Reviewed, edited and approved by Editor-in-chief Daniel Sköld · Updated 21 October 2026
Contents
You ask an AI chatbot to tidy an email. It does. You ask it to explain a baffling letter. Excellent. Then it writes your reply, chooses your argument and tells you what to think about the news. At some point, the convenience starts to look suspiciously like surrender.
Is AI making us less clever? There is no good evidence that using a chatbot causes a general decline in intelligence. There is, however, a practical question worth asking: which bits of thinking are you no longer doing for yourself?
AD
Does using AI make you worse at thinking?
Generative AI — software that produces text, images or other material in response to your instructions — has made it remarkably easy to skip from question to finished answer. Ofcom’s 2026 research found that 54% of UK adults said they used AI tools. Three-quarters of online adults said they read AI-generated search summaries. You may be using one without ever opening a chatbot.
That matters because the summary can feel like the end of the job. Previously, you might have compared two search results, noticed they disagreed and worked out why. Now a neat paragraph appears above them. It might be useful. It might also hide the disagreement you needed to see.
A 2025 Microsoft Research survey asked 319 knowledge workers about tasks they had completed with generative AI. Greater confidence in the AI was associated with less reported critical thinking; confidence in their own abilities was associated with more. The researchers also found that effort shifted towards checking and integrating AI output. This was a survey of people’s accounts of their work, not a test showing that their intelligence had fallen.
That distinction is important. Handing a tedious task to a machine is not intellectual decay. Letting its plausible answer replace your own judgement, especially when you do not understand the subject well enough to spot a mistake, is a different bargain.
What should you let AI do for you?
Give it jobs where a rough first attempt has value and errors are easy to catch. Ask it to turn your scattered notes into an agenda, suggest questions for a tradesperson, or explain an unfamiliar term in plain English. Have it offer three approaches to a difficult email, then write the final version yourself. Your voice will probably survive the experience.
It can also be a patient sparring partner. If you are weighing up a home renovation, ask what assumptions your plan depends on, what costs you might have missed and what would change its recommendation. You are using it to widen the conversation, not appointing it household chancellor.
Keep the work that gives the task its meaning. If you want to understand a subject, first try to explain it in your own words. Then ask the chatbot to challenge your explanation. If you are deciding how to respond to a friend, think about what you actually mean before requesting a polished message. An impeccable paragraph is no substitute for knowing what you want to say.
Use AI to challenge your thinking, not to save you from having any.
A small experiment helps: before you open the chatbot, write down your answer or your first three questions. Compare its response with your starting point. What did it add? What did it miss? That pause costs very little and keeps you in charge of the decision.
AD
Can you trust an AI answer?
Not on its presentation alone. A large language model, or LLM, generates a response by finding patterns in the material it has been trained on and the information supplied to it. Fluency is one of its strengths; proof is not. The UK’s National Cyber Security Centre warns that these systems can produce incorrect facts while sounding convincing.
The stakes should decide how hard you check. For a dinner-party quiz, a dubious answer is mildly embarrassing. For a tax return, pension decision or health concern, it is a reason to stop. Ask AI to identify the questions you need to investigate, then check the answer with the relevant authority or a qualified professional.
For example, a chatbot can help you understand the difference between a State Pension forecast and what you might actually receive. It cannot see your record unless you provide it, and you should not take its guess as your entitlement. Use the Check your State Pension forecast service on GOV.UK for your own figures. The same principle applies to tax: check current information with HMRC rather than relying on a convincing explanation that may omit an important detail.
Checking means opening the underlying source, confirming it says what the AI claims and looking at its date. Ask whether the answer depends on your circumstances or where you live. Do not assume one account of a UK service applies identically across England, Scotland, Wales and Northern Ireland. If the chatbot cannot show where a consequential claim came from, treat it as a lead, not a finding.
Where is the line between help and dependency?
You can usually tell by what happens when the tool is unavailable. Could you still sketch the email, explain the conclusion or say why you made the choice? If not, the chatbot may be doing more than clearing away drudgery.
There is a privacy line, too. The NCSC advises against putting sensitive information into prompts to public AI services. Before pasting in a medical letter, a customer complaint or a work document, consider whose information it contains and whether you have permission to share it. Removing names is sensible, but other details can still identify someone. Check your employer’s policy rather than assuming a chatbot is a private notebook.
Research offers a more hopeful alternative to either blind trust or blanket refusal. In a 2026 controlled experiment on AI-assisted writing, Microsoft researchers found that asking people to examine the assumptions in an AI-generated plan reduced over-reliance without increasing reported cognitive load. It is a narrow finding about a particular task, not a universal cure. Still, the habit travels well: ask, ‘What are you assuming?’ and ‘What would prove this wrong?’
You do not need to reject a tool because it makes a job easier. Nor must you use it for everything because it is sitting in the search box. Keep your own first attempt for things you want to learn; let AI help with the laborious middle; inspect the result before you act. The point of saving time is to have more of it for decisions that are actually yours.
In short
AI does not automatically make you less clever. It becomes a problem when you let a polished answer replace checking, learning or deciding for yourself. Use it for drafts, explanations and challenges to your ideas; verify important claims and keep sensitive information out of public chatbots.
Sources
Ofcom – Adults’ Media Use and Attitudes 2026 report ([ofcom.org.uk](https://www.ofcom.org.uk/siteassets/resources/documents/research-and-data/media-literacy-research/adults/adults-media-use-and-attitudes-2026/adults-media-use-and-attitudes-2026-report.pdf?v=415430))
Microsoft Research – 2025 study of AI and critical thinking at work ([microsoft.com](https://www.microsoft.com/en-us/research/wp-content/uploads/2025/01/lee_2025_ai_critical_thinking_survey.pdf)); 2026 experiment on checking AI-generated plans ([microsoft.com](https://www.microsoft.com/en-us/research/publication/an-experimental-comparison-of-cognitive-forcing-functions-for-execution-plans-in-ai-assisted-writing-effects-on-trust-overreliance-and-perceived-critical-thinking/?lang=ko-kr))
National Cyber Security Centre – Guidance on chatbot accuracy and sensitive prompts ([ncsc.gov.uk](https://www.ncsc.gov.uk/blog-post/chatgpt-and-large-language-models-whats-the-risk?adhoc_referrer=568483602))
Alex Innes explains technology for men 45 and over who have seen every gadget since the ZX Spectrum and are not easily impressed. AI and ChatGPT, phones, cars, the smart home, the subscriptions worth keeping and staying safe online – what it does, whether it's worth the money, and whether you need it at all.
One of Midcent's writers. Spotted something? Write to alex@midcent.uk.
You do not need to memorise dozens of complicated passwords. Secure the accounts that unlock everything else, let a password manager handle the rest, and plan for a lost phone.
A robin lands while the kettle boils; your phone catches the action. Pick the right feeder, place it sensibly and dodge Wi-Fi grief and pointless subscriptions.
