AEO-REX Special Report · Wednesday 5 August 2026 · Search
The regulator has a name for what is happening to search. Most British businesses have never heard it.
Ofcom calls it the era of answer engines. UK data says businesses are using AI to write, while their customers have quietly started using it to choose. The gap between those two facts is where the next five years of enquiries will be won or lost.
Ofcom published a discussion paper in November 2025 called The Era of Answer Engines. It set out how generative AI search works, how it differs from traditional search, and who the key players shaping the industry are. A national regulator, writing formally, about a shift that most business owners in this country still think of as "that ChatGPT thing."
I have spent the past year building a consultancy around that shift, and the gap between what the data says and what small businesses believe is the widest I have seen in twenty years of working with search.
If you sell to people who ask questions before they buy — and almost every business does — this is the most consequential change to how customers find you since Google itself arrived. So let me lay it out, and then tell you what I have built to deal with it.
- 35%
UK businesses using AI, June 2026 — up from 12% in late 2023.
(ONS)
- 1.6
Average AI tools per adopting business. Wide, not deep.
(ONS)
- 693%
Growth in AI referral traffic to retail, holiday 2025.
- 31%
Better conversion from AI referrals than non-AI traffic.
Section 01
What the numbers show
Start with adoption, because that is where most people go wrong. The Office for National Statistics reports that self-reported AI use among UK businesses with ten or more employees has risen from roughly 12 per cent in late 2023 to around 35 per cent by June 2026.
That sounds like a story about rapid uptake. It is not. The same analysis shows adoption has been shallow. The average number of AI technologies used per adopting business has crept from about 1.4 to 1.6 over the same period. Businesses picked up one tool, usually a chatbot for writing, and stopped.
Now the part almost nobody has priced in. While businesses were experimenting with AI as a writing tool, their customers started using it as a search tool.
Previsible's analysis of 6.77 million AI-driven sessions across 166 websites, running from November 2024 to May 2026, found growth of almost tenfold in nineteen months. And crucially, the researchers note that AI discovery happening inside Google's own results, including AI Overviews, almost certainly represents a larger volume of traffic than all standalone AI platforms combined. They excluded it because it does not produce trackable referrals in the same way.
The largest channel in this shift is the one nobody can measure properly.
The commercial signal is stronger still. Adobe Digital Insights found that AI referral traffic to US retail sites grew 693 per cent year on year during the 2025 holiday season (data 1 November to 31 December 2025), and that those AI referrals converted 31 per cent better than non-AI traffic.
Small volume, high intent. That is the shape of a channel in its early phase, and exactly when it is cheapest to win.
Section 02
The finding that should stop you
The number that changes the budget
- 11%
Only 11% of the URLs AI assistants cite appear in Google or Bing's top 10 for the same query. 80% don't rank anywhere in Google for that query at all.
- 76%
Of Google AI Overview citations come from pages ranking in Google's top 10. AI Overviews follow the search results. AI assistants do not.
That single finding dismantles the assumption underneath most UK marketing budgets, which is that if you invest in SEO, AI visibility follows. For AI assistants, it does not. They are different systems, making different decisions, using different signals.
But the picture depends entirely on which surface your customer is using. The same research found 76% of Google AI Overview citations come from pages ranking in Google's top 10. AI Overviews follow the search results. AI assistants do not. So a business can be well served by its SEO inside Google's own AI answers and completely absent from ChatGPT's — and it will have no way of knowing which.
Google Search Console will never tell you which of those you are. It does not measure it. There is no report, no dashboard, no alert.
A separate Semrush study measuring something different — whether the most-cited domain in a category was also the most-mentioned brand — found agreement in only 21% of cases. Citation and mention are not the same signal. For the full analysis, see AEO REX Weekly Issue 02: being mentioned is not being recommended.
Section 03
Why "optimise for ChatGPT" is already outdated advice
The other assumption worth killing is that this is a one-platform problem.
Similarweb data shows ChatGPT's share of worldwide generative AI web traffic sliding from roughly 76 per cent a year ago to around 53 per cent today, while Gemini has climbed past a quarter of all traffic and Claude has become the fastest-growing platform in the category. Optimising for one AI platform used to cover most of the addressable audience. It no longer does.
Different studies produce different splits depending on what they measure, and that disagreement is itself instructive. This is a market still moving fast enough that the measurement has not settled. But the direction is not ambiguous.
Section 04
What I have been building
Everything above describes a measurement problem wrapped around an execution problem. You cannot see whether AI recommends you. If you could see it, you would not know why. If you knew why, you would still need someone to fix it. And once fixed, you would have no way of proving it worked.
Every tool I have evaluated solves the first of those four and stops. It tells you that you are invisible, then hands you a dashboard and an invoice.
So I have been building the AEO-REX Command Centre, an operating system that runs the whole loop rather than the first quarter of it.
- It asks. The questions your customers actually type, run repeatedly across every major engine, with raw responses stored as evidence. A single run is an anecdote, not a measurement.
- It diagnoses. Whether AI crawlers can reach your site at all — and a surprising number of UK businesses have quietly blocked them without knowing.
- It fixes. The system drafts the changes, but nothing reaches your website without your approval. In regulated sectors, an AI publishing unreviewed claims about your practice is not efficiency. It is a liability.
- It proves. It re-runs the same prompts afterwards and shows you the movement, before and after, against the questions you agreed to track.
I am being straight about where this sits. The platform is in build, with the measurement and diagnostic layers running. I would rather tell you that than sell you a screenshot.
Section 05
Why I care about this fight
I teach Applied Science and Physics at BMet College alongside running AEO-REX. That background is why I am allergic to the way this field talks: too much confident assertion, too little method. We are early, and what that demands is not certainty. It is rigour — repeated measurement, stored evidence, and a clear line between what was verified and what was estimated.
The businesses that win the next five years will not be the ones ranking highest. They will be the ones getting named.
AEO-REX was the first Answer Engine Optimisation consultancy in the UK built specifically for small and medium businesses. Not an SEO agency that added a service page in 2025. This has been the whole job from day one.
Section 06
Sources
- Ofcom, The Era of Answer Engines, discussion paper, November 2025.
- ONS, Artificial intelligence in UK businesses: 2023 to 2026.
- Previsible, 2026 AI Traffic Report, 6.77m sessions, 166 properties.
- Adobe Digital Insights, January 2026 (data 1 November to 31 December 2025).
- Similarweb, AI Search Stats 2026, July 2026.
- Ahrefs (Linehan & Guan), AI search citation overlap, 15,000 prompts, 11 August 2025. Added in the web edition with the corrected finding above.
- Semrush, 50,000 brands across 1,094 US categories, ChatGPT, January to June 2026. Added in the web edition.
Figures are dated snapshots of a fast-moving field. Last reviewed .