AI isn't the problem, humans are

The worry most small businesses bring to AI is about the stack: which tools, which subscriptions, whether everyone else has got further. The evidence points somewhere less dramatic. People are already using AI, mostly to save time, and the measured gains are smaller and less certain than the headlines. So the useful questions are human ones: which task you already do that it could help with, whether it really helps once you time it, and what you do with the minutes it gives back. A cup of tea is a legitimate answer. So is the dog.

Our interest, stated first. We sell a tool that scores you. Our free AI score gives a business a headline number out of 100, and we sell AI search visibility work on the back of it. A score is anxiety-inducing; nobody enjoys being graded, and a low number feels like a verdict. It should not, because of what it measures: whether AI assistants find, describe and recommend a business, "That and nothing else. Not your ads, email, reporting or social media", as the page says. It does not score the business, the quality of the work or the person running it. The check "reads your live site and asks an AI model what it knows about your business", so a low number describes what an assistant said on the day. That is a thing to fix, not a judgement on you.

We went looking for evidence

Evidence on AI and wellbeing at work is thin, and much of it is published by people selling AI or selling a fix for it. Here is what happened to the figures you are most likely to meet.

The claim that AI makes experienced developers 19% slower. A randomised trial by METR, a research non-profit, did find that in early 2025 developers "take 19% longer" with AI tools (METR, July 2025). The same page now says "We believe these historical results no longer reflect the current impact of AI models", and its February 2026 update calls its newer data "an unreliable signal" because developers increasingly refused to work without AI (METR, February 2026). Do not quote it as current.

The claim that 95% of AI pilots fail. From a report by MIT's NANDA project, which says "95% of organizations are getting zero return". Its own limitations note says its figures are "directionally accurate based on individual interviews rather than official company reporting", from 52 organisations and 153 survey responses at industry conferences (State of AI in Business 2025). A direction, not a statistic.

The claim that 77% say AI has added to their workload. The source, a survey for a freelance marketplace, says "Over three in four (77%) say AI tools have decreased their productivity and added to their workload in at least one way" (Upwork Research Institute, 2024). "In at least one way" matters, and the same report makes the case for hiring freelancers, which is what the company sells.

The claim that 75% of knowledge workers use AI. From 2024: a survey of 31,000 people across 31 countries, run by a research firm for Microsoft and LinkedIn, which sell AI tools (Microsoft Work Trend Index). Two years old, and paid for by a seller.

What survived

The UK's own official statistics, from fieldwork between November 2025 and March 2026: 59% of adults had used generative AI in the last three months; of those users answering online, 54% used it for work, and 65% "used it as a way to save time" (DSIT Public Engagement Survey 2025/2026). People are not waiting for the right stack. They are already working AI into what they do.

The government's own trials are the best UK evidence on time. In a cross-government trial of an AI assistant, users "saved an average of 26 minutes a day", but that was people estimating their own savings in ranges, with 7,115 survey responses from 20,000 licences (GDS, June 2025). The Department for Business and Trade's evaluation of the same tool found small time savings on most tasks, some tasks taking longer, and "did not find evidence that time savings have led to improved productivity" (DBT, August 2025). One civil servant in the Department for Work and Pensions trial put the wellbeing question honestly: "I'm not sure it's improved my well-being, but I do enjoy using, like, AI" (DWP, January 2026).

And the METR trial left one finding that has not dated: its developers expected AI to speed them up by 24%, and afterwards believed it had by 20%, when it had slowed them down (METR, July 2025). How AI feels and what it does can be different things.

What to do with that

• Start with one task you already do every week. Drafting replies, summarising a long document, a first go at a proposal. Not a new process built around a tool.
• Time it with a clock, not a feeling. Do it a few times each way and write the minutes down. The trials above show how easily people misjudge this.
• Keep the human check. The government evaluations report users stressing the need to review AI output rather than send it as it comes.
• Decide in advance what the saved minutes are for. If they vanish into more of the same work, the gain has gone to the to-do list. Tea, the dog, or leaving on time are all reasonable answers.

The same principle applies to being found: do small businesses need to worry about AI search? and is SEO dead now that everyone uses AI? come to a similar, calmer answer, and can you pay to be recommended by ChatGPT? deals with the people selling certainty.

Where the evidence is thin

Almost everywhere. We found no independent study of AI and wellbeing in small businesses specifically. The UK's best time evidence is self-reported, the one rigorous trial we found has been set aside by its own authors, and the most quoted figures come from companies with something to sell. That is the finding, and we would rather tell you so than give you a confident number.