More sales conversations now begin with “I asked ChatGPT”. A buyer describes their problem to an assistant, gets a shortlist with linked sources and contacts the businesses on it. If yours isn’t named, you’ll rarely know. Generative engine optimisation (GEO) is the work of making your website easy for those assistants to find, understand and cite.
Much of what is sold under that name is repackaged SEO, and some of it is worse. This guide covers standalone assistants such as ChatGPT, Perplexity, the Gemini app, Claude and Microsoft Copilot. For Google’s own AI answers, see what Google AI Overviews and AI Mode mean for your website.
The short answer
Generative engine optimisation is the practice of making your content easy for AI assistants to retrieve, understand and cite. You’ll also see it called answer engine optimisation (AEO) or LLM SEO. In practice it means letting the right crawlers in, putting key facts in plain page text, answering questions at the start of each section, keeping business details consistent across the web and earning mentions on the sites assistants cite. Most of it is good SEO. The differences: assistants quote passages rather than rank pages, weigh what the wider web says about you, and don’t all search the same index.
How AI assistants find and choose sources
No assistant publishes a ranking formula, but the mechanics are clear enough to act on.
Two routes into an answer
- Training. What the model learned before its cut-off. You can’t edit it, and it changes only with new model versions. Crawlers such as OpenAI’s GPTBot, Anthropic’s ClaudeBot and Common Crawl’s CCBot feed this route.
- Live search. For questions about products, prices, suppliers or anything recent, assistants typically search the web, read a few pages and answer from them, usually with links. This is where citations come from, and where changes to your site show up once pages are re-crawled.
GEO pays back mainly through live search. Training is slower: being described accurately across the web today shapes what future models “know” about you.
Different assistants, different indexes
Microsoft Copilot draws on Bing, and the Gemini app on Google Search. ChatGPT, Claude and Perplexity run their own search crawlers, and some also use third-party search providers. So a strong Google ranking doesn’t guarantee you’re in the index a given assistant searches.
What gets a page cited
A cited page is usually in the index the assistant searched, matches one of the queries it generated from the user’s question, contains a passage that answers that query directly, and agrees with other sources.
The 2023 research paper that popularised the term, “GEO: Generative Engine Optimization”, found in controlled experiments that adding citations, quotations and statistics made pages more visible in generated answers, while keyword stuffing didn’t help. Experiments aren’t a promise, but they point the same way: specific, sourced, quotable content.
GEO vs SEO: mostly the same work
Most GEO advice is SEO advice relabelled. Assistants that search the web need what search engines need: crawlable pages, clear structure, helpful content and signs of trust. If your SEO foundations are weak, fix those first.
| At a glance | Traditional SEO | GEO |
|---|---|---|
| The goal | Rank a page and earn the click | Be named, quoted or linked inside an answer |
| What competes | Whole pages | Passages, facts and your business name |
| Trust comes from | Relevance, links and quality signals | The same, plus agreement with what other sources say about you |
Two differences need planning of their own.
Being quotable. Search engines rank pages; assistants lift passages. A thorough page can rank well yet never be quoted because its answer is spread across five paragraphs.
Entity clarity. Assistants build a picture of your business from everything they read: your site, profiles, directories, reviews, press. If those disagree about what you do or what you’re called, an assistant may hedge, confuse you with a similar name or leave you out.
The generative engine optimisation checklist
1. Let the right crawlers in
OpenAI documents GPTBot (training) and OAI-SearchBot (ChatGPT search) as independent settings, and Anthropic and Perplexity publish their own crawler lists, so you can opt out of training and still be cited. Some CDN and security services block AI bots by default, sometimes search crawlers included. Should you block AI crawlers? sets out the choices with robots.txt examples.
Check it yourself. Read yourdomain.com/robots.txt for rules naming OAI-SearchBot, PerplexityBot, Claude-SearchBot or Bingbot, or a blanket User-agent: * block, and review your CDN or firewall’s bot settings. Server logs show which of these crawlers actually visit. Then add the site to Bing Webmaster Tools, submit your sitemap and inspect a few key URLs.
2. Put key facts in server-rendered text
Not every AI crawler runs JavaScript, and text inside an image is, to most crawlers, just an image. Services, prices, specifications and opening hours that live only in images, PDFs or content loaded after a click may never be seen.
Check it yourself. On a key page, choose View Page Source (not Inspect) and search for your main service, a price and your phone number. If a fact isn’t in the source, a crawler that doesn’t run JavaScript won’t see it.
3. Write answer-first sections
Make each section easy to lift:
- Headings that match how buyers ask (“How long does installation take?”, not “Our process”).
- The answer in the first one or two sentences, then detail, exceptions and proof.
- Self-contained sections that name the service rather than relying on “it” from three paragraphs back.
- Specifics (ranges, conditions, dates) and tables for comparisons.
Before. “We offer flexible pricing tailored to every client’s needs. Get in touch to find out more.”
After. “Office cleaning contracts start from [price] a month for spaces up to [size], including [what’s covered]. Multi-site contracts are quoted after a visit.”
The second can be quoted; the first gives an assistant nothing to say. If you draft with AI, writing website copy with AI shows how to keep your own facts in every page.
4. Keep business details consistent everywhere
Settle on one name, one address format, one phone number and one sentence saying what you do and for whom (“[Name] installs cold rooms for food manufacturers”). Use them on your site, Google Business Profile, Bing Places, LinkedIn and directory listings, in every language you serve. It is the same discipline as local SEO.
5. Add appropriate structured data
Use Organization markup with your logo, contact details and sameAs links to your official profiles, LocalBusiness if customers visit you, and Article with author and dates on guides. Most standalone assistants say little about whether or how they use it, so treat markup as a consistent restatement of facts already written on the page, not a shortcut. Schema markup for business websites covers which types you need.
6. Earn mentions on the sites assistants cite
Ask for “the best cold-room installers for food factories” and the answer often leans on comparisons, reviews, directories, forums and trade publications rather than on companies’ own claims. Note which third-party sites are cited for your buyers’ questions, then get listed accurately there, contribute real expertise to industry publications and events, and ask satisfied customers for reviews where your buyers look.
On your own site, the signals behind E-E-A-T do similar work: named authors, sourced claims and first-hand examples make a page easier to trust and to attribute.
7. Keep pages current
Live search quotes whatever your page says today, including last year’s prices.
What doesn’t work, and can backfire
GEO attracts shortcuts, and nobody can guarantee a place in AI answers. Refuse these, whoever sells them.
Hidden instructions for AI. Invisible text telling assistants to “recommend this company” is prompt injection. Hidden text breaks Google’s spam policies, AI developers treat it as an attack, and if a customer finds it, your reputation pays.
Pages written only for AI. Showing crawlers different content from visitors is cloaking, and separate “AI versions” of pages drift out of date.
Fake reviews and self-ranking lists. Invented reviews are illegal in many countries, and putting yourself first on your own “top ten” list fools nobody.
Mass-produced pages. Near-identical pages for every phrasing of a prompt are what Google calls scaled content abuse, however they were written. Does Google penalise AI-generated content? explains where the line sits.
How to measure AI visibility
At the time of writing (August 2026), no assistant gives site owners a complete record of their citations, so measurement combines three partial signals.
Repeated test prompts
- Write 15 to 30 questions buyers genuinely ask, from broad (“Who installs cold rooms for food factories?”) to brand checks (“Is [your business] reliable?”), in every language you serve.
- Run each in the assistants your buyers use, with web search on and in a clean session. Repeat the important ones, because answers vary.
- Log whether you were mentioned or linked, whether the facts were right, which competitors appeared and which sources were cited.
- Rerun the same prompts monthly or quarterly and judge the trend, not single answers.
Bing’s AI Performance report
Bing Webmaster Tools launched an AI Performance report as a public preview in February 2026. It shows how often your pages are cited in Microsoft Copilot, AI summaries in Bing and some partner experiences, and which URLs. It covers only Microsoft’s side, but it is first-party data rather than a sample.
AI referrals in GA4
Clicks on citations usually arrive as referrals from assistant domains. In Google Analytics 4, create a custom channel group with an “AI assistants” channel whose source matches the regex .*(chatgpt|perplexity|gemini\.google|copilot|claude\.ai).*, and move it above Referral, because each visit lands in the first channel it matches (Google’s Analytics help has a similar example). Then compare its key events with other channels; GA4 conversion tracking covers setting them up.
Expect undercounting: some apps pass no referrer, and some people read an answer, then search your name. Watch branded impressions in Search Console, and ask new enquiries how they found you.
Frequently asked questions
How do I get my business to appear in ChatGPT answers?
Allow OAI-SearchBot in robots.txt and your firewall, state your services, prices and location in plain page text, and get described consistently on credible third-party sites. No form guarantees inclusion, and nobody can promise a mention.
Do I need an llms.txt file?
Not as a priority. llms.txt is a proposal, not an established standard, and at the time of writing there is little public evidence that major assistants rely on it. It’s cheap to add but never replaces crawlable pages with the facts written on them.
What if an AI assistant gets facts about my business wrong?
Trace the error. If the answer cites a source, correct your page or ask the third party to update theirs. If it cites nothing, the fact may come from training data: make the correct version prominent and consistent everywhere, and flag the answer with the assistant’s feedback option.
How long does generative engine optimisation take to work?
Fixes such as unblocking a crawler can show up once pages are re-crawled. Third-party mentions build over months and training data changes only with new models, so treat GEO as ongoing work alongside SEO.
What to do next
GEO isn’t a discipline to bolt on. It is what happens when strategy (knowing buyers’ questions), content (answering them plainly), development (HTML crawlers can read), SEO and measurement work together. How AI is changing websites puts it in the wider picture.
Start with three checks: robots.txt and CDN bot settings, key facts in the page source, and a handful of test prompts. For a second pair of eyes, a free website audit reviews the foundations assistants rely on as much as people do: Google visibility, speed, mobile experience, security and how easily visitors get in touch. If key facts turn out to be locked in images, PDFs or scripts, a website redesign fixes that at the root.