Only 8% of the pages ChatGPT cites rank in Google's top ten for the same question, and for Perplexity the figure is 28.6%.
Those two numbers, from an Ahrefs study of 15,000 prompts, are the reason "ChatGPT SEO" cannot be one job. Each engine goes looking in a different place and comes back with a different kind of source.
Most advice on the subject treats the four big engines as one target. ChatGPT, Perplexity, Gemini and Google AI Mode retrieve differently, prefer different source types and show their citations in different ways. A UK SME with a small content budget has to choose where to spend it. The one thing that travels across all of them, the survey of 45 generative engine optimisation studies found, is topical relevance, not any per-engine trick.
This piece lays the four out side by side, with the sources behind each column, and ends with what to do for each one.
The four engines at a glance
Four things separate the engines. Where each gets its candidate pages, which domains it ends up citing, how visible the citation is to the reader, and what a small business can do about it.
| ChatGPT (with search) | Perplexity | Gemini | Google AI Mode | |
|---|---|---|---|---|
| Where it retrieves from | Third-party search providers plus partner content (OpenAI's words). Yoast and others say it leans on Bing's index | Its own continuously refreshed index, searched live for every answer | Google's index and the open web, grounded in Google Search | Google's index, using query fan-out to split one question into many searches |
| Overlap with Google top 10 | 8% in-text, 6.1% in references | 28.6%, the highest of the chatbots | Roughly 8 to 9% (Ahrefs chart, not labelled in the text) | Not in the Ahrefs study. Earlier Ahrefs data put AI Overviews at 76% |
| Most-cited domains | Wikipedia, Reddit, Forbes, TechRadar, LinkedIn | Reddit, YouTube, LinkedIn, Wikipedia, G2 | Reddit, YouTube, Wikipedia, Medium, PCMag and Forbes | YouTube, Reddit, Facebook, LinkedIn, Yelp |
| Share of answers that cite LinkedIn | 14.3% of responses | 5.3% | Not measured | 13.5% |
| How citations appear | A Sources button under the answer opens a sidebar; some inline links | Numbered footnotes on every answer | Selective. Source links shown for grounded answers, often fewer than Perplexity | Inline link icons at the end of paragraphs plus a side panel of sources |
| What a UK SME should do first | Get indexed and crawlable by Bing, allow OAI-SearchBot, publish on LinkedIn under a named person | Answer specific questions in full on one page, keep it fresh, show up in Reddit and review sites | Standard Google SEO, structured data that matches the page, entity consistency | Cover the sub-questions a buyer would ask, not just the head term; keep Business Profile current |
The domain lists come from Peec AI's US dataset of 30 million cited sources, published in March 2026. The LinkedIn figures come from Semrush's analysis of 325,000 prompts in January and February 2026. Neither is UK-specific, which I will come back to.
ChatGPT: mostly a Bing problem
ChatGPT with search retrieves from what OpenAI calls third-party search providers, plus content licensed directly from partners such as Associated Press and Reuters. OpenAI does not name the provider. Yoast and most of the SEO trade press say plainly that Bing's index is central to it, and that a site missing from Bing is unlikely to appear in ChatGPT's results.
That matters for UK firms because almost nobody checks Bing. Plenty of small business sites have never been submitted to Bing Webmaster Tools and carry crawl errors there that would have been fixed within a week on Google.
The Ahrefs data makes the point another way. ChatGPT's cited URLs overlapped with Google's top ten only 8% of the time, and with Bing's top ten 8.1%. So ChatGPT is pulling pages that rank well on neither. It is reading deeper into the index than a human searcher ever would.
What it then cites is heavily skewed to a few domains. An analysis of 680 million citations by Profound, covering August 2024 to June 2025, put Wikipedia at 7.8% of all ChatGPT citations, with Reddit at 1.8% and Forbes at 1.1%. The same study found 80.41% of ChatGPT-cited domains were .com, and only 2.16% were .uk.
That last figure is the one I would want every UK owner to see. A .co.uk domain is not a disadvantage in itself, but ChatGPT is mostly drawing on American and global sources. If your competitors for a query are US sites, the engine will probably cite them unless your page is clearly the better answer for a UK buyer.
ChatGPT's source mix also moves without warning. Semrush tracked 230,000 prompts over 13 weeks and found Reddit appeared in close to 60% of ChatGPT responses in early August 2025 and around 10% by mid-September. That swing in Reddit's share is worth watching, because the engines lean on it harder than on your own site. Wikipedia fell from about 55% to under 20% in the same window. Semrush's Sergei Rogulin put it down to "an attempt to avoid over-citing on certain websites". Anyone who built a ChatGPT plan around Reddit in July was dealing with a different engine by October.
Perplexity: the easiest to read and the hardest to fake
Perplexity searches live for every query and shows numbered footnotes linking to the original sources on each answer. You can see exactly which page supported which sentence. The LLM Pulse GEO guide calls its citation behaviour the most transparent of any major engine, which also makes it the easiest to measure.
It is also the engine closest to Google. Its 28.6% top-ten overlap in the Ahrefs study was more than three times ChatGPT's. If you already rank for a question, Perplexity is the engine most likely to notice.
What it favours is opinion and experience. Reddit was its top cited domain in both the 680 million citation dataset and the Peec one, at 6.6% of all Perplexity citations, with YouTube at 2% and Gartner at 1%. G2 appears in Perplexity's top five and in nobody else's. For a UK SME, review platforms and forum threads about your category are doing work on Perplexity that your own site may not be.
One more Perplexity quirk from the Semrush LinkedIn study: when it cites LinkedIn, 59% of those citations are company pages. ChatGPT and AI Mode go the other way, citing individual members 59% of the time. So a neglected company page costs you more on Perplexity than anywhere else.
Gemini: Google's index with a quieter citation habit
Gemini retrieves from Google's index and the open web, and shows source links when an answer is grounded in search. Its citation behaviour is more selective than Perplexity's, so fewer sources appear and the ones that do tend to be the same domains Google already trusts.
The Peec top five for Gemini was Reddit, YouTube, Wikipedia, Medium, then PCMag and Forbes together. That list is closer to ChatGPT's than to AI Mode's, even though Gemini and AI Mode share an index. BrightEdge data, reported by Frase, found Gemini shares only 27% of its sources with Google AI Mode.
There is very little Gemini-specific tactic worth paying for. Its top-ten overlap with Google in the Ahrefs study sits at roughly 8 to 9%, so ranking alone does not get you cited. The sensible approach is to treat it as a by-product of the Google work you do for AI Mode, and check it monthly.
Google AI Mode: one question becomes twenty searches
AI Mode uses what Google calls query fan-out, "breaking down your question into subtopics" and running multiple searches at once. A buyer asking "best accountant for a limited company in Northampton" triggers searches on fees, specialisms and reviews that the buyer never typed.
The page that gets cited is the one that answers one of those sub-queries well, which is frequently not the page ranking first for the head term. The LLM Pulse guide cites unnamed 2025 studies putting around 68% of AI Mode citations outside the organic top ten. I could not trace that figure to a named study, so treat it as a direction, not a number.
Google's own guidance from John Mueller in May 2025 says there is nothing special to do:
"Meeting the technical requirements covers you for search generally, including AI formats." Google Search Central
The same post asks for structured data that matches visible content, up-to-date Business Profile and Merchant Center information, and warns that nosnippet and max-snippet settings will limit how a page appears in AI results. For a local UK firm, the Business Profile line is the one most often ignored.
What AI Mode actually cites looks more social than the other three. Peec's top five was YouTube, Reddit, Facebook, LinkedIn and Yelp. Semrush found LinkedIn in 13.5% of AI Mode responses, the highest LinkedIn share of the three engines it tracked. Pew's study of AI Overviews, the sibling product, found government sites made up 6% of cited sources against 2% in standard results, which for UK firms means gov.uk pages will often beat yours on anything regulatory.
Pew also measured the cost of the format. Users clicked a traditional result on 8% of visits with an AI summary against 15% without, and clicked a link inside the summary on just 1% of visits. An AI Mode citation is worth having. It is not a traffic strategy.
The UK gap in all of this
Every dataset above is American or global. Peec's is explicitly US. The 2.16% .uk share in the 680 million citation study tells you how small a slice of ChatGPT's citations UK domains take. There is no published per-engine citation study for UK prompts that I could find, and any agency claiming to know "what ChatGPT cites in the UK" is extrapolating from the same US numbers.
That leaves two honest options. Run your own prompt set and record who gets cited for your category, in your town, in each engine, every month. Or pay for a tool that does the same. Either way you are building the UK dataset yourself.
What to do for each engine, in order
The sequence that makes sense for a UK SME on a small budget is this.
- Fix Bing first. Verify the site in Bing Webmaster Tools, submit the sitemap, check that OAI-SearchBot is not blocked in robots.txt. This is the cheapest ChatGPT SEO task there is and almost nobody has done it.
- Publish on LinkedIn under a named person, 500 to 2,000 words for articles and 50 to 299 words for posts, which were the most-cited lengths in Semrush's study. Keep the company page populated for Perplexity.
- Rewrite your three most important service pages so each one answers the sub-questions a buyer would ask (price, timescale, who it suits, who it does not). That is the AI Mode fan-out job and it also helps Gemini.
- Claim and complete the review profiles for your category. Perplexity cites G2 and Reddit; AI Mode cites Yelp and Facebook. UK equivalents vary by sector, so find where your competitors are reviewed and be there.
- Add structured data that matches what is on the page, and check Google Business Profile hours, categories and services are current.
- Only then start a monthly prompt log across all four engines, so you know which of the above is moving anything.
Everything in that list is ordinary search work with a different emphasis, which is why we fold it into the same AI visibility retainer instead of selling it as a separate product.
The decision rule I would give an owner is simple. If your buyers mostly ask Google, do steps 3, 5 and 6 and let Gemini and AI Overviews follow. If you sell to people who research on their phone in ChatGPT, do steps 1 and 2 before anything else. If you have no idea which, step 6 comes first, and you spend a month finding out before you spend anything on the rest.
If you would like help setting up that prompt log and working through the list for your firm, our generative engine optimisation service covers all four engines. We run it most for professional services firms whose clients now compare them across several engines before making contact, and you can reach us through the contact page.