GEO, AEO and “AI SEO” are three names for one job: making your business a source that AI-written answers pick up, quote and link to. That covers Google’s AI Overviews and AI Mode, ChatGPT search, Perplexity and Microsoft Copilot. Most of the work is ordinary SEO done properly, plus a handful of genuinely new checks around crawler access, how answers are put together and how little of the result you can measure.
Everything below is taken from what the platforms themselves document, with dates, and I say so wherever I am giving my own judgement instead. Where the guidance is thin, the honest answer is that nobody outside these companies knows.
The vocabulary, without the jargon
New labels arrive in this area faster than new techniques, so it helps to pin down what each one usually means before anyone sells you one.
| Term | What people usually mean by it | Where it applies |
|---|---|---|
| GEO (generative engine optimisation) | Getting a brand retrieved, mentioned and cited when an AI tool writes an answer | ChatGPT, Perplexity, Gemini, Copilot, Claude |
| AEO (answer engine optimisation) | Shaping pages so a system can lift a direct answer from them. It is older than GEO and started with featured snippets and voice assistants | Featured snippets, People Also Ask, voice answers, AI answers |
| AI SEO, LLMO, AI search optimisation | Umbrella labels covering both of the above | Depends on who is using the term |
| AI Overviews | The AI-written summary Google shows above some ordinary results, with links to supporting pages | Google Search |
| AI Mode | Google’s conversational search tab, built for longer questions and follow-ups | Google Search |
The glossary has fuller definitions of generative engine optimisation and answer engine optimisation. In practice the line between them is blurred, and I would not hire two different people for them. Both reward the same things: a page that can be found, understood and trusted, with answers that still make sense when quoted on their own.
Both of Google’s features are live here. Google announced AI Mode for UK users on 28 July 2025, and AI Overviews appear on some UK searches and not others.
How AI answers choose their sources
None of these tools publishes a ranking formula, but the documentation they do publish describes a consistent pattern, and it shows where effort pays off.
Training and live retrieval are two different routes
An AI model knows things in two ways. Part of its knowledge comes from training: text it read months or years ago, fixed inside the model and out of your reach. The rest comes from retrieval, where the tool runs a search at the moment someone asks, reads the pages that come back and writes its answer from them. That second route is called retrieval-augmented generation, or grounding.
Only retrieval produces citations you can work on. When an answer links to a page, that page was fetched for the question, so it had to be crawlable, indexed and relevant enough to be found by a search. This is why so much of GEO turns out to be search engine optimisation under another name.
Each tool searches a particular index
Google’s AI Overviews and AI Mode draw on Google’s own index. Google’s page on AI features and your website (Google Search Central, last updated 10 December 2025) says a page must be indexed and eligible to show in Search with a snippet before it can appear as a supporting link. ChatGPT search depends on OpenAI’s own crawler, OAI-SearchBot. Perplexity uses PerplexityBot, and Microsoft Copilot’s web answers draw on Bing.
One question becomes several searches
Google describes a technique it calls query fan-out: the system splits a question into sub-topics, runs several related searches and builds the answer from what comes back. A question such as whether a heat pump suits a Victorian terrace in Manchester might set off separate searches on running costs, available grants, insulating older houses and installers nearby. A page that answers one of those smaller questions clearly can be cited even if it would never rank for the full question.
Passages get quoted, not whole pages
An AI answer usually takes a few sentences from each source rather than the whole page. A paragraph that only makes sense after the three before it is hard to use. A heading that states the question, followed by a direct answer in the first sentence or two, gives the system something it can lift without misrepresenting you. This is the part of AEO that is genuinely about writing, and it makes pages easier for people to scan as well.
Answers that name businesses lean on other people’s pages
When someone asks for a recommendation, such as conveyancing solicitors in Leeds or a boiler engineer in Bristol, the answer is built from whatever pages the search returned: directories, review platforms, comparison articles, local news, trade body listings and the businesses’ own sites. My reading of the mechanism is that a business described consistently across several of those pages is easier to name with confidence than one that only describes itself on its own website. That is my judgement from how retrieval works, not a rule any platform has published.
Training crawlers and search crawlers are separate switches
This is the most useful technical point in the subject and the easiest to get wrong. Several AI companies run one crawler to gather training data and another to fetch pages for live answers. Blocking the training crawler keeps your content out of future model training. Blocking the search crawler takes you out of that tool’s answers altogether, and a broad “block AI bots” setting in a hosting panel or CDN can do both without the owner realising.
| Crawler or token | Run by | What it is for | If you block it |
|---|---|---|---|
| Googlebot | Crawling for Google Search, AI Overviews and AI Mode included | You leave Google, AI features and all | |
| Google-Extended | A robots.txt token, not a separate crawler. It governs whether your content is used to train Gemini models and to ground answers in the Gemini apps and Vertex AI | Google says your inclusion and ranking in Google Search are unaffected | |
| OAI-SearchBot | OpenAI | Finding pages to show and link in ChatGPT search | You are not shown in ChatGPT search answers |
| GPTBot | OpenAI | Collecting content that may be used to train OpenAI’s models | Your content is kept out of training; ChatGPT search is controlled separately |
| PerplexityBot | Perplexity | Finding and linking pages in Perplexity’s results; Perplexity says it is not used to train foundation models | You lose visibility in Perplexity’s results |
| Bingbot | Microsoft | Crawling for Bing, which Copilot’s web answers draw on | You drop out of Bing and the answers built on it |
The table follows Google’s list of its common crawlers (last updated 14 July 2026), OpenAI’s overview of its crawlers and Perplexity’s crawler documentation, the last two checked on 4 October 2026. OpenAI and Perplexity also run fetchers that visit a page when a person asks about it (ChatGPT-User and Perplexity-User), and both say those visits may not follow robots.txt rules.
Whether to allow training is a decision about your content, not an SEO question, and sensible businesses land on both sides of it. Whatever you choose, set it deliberately in your robots.txt file and check that your firewall settings agree, rather than finding out months later that a default switched you off. If the crawling and indexing problems go deeper than a few lines in that file, the fix is technical SEO work rather than anything specific to AI.
What is genuinely new, and what is SEO with a new name
Google’s guidance on its own AI features is blunt. It says no special optimisations are needed, that you do not need new machine-readable files or AI text files, and that there is no special schema markup to add. What it recommends instead would be familiar to any competent SEO: allow crawling, link to important pages internally, put key content in text, keep structured data consistent with what the page visibly says, and keep your Business Profile up to date.
So most of what is sold as GEO is existing work: pages that can be crawled and indexed, content that answers real questions better than the alternatives, visible experience and expertise, and a reputation that shows up on other people’s sites. If those foundations are missing, they come first, and my SEO guide for UK small businesses takes you through them in order.
A few things have genuinely changed, though.
- More crawlers to let in. Visibility now depends on OpenAI’s, Perplexity’s and Microsoft’s crawlers as well as Google’s, and each has its own settings.
- Bing matters again. Plenty of UK sites have never been set up in Bing’s tools because Bing sends them little traffic, yet a site Bing has not properly indexed cannot be cited in Copilot’s answers.
- Mentions without links carry more weight. In classic SEO, a mention of your business that does not link to you does little directly. When answers are assembled from retrieved pages, being named and described accurately on those pages is likely to matter whether or not they link.
- Answers can be wrong about you. AI tools sometimes state outdated or invented facts: an old price, a branch that has closed, a service you never offered. Checking what they say about your business is now part of looking after its reputation.
- Simple questions may not need a click. If the answer on the results page satisfies someone, they have no reason to visit a site. That is most likely for short factual questions, so I put more effort into pages that help someone choose, compare or buy than into bare definitions.
I would be cautious about anyone who promises a fixed position in ChatGPT, sells a special “AI schema” package or presents a new file on your server as the main fix. None of the platforms offers a way to secure a citation, and answers change from one person and one day to the next. Using AI tools to write your content is a separate question with its own risks, which I cover in using AI for marketing without hurting SEO.
What you can measure, and what you cannot
Measurement is the weakest part of this subject. Anyone reporting a precise “AI visibility score” should be able to explain exactly how it was produced and from which questions. Here is what a UK business can actually see today.
What you can measure
- Google Search Console. Google says traffic from its AI features is counted in the overall figures in the Performance report, under the Web search type. It is not shown separately, so you see the combined effect rather than the AI share.
- Bing Webmaster Tools. Microsoft announced an AI Performance report in public preview on 10 February 2026. It counts how often your pages are cited in Copilot and in Bing’s AI summaries and lists sample grounding queries, but it does not show clicks. It is the closest thing to citation data a major platform gives site owners, and a good reason to verify your site in Bing Webmaster Tools if you have not already.
- Referrals in GA4. Visits from links in AI answers show up as referrals from domains such as chatgpt.com and perplexity.ai. OpenAI’s publishers and developers FAQ says ChatGPT adds utm_source=chatgpt.com to the links it sends, which makes that traffic easy to separate.
- Server logs. Your hosting logs show whether OAI-SearchBot, PerplexityBot, Bingbot and Googlebot are reaching your pages, which settles any doubt about blocking.
- A fixed set of test questions. Write down 10 to 20 questions your customers genuinely ask, run them in each tool once a month and record whether you are mentioned, cited or described wrongly.
What you cannot measure
You cannot see how many people were shown your business inside ChatGPT, Perplexity or the Gemini app, because none of them currently gives ordinary site owners impression figures. You cannot split AI Overviews traffic from the rest of your Google traffic. You cannot see mentions that carry no link, and you cannot see the customer who read an AI answer, remembered your name and rang you a week later.
Answers also vary. The same question can bring up different sources for different people, at different times and with slightly different wording, so one screenshot proves very little either way. Treat your monthly test questions as a trend, and judge the work on what you can count reliably: enquiries, sales, searches for your business name and the referral traffic above.
Where a UK business should start
This is the order I work in on a typical small or medium-sized UK business site. Each step depends on the one before it, so there is little point polishing passages for AI answers while a search crawler is blocked.
- Confirm you are indexed in Google and Bing. Verify the site in Google Search Console and Bing Webmaster Tools, submit your XML sitemap to both, and check that your important pages are indexed in each.
- Check who you are blocking. Compare your robots.txt file and your firewall or CDN bot settings with the crawler table above. Decide on training crawlers deliberately, and leave search crawlers open unless you have a specific reason not to.
- Make your business details match everywhere. Use the same trading name, address, phone number and opening hours on your site, your Google Business Profile and the UK directories and review platforms in your sector. If your registered name at Companies House differs from the name you trade under, state both on your about or contact page so the two can be connected. Add LocalBusiness structured data (or schema.org’s general organisation type if customers never visit you) that matches what the page visibly says.
- Rewrite your key pages answer-first. Put the question your customer asks in a heading and answer it in the next sentence or two, then add the detail: prices in pounds where you can publish them, the areas you cover, and what is and is not included.
- Earn mentions on the pages answers draw from. Search your main questions yourself, note which directories, comparison articles, trade bodies and local publications keep appearing, and work out an honest route onto them. Reviews on Google and on the platforms your customers use belong in this step too.
- Check what AI tools say about you. Ask each tool about your business by name and by service. Where an answer is wrong, find the page it came from, correct your own site, and ask the owner of any outdated listing to update it.
- Set up measurement before you report on anything. Add the referral checks, Bing’s report and the monthly question log described above, so you have a baseline before you change anything else.
If you work in a regulated sector, add one more check. An AI answer can repeat a claim from your site without the qualifications that sit around it, so financial promotions under FCA rules, health and beauty claims under the CAP Code and the way a law firm describes its services all deserve a careful read with that in mind. Clear, accurate claims on your own pages are the best protection you have.
If you would rather have help, my AI search optimisation service covers this work from crawler access through to measurement. The simplest first step is a free SEO audit: send me your site and I will tell you which of the steps above I would fix first.
Updated