AI Should Make You Think Harder, Not Think Less: a thoughtful woman beside the article title and Liz Baker branding

Let AI handle some of the effort. Keep your judgement firmly involved.

There is something rather satisfying about watching an empty document become a neatly structured plan in seconds. The headings appear, recommendations follow, and suddenly a task that felt substantial looks almost finished.

But before we start ticking boxes, we should ask ourselves what has actually been completed.

Have we explored the problem? Checked the evidence? Understood the consequences? Or have we acquired a convincing document?

In Beyond the AI Echo Chamber, I explored how different perspectives can strengthen decisions. The next step is considering how we work with those answers.

AI can help us produce something quickly. Making that output useful still requires thought.

Saving effort without surrendering judgement

Where the thinking goes
AI assists
Organise · Compare · Draft
You judge
Question · Verify · Decide

Some activities are sensible candidates for assistance: organising notes, comparing documents, suggesting questions or turning an approved outline into an initial draft.

The difficulty comes when we quietly hand over judgement along with the task.

Consider a marketing report. Asking AI to summarise campaign results may save considerable time. Accepting its explanation for a decline without examining the underlying figures is a different decision.

A summary describes what the information appears to show. An explanation introduces a claim about why something happened. That distinction deserves our attention.

Research offers a useful caution. A 2025 study involving 319 knowledge workers found that greater confidence in generative AI was associated with less reported critical thinking. It also identified a shift towards checking information, integrating responses and overseeing the work. These were self-reported findings, rather than proof that using AI causes people to lose thinking skills.

The practical question is whether we recognise that change in responsibility.

Less time producing a document does not automatically mean less work deciding whether its contents deserve to be used.

Start with your own view

Before opening a chat, take a moment to write down what you think.

Nothing elaborate. A few sentences will do:

  • What decision are we trying to make?
  • Which facts do we already have?
  • Where are we uncertain?
  • What would a useful outcome look like?

This gives us something to assess the response against.

For example, “Create a customer retention strategy” leaves considerable room for an assistant to fill the gaps. Without further context, its recommendations could suit almost any business.

A more useful starting point might be:

“Our first-time customers buy regularly during promotional periods but few return at full price. Before recommending a retention programme, identify the information needed to distinguish discount dependence from problems with product suitability or the purchasing experience.”

We have defined an observation without pretending to know its cause.

That connects directly with Journey Mapping That Guides Spend: understand the evidence before selecting activities.

Review the reasoning before polishing the wording

A familiar temptation is to start editing an AI response immediately.

We shorten the introduction, replace an awkward phrase and make the headings sound more like us. Those improvements matter, but a beautifully edited recommendation can still rest on an unsupported assumption.

Before adjusting the language, we should examine the argument.

  • Which statements are supported? The evidence behind the recommendation.
  • What has been assumed? Gaps presented as settled information.
  • Which alternatives were considered? Whether the proposal is unnecessarily narrow.
  • What would change the conclusion? How dependent it is on particular conditions.
  • Who needs to review this? The expertise or authority still required.

Emerging research supports this approach. A January 2026 preprint examining AI-assisted writing found that a review method focused on analysing assumptions reduced overreliance most effectively among the conditions tested. That is evidence from a particular experiment, rather than a guarantee for every workplace task, but it supports making assumption checks part of our routine.

A practical example: should we offer a discount?

Before choosing a discount
Repeat purchases have slowed
Purchase timingCheck normal buying intervals
Customer experienceReview feedback and delivery
Promotion dependenceCompare buying patterns
Check the evidence before choosing the response

Imagine an online retailer considering a discount because repeat purchases have slowed. This is an illustrative example, rather than a client case study.

An initial AI response might propose a percentage reduction, an email sequence and a deadline. Everything looks actionable.

Before approving it, we should investigate whether another promotion addresses the problem.

Perhaps customers bought products that naturally last several months. Delivery delays could have weakened their confidence. The original promotion may have attracted people unlikely to buy again without another offer.

Each explanation suggests different work.

The equivalent B2B question might concern a service renewal. A lower price will not necessarily resolve poor onboarding or uncertainty about the value received.

Ask the assistant to identify competing explanations and the records needed to investigate them. Then check actual purchase intervals, customer feedback and relevant service information.

The useful output may be a research checklist rather than a campaign.

That can feel less immediately productive. It could also prevent an expensive mistake.

Make room for useful friction

Speed matters, particularly when workloads are heavy. We do not need a committee meeting before changing a subject line.

However, the amount of review should reflect the consequences of being wrong.

A draft social caption, a budget recommendation and an automated customer communication deserve different levels of scrutiny.

A deliberate pause before action
01Explain itDescribe the recommendation
02Check itVerify the supporting evidence
03Challenge itTest the assumptions
04Own itDecide and review the result

For consequential work, introduce a deliberate pause:

Explain it. Can you describe the recommendation without reading the generated answer?

Check it. Have you examined the sources and confirmed their relevance?

Challenge it. Which realistic condition would make the proposal unsuitable?

Own it. Who will approve the decision and review its outcome?

If we cannot explain why a recommendation makes sense, we are probably not ready to implement it.

Use AI to develop understanding

An assistant can also help us practise.

Ask it to question your interpretation, work through a contrasting example or present a problem without revealing its proposed solution immediately.

For someone learning campaign analysis, that could mean examining a small set of results, explaining their interpretation and then comparing possible conclusions.

The objective is to become better at recognising what matters. Completing another exercise is less valuable if we cannot apply the learning independently.

Try this prompt:

A prompt to help you think further

I am considering [decision]. Before recommending an action, ask me three questions about the objective, available evidence and constraints. Then separate supported facts from assumptions, explain alternative interpretations and identify what requires independent verification. Finish by asking me to explain my decision in my own words.

What this means for marketing teams

For agencies: Include the reasoning behind recommendations in internal reviews. Clients should receive advice someone can explain and defend, not a presentation everyone has merely proofread.

For in-house departments: Bring operational knowledge into the discussion. Sales, service and finance colleagues may hold information missing from the original brief.

For independent marketers: Use AI to expose gaps in your thinking, then seek appropriate evidence or specialist input. A simulated expert perspective can suggest questions; it can't provide access to information you haven't provided.

Across every setting, measure more than output volume. Consider corrections, avoidable rework, implementation results and whether people understand the decisions being made.

A useful habit for your next task

Choose one piece of AI-assisted work this week. Before editing its wording, identify the central recommendation, check its most consequential assumption and explain why you would accept or reject it.

Record anything that changed your mind. That is useful learning for the next brief.

AI can take some effort out of our workday. The opportunity is to spend that recovered capacity on better questions, closer scrutiny and more considered decisions.

Intelligence behind every interaction