

• whether it is distinctive enough to matter

• whether it should be used at all

Producing the work becomes easier.

Deciding what is worth using does not.

Subscribe now

The real risk is confidently adequate output

Obviously bad AI output is relatively easy to deal with.

An invented fact can be checked.

An awkward sentence can be rewritten.

A completely irrelevant recommendation can be discarded.

The more difficult problem is output that is confidently adequate.

It reads well.

It is logically structured.

It uses the right vocabulary.

It sounds like something a competent professional might say.

Nothing is obviously wrong.

But nothing is particularly true, distinctive or useful either.

This kind of work moves quickly through organisations because it creates very little resistance.

It does not look risky enough to challenge.

It does not look weak enough to reject.

It is simply good enough to pass.

And when teams are under pressure to produce more, “good enough to pass” can become the standard surprisingly quickly.

Fluency is not evidence

AI can make an assumption sound like a conclusion.

That is one reason human review needs to go beyond proofreading.

Consider an AI-generated audience summary that says:

This audience values authenticity, convenience and purpose-driven brands.

It sounds reasonable.

It may even be true.

But it is also broad enough to describe a significant proportion of the internet.

Where did the insight come from?

Which audience behaviour supports it?

Does it hold across markets?

What would it change about the strategy?

What would we do differently if the opposite were true?

A polished sentence can hide the fact that no real diagnosis has happened.

The first review question should not be:

Does this sound good?

It should be:

What evidence supports it, and what decision does it help us make?

Subscribe now

Generating options is not the same as strategy

AI is extremely useful for expanding the option set.

It can help a team consider:

• different audience framings

• alternative creative hooks

• objections that have not been addressed

• new content structures

• variations for different markets

• weaknesses in an early idea

• arguments against its own recommendation

But strategy is not the existence of many options.

Strategy is choosing between them.

That choice requires an understanding of organisational priorities, audience behaviour, evidence, risk, capacity, timing and trade-offs.

There is rarely one objectively correct marketing answer.

Should the campaign prioritise immediate conversion or attracting a new audience?

Should a global team insist on consistency or allow more market flexibility?

Should the message use urgency or create more room for consideration?

Should the organisation act with incomplete evidence or wait for a stronger signal?

A model can help surface the options.

It can help explain the trade-offs.

It cannot take responsibility for the choice.

The organisation still has to decide what matters most.

Subscribe now

The generic answer usually breaks when it meets the real context

The same piece of copy can work in one context and be careless in another.

The same creative technique can build urgency for one audience and mistrust for another.

A message written for an existing customer may make no sense to someone encountering the organisation for the first time.

A global campaign line may be easy to translate and still feel completely irrelevant in the market.

A tactic that is acceptable for a low-risk consumer decision may be inappropriate in healthcare, finance, humanitarian communication or any context involving vulnerable people.

Context is not a final editing layer.

It changes the answer.

That is why experienced people sometimes appear slower than an AI system.

They are not only producing the line.

They are considering what the line means here, for this audience, at this moment, with these possible consequences.

Subscribe now

When competent content is everywhere, distinctiveness matters more

AI is raising the baseline quality of a lot of professional communication.

That is mostly a good thing.

Weakly structured writing can become clearer.

Teams with limited resources can produce stronger first drafts.

People can explore ideas they may not have had the time or confidence to develop before.

But as competent content becomes easier to produce, competence alone becomes less noticeable.

A technically sound article can still be forgettable.

A clean campaign concept can still look like everything else in the category.

A perfectly structured LinkedIn post can still give no indication that a person with real experience wrote it.

Distinctiveness does not mean being provocative for attention.

It means having something specific enough to say that another informed person could disagree with it.

It may come from:

• a pattern you have noticed repeatedly

• a trade-off most advice ignores

• a detail that only appears when strategy reaches execution

• a market reality that contradicts global best practice

• a decision you would make differently and why

AI can help articulate a point of view.

It cannot manufacture the experience that made the point of view worth having.

Subscribe now

“Human in the loop” is too vague

Most organisations now understand that AI output should receive human review.

The phrase sounds reassuring.

It is not always operationally useful.

Which human?

Reviewing what?

Against which standard?

At what stage?

With the authority to change or stop the work?

NIST’s guidance on generative AI recommends defining ownership, roles and responsibilities for human oversight rather than treating human involvement as a vague safeguard.[2]

That principle applies well beyond technical AI systems.

A low-risk brainstorm does not need the same review process as:

• a public factual claim

• customer-facing advice

• content based on confidential information

• communication in a politically sensitive market

• health or humanitarian messaging

• a recommendation influencing significant investment

“Someone checked it” is not an operating model.

Teams need to know what kind of judgment the work requires and who owns the final decision.

Subscribe now

A practical checklist for reviewing AI-assisted marketing work

This is the checklist I find useful when reviewing AI-assisted work.

1. Are we solving the right problem?

AI responds to the brief it receives.

It does not automatically know that the brief is wrong.

A request for “more emotional copy” may really be a proposition problem.

A request for “more top-of-funnel content” may actually be a distribution problem.

A request for “better-performing ads” may be hiding poor audience quality, a weak journey or low retention.

Before reviewing the answer, review the question.

2. What evidence supports it?

Check the:

• facts

• sources

• dates

• examples

• quotations

• causal claims

• audience assumptions

• market-specific statements

Plausibility is not verification.

A confident answer can still be built on weak or missing evidence.

3. Does it reflect the actual audience?

Ask:

• What does the audience already know?

• What stage of the journey are they in?

• What might they misunderstand?

• What would make this relevant?

• What would make it feel obvious, manipulative or out of touch?

The same content should not be expected to attract a new audience, reassure someone considering the offer and close someone ready to act equally well.

Different stages require different work.

4. What changes because of the context?

Review the work for:

• cultural meaning

• language and translation

• political sensitivity

• legal or policy requirements

• power dynamics

• risk to vulnerable audiences

• what the organisation can realistically promise

A generic answer can be competent and still be wrong for the situation.

5. Could any organisation publish this?

If the answer is yes, the work may need more human input.

Look for:

• a clearer point of view

• a real example

• a specific trade-off

• a non-obvious observation

• language that sounds like the organisation

• a reason this content should exist

The goal is not novelty for its own sake.

It is relevance that could not be produced simply by replacing the brand name.

6. Who owns the final decision?

The person publishing or acting on the work remains responsible for:

• accuracy

• appropriateness

• risk

• brand impact

• audience consequences

• the decision to use the output

AI can assist.

It cannot be accountable.

Subscribe now

Treat AI like a very fast intern, not the boss

The analogy is imperfect, but useful.

A capable intern can:

• research a starting point

• organise information

• generate options

• draft material

• identify questions

• get through repetitive work quickly

But you would not give an intern a vague brief, remove all context, accept the first answer and allow them to make a high-stakes decision alone.

You would explain the objective.

Provide the right background.

Check the sources.

Challenge the assumptions.

Give feedback.

Decide what is ready.

And remain accountable for what goes out.

The same should be true with AI.

Use it for speed.

Use it for range.

Use it to make the blank page less blank.

Use it to show you an argument you disagree with clearly enough to improve your own thinking.

Just do not quietly promote it above its level of judgment.

Microsoft’s 2026 Work Trend Index found that 86% of surveyed AI users treated AI output as a starting point rather than a final answer. Respondents also identified quality control and critical thinking as increasingly important human capabilities.[3]

That sounds less like AI replacing the work and more like the work changing.

The first version becomes easier.

Evaluating, refining and owning it becomes more important.

Subscribe now

Judgment is not one mysterious human quality

“Human judgment” can sound abstract.

In practice, it is a combination of capabilities:

• critical thinking: assessing whether a conclusion follows from the evidence

• audience understanding: recognising what is relevant and appropriate

• taste: knowing what feels coherent, distinctive and worth attention

• context: understanding the environment in which the work will land

• prioritisation: deciding what deserves time and investment

• risk awareness: identifying what could go wrong and who could be affected

• accountability: being willing to own the decision

• restraint: knowing when not to produce, publish or automate

These are not anti-technology skills.

They are what make the technology useful.

Using AI well does not mean using it for everything.

It means knowing where it adds value, where it introduces risk and where a person needs to remain closely involved.

Subscribe now

Authenticity does not mean pretending AI was never involved

I do not think the goal is to make work look as though AI never touched it.

Nor does using AI automatically make the work inauthentic.

Authenticity does not come from manually typing every word.

It comes from whether the thought is real, the judgment is visible and someone is prepared to stand behind it.

In an AI-heavy feed, the most valuable thing may be evidence that a human was actually there.

Not a personal story added to every post.

Not deliberate typos.

Not fake vulnerability.

Evidence of real thought.

A detail that could only come from doing the work.

A trade-off that does not resolve itself into a perfect lesson.

A sentence that sounds like the person who wrote it.

A view specific enough that someone else could disagree.

AI can make writing cleaner.

It can also sand off the things that make a person recognisable.

The restraint.

The humour.

The odd rhythm.

The hesitation before making a claim.

The context that stops a strong line from becoming an irresponsible one.

The job is not to avoid the tool.

It is to use it without letting the work lose the person behind it.

Subscribe now

Output is abundant. Judgment is still scarce.

AI has made it easier to produce something.

It has not made it easier to know what is true, what matters, what fits the context or what deserves to be published.

Organisations will not create more value simply by generating more material.

They will create it by making better choices about:

• what problem to solve

• who to serve

• what to say

• what to test

• what to stop

• what risk to accept

• what deserves to move forward

That is the work.

And as output becomes cheaper, it is becoming more valuable.