What is GEO?
Visible to AI assistants.
GEO, short for generative engine optimisation, is the work of making sure AI assistants such as ChatGPT, Claude, Perplexity, Gemini, Copilot and Google's AI Overviews can find, read and correctly describe your business. It is a marketing term, not a geography one. The symptom of neglecting it is quiet: a customer asks an assistant who to call instead of scrolling a list of links, rings whoever it names, and the enquiry you never received leaves nothing in your analytics to notice.
Every claim here about how Google, OpenAI, Anthropic, Perplexity and Microsoft behave is cited to their own documentation or to the original research, all read on 23 September 2026. Where they do not say how they choose, neither do we. Updated 24 September 2026. 9-minute read. How we write these guides.
The short answer.
GEO, or generative engine optimisation, means making a business visible to AI assistants such as ChatGPT, Claude, Perplexity, Gemini, Copilot and Google's AI Overviews. The term comes from a 2023 research paper[1]. In practice it rests on pages a crawler can read without running JavaScript, AI crawlers allowed by name in robots.txt, and the same business facts everywhere assistants look. No one controls which business an assistant names. GEO can remove the reasons yours gets ruled out.
Where did the term GEO come from?
From a research paper first posted in November 2023 by researchers at Princeton University and IIT Delhi, with two independent co-authors, and accepted to the KDD 2024 conference[1]. It named systems such as Bing Chat, Google's Search Generative Experience (SGE) and Perplexity 'generative engines', because they search for information and write one answer from several sources[1].
Their test engine took the top five Google results for a question and had gpt-3.5-turbo write the answer. They built a benchmark of 10,000 queries, some from existing datasets and some generated, and tested on its 1,000-query test split. They rewrote the source pages nine ways and scored visibility mainly by how much of each answer drew on a page, weighted by how high it was cited[1].
| Change made to the page | Result |
|---|---|
| Citing sources, adding quotations, adding statistics | Best of the nine: a 30 to 40 per cent relative gain on that measure[1] |
| Making the text more fluent or easier to understand | A gain of 15 to 30 per cent[1] |
| Rewriting in a more persuasive, authoritative tone | No significant improvement[1] |
| Keyword stuffing, widely used in SEO | Little to no improvement[1] |
Lower-ranked pages gained the most. In the paper's test, citing sources lifted the visibility of pages ranked fifth on Google by 115.1 per cent, while the top-ranked page lost 30.3 per cent on average. [1]
These are relative gains on the authors' own measures, mostly on an engine they built in 2023, though the methods also helped on Perplexity itself[1]. The authors note that methods 'may need to adapt over time'. The lasting lesson is the direction: pages with something specific to quote beat pages padded with keywords.
How do AI assistants find you?
Two ways. The first is memory: a model learns from text collected up to a fixed date, its training data cut-off. Anthropic, for one, publishes that date for its Claude models[2]. OpenAI's GPTBot and Anthropic's ClaudeBot gather web content that may be used in training[3][4]. If your phone number changed last year, a model may still hold the old one.
The second is live retrieval: an assistant with search looks things up as the question is asked and answers from the pages it reads.
ChatGPT
OpenAI says OAI-SearchBot surfaces websites in ChatGPT's search features, and sites that block it will not be shown in ChatGPT search answers, though they can still appear as navigational links[3].
Claude
Anthropic says Claude-SearchBot indexes content to improve search results for users, and Claude-User visits a site when someone asks Claude a question. Blocking either may reduce a site's visibility[4].
Perplexity
Perplexity says PerplexityBot surfaces and links websites in its search results, and is not used to crawl content for AI foundation models[5].
Gemini and AI Overviews
Google says that to be eligible as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to show with a snippet, with no additional technical requirements[6]. Content Google crawls can also ground answers in the Gemini apps[7].
Microsoft Copilot
Microsoft says Copilot turns a prompt into a short search query, sends it to the Bing search service and grounds its answer in what comes back[8].
What do some AI crawlers miss?
Some sites send a near-empty page and build the content with JavaScript in the browser. Google says it can process JavaScript content as long as it is not blocked[9]. AI crawlers are another matter. When Vercel and MERJ measured crawler traffic on Vercel's network in late 2024, the crawlers from OpenAI, Anthropic, Meta, ByteDance and Perplexity did not render JavaScript. ChatGPT's and Claude's crawlers downloaded JavaScript files without running them. Google's Gemini, which uses Googlebot's infrastructure, rendered JavaScript in full[10].
We saw the cost ourselves. Until pixelategroup.com was rebuilt in August 2026, the live site injected its navigation and footer with JavaScript, so a crawler that does not run JavaScript saw orphan pages.
| In the raw HTML | Before the August 2026 rebuild | Now |
|---|---|---|
| Links | 12 | The full navigation and footer |
| Navigation and footer elements | None at all | On every page, server-rendered |
View the source
Open your home page and choose view source, not the element inspector.
Search for words you can see
Find a sentence from your footer, a service from your menu and your phone number.
Read the gap
Words on screen but missing from the source come from JavaScript. Google will probably still read them. Assume the crawlers from those five companies cannot.
Which AI crawlers should I allow?
Your robots.txt file tells crawlers, by name, what they may fetch. AI crawlers come in three kinds: search crawlers fetch pages so an assistant can answer and cite you, training crawlers collect material for future models, and user-triggered agents fetch a page because someone asked. Each is a separate decision.
| Name | Company | What it is for | Kind |
|---|---|---|---|
| OAI-SearchBot | OpenAI | Surfacing websites in ChatGPT search[3] | Search |
| GPTBot | OpenAI | Content that may train its models[3] | Training |
| ChatGPT-User | OpenAI | Actions users ask for; robots.txt rules may not apply[3] | User-triggered |
| Claude-SearchBot | Anthropic | Improving search results for users[4] | Search |
| Claude-User | Anthropic | Fetching pages when someone asks Claude[4] | User-triggered |
| ClaudeBot | Anthropic | Content that could contribute to training[4] | Training |
| PerplexityBot | Perplexity | Surfacing and linking websites in its results[5] | Search |
| Perplexity-User | Perplexity | Pages users ask about; generally ignores robots.txt[5] | User-triggered |
| Google-Extended | A control token, not a crawler: Gemini training and grounding in the Gemini apps[7] | Training and grounding |
Blocking Google-Extended does not take a site out of Google Search. Google says the token does not affect inclusion in Search and is not used as a ranking signal. [7]
The switches are independent: OpenAI notes a site can allow OAI-SearchBot and still block GPTBot[3]. Our own robots.txt allows PerplexityBot, OAI-SearchBot, Claude-SearchBot and Claude-User. As a separate decision, explained in its comments, it also allows GPTBot, ClaudeBot and Google-Extended.
Why do consistent details matter?
A generative engine writes one answer from several sources, and some will not be yours[1]. When directories, review sites, your Google Business Profile and old listings disagree about your name, phone, area or services, the machine has to pick one or blend them.
Similar names blur. In the 28 days to 20 September 2026, Google Search Console showed pixelategroup.com in results 60 times for 'pixalate pricing'. Pixalate is a different company, one letter away. That is an entity problem, and the part any business controls is the same: one name, one address or service area and one description, repeated everywhere.
- Your website, social profiles, directories and review sites, all matching word for word.
- Your Google Business Profile, which Google says can help your services show in AI responses as well as ordinary results[9].
- Structured data, which Google says can help it tell your business apart in search results, with sameAs linking your other profiles[11].
Google also says structured data is not required for its generative AI features[9], so treat it as a way to be unambiguous, not a switch. Genuine mentions on other sites give an assistant more places to find you; Google warns that chasing inauthentic ones is not as helpful as it might seem[9].
What is llms.txt, and do I need one?
The llms.txt file is a proposal, published by Jeremy Howard in September 2024 and revised in August 2026: a plain Markdown file that summarises a site and points AI agents to its key pages, mainly for use when an assistant answers a question[12]. It is an emerging convention, not a ranking factor.
Google says Google Search does not use AI text files like llms.txt[9]. Chrome's Lighthouse, meanwhile, checks for one in its agentic browsing audits, and marks a missing file not applicable because providing one is 'optional at the moment'. [15]
Publish one only if you will keep it current, because a stale file hands assistants old facts. Ours is at /llms.txt.
How do I check what assistants say?
Ask the way a customer would
In each assistant, ask for the service and suburb, not your name. Use a fresh chat, note who is named and which pages are cited, and ask again another day.
Ask about yourself by name
Check the phone number, area, hours and services it gives. If a detail is wrong, search the web for it: the page still carrying it is the one to fix.
Open Search Console
Google's Generative AI performance report counts how often links to your site were shown in AI Overviews and AI Mode. Not every property has it yet[13].
Open Bing Webmaster Tools
Its AI Performance report, launched in preview in February 2026, shows how often your pages are cited in Copilot and Bing's AI summaries, and the queries behind them[14].
Run the free audit
The free audit runs thirty weighted checks live in your browser. We then scan your own site at our end, including crawler access and llms.txt.
Is GEO just SEO with a new name?
Partly, and Google says so: from its point of view, work aimed at its generative AI features is still SEO, because they rest on its core ranking and quality systems[9]. Microsoft uses the term too, calling its Bing report an early step towards GEO tooling[14]. The difference is the audience. GEO also has to account for crawlers that skip JavaScript, make a separate decision for each search and training bot, and get facts to agree across sites you do not control. SEO vs AEO vs GEO sets out how the three fit together, and the AEO guide covers how a page gets quoted.
No one controls which business an assistant names, and anyone who claims to is selling what they cannot deliver. GEO work goes after the reasons a machine cannot read you, confuses you with someone else, or has been told to stay away. Our SEO page sets out that work.
Questions and answers.
Is GEO the same as AEO?
They overlap. AEO is being the source a system quotes when it answers a question directly. GEO is being visible to the assistants at all: allowed in, readable, and described the same way everywhere. The AEO guide covers the quoting side.
Should I block AI crawlers from my website?
Decide by kind, not all at once. OpenAI says blocking OAI-SearchBot keeps a site out of ChatGPT search answers except as a navigational link, and Anthropic says blocking Claude-SearchBot may reduce visibility[3][4]. Blocking GPTBot or ClaudeBot is a separate choice about training. Google says Google-Extended does not affect Google Search[7].
Will an llms.txt file get my business into ChatGPT?
No company cited here says so. It is a 2024 proposal, not a standard, and the OpenAI, Anthropic and Perplexity crawler pages do not say their crawlers look for one[3][4][5]. Google says Google Search does not use such files[9]. Keep one current as a courtesy, not a lever.
How long does GEO take to work?
Some parts are quick: OpenAI and Perplexity say robots.txt changes reach their systems within about a day[3][5]. A fixed page can feed live answers once recrawled. A model's memory changes only when a newer model with a later training cut-off arrives[2]. No one can put a date on being named.
Does Pixelate build GEO into its websites?
Yes. Essentials, Growth and Pro websites all include the search and AI visibility layer: structured data, answer-first page structure, FAQ markup, llms.txt and AI crawler rules. For a site you already own, the SEO page sets out the work, and the free audit shows where you stand first.
Sources.
- Aggarwal et al., the GEO paper, arXiv 2311.09735, accepted to KDD 2024, arXiv. Read 23 September 2026.
- Claude models overview, training data cut-offs, Anthropic. Read 23 September 2026.
- Overview of OpenAI Crawlers, OpenAI. Read 23 September 2026.
- Does Anthropic crawl data from the web, and how can site owners block the crawler?, Anthropic. Read 23 September 2026.
- Perplexity Crawlers, Perplexity. Read 23 September 2026.
- AI features and your website, Google Search Central. Read 23 September 2026.
- Google's common crawlers, including Google-Extended, Google Search Central. Read 23 September 2026.
- Data, privacy, and security for web search in Microsoft Copilot and Microsoft Copilot Chat, Microsoft Learn. Read 23 September 2026.
- Optimizing your website for generative AI features on Google Search (updated 10 July 2026), Google Search Central. Read 23 September 2026.
- The rise of the AI crawler, Vercel and MERJ. Read 23 September 2026.
- Organization structured data, Google Search Central. Read 23 September 2026.
- The /llms.txt file, v2, llmstxt.org, Jeremy Howard. Read 23 September 2026.
- Generative AI performance report (Search), Google Search Console Help. Read 23 September 2026.
- Introducing AI Performance in Bing Webmaster Tools Public Preview, Microsoft Bing Webmaster Blog. Read 23 September 2026.
- Lighthouse agentic browsing audit: llms.txt, Chrome for Developers. Read 23 September 2026.
Find out what an assistant can read.
Thirty weighted checks run live in your browser, and we scan your own site at our end, including crawler access, llms.txt and structured data. There is no call attached, and the findings are yours either way.