GEO (Generative Engine Optimization) is the work of getting ChatGPT, Perplexity, Copilot or Google's AI Overviews to use and cite your website when they answer. It is not a separate discipline from SEO: Google says that, for its search engine, it is "still SEO" and that there are no additional requirements. What GEO adds is letting each AI's crawlers in, answering clearly and early, and measuring. Nobody can guarantee a citation.
What is GEO and where does the name come from?
The term comes from an academic study, GEO: Generative Engine Optimization, by Pranjal Aggarwal and five co-authors, published on arXiv in November 2023 and accepted at the KDD 2024 conference. It calls "generative engines" the search tools that, instead of a list of links, gather information from several sources and summarise it with a large language model.
Today the search engines use the word themselves. For Google, AEO (answer engine optimisation) and GEO are terms for work focused on improving visibility in AI search experiences. For Bing, GEO "focuses on content eligibility for grounding and reference in AI responses".
Do not confuse it with geography (geo-targeting, local SEO), although the two overlap: according to Google, AI answers can include information about local businesses, and keeping your Google Business Profile up to date can help your services appear in them.
What did the original study show, and what didn't it?
The authors built GEO-bench, a set of 10,000 queries across 25 domains, and simulated a generative engine: for each query, a model (GPT-3.5 Turbo) wrote the answer from Google's top five results. They then rewrote one of those sources in nine different ways and measured how much weight it carried in the answer. They also repeated tests on Perplexity, uploading the texts as files.
| Change to the source | Result in the study |
|---|---|
| Citing sources, adding quotations or adding statistics | The best: a 30% to 40% relative improvement on one of its two metrics and 15% to 30% on the other |
| Writing more fluently or in simpler language | Also improved, by 15% to 30% |
| A more persuasive, "authoritative" tone | No significant improvement |
| Adding more keywords | Little or no improvement |
The effect varied by topic and helped lower-ranked sources most: by citing sources, the site ranked fifth on Google gained 115.1% in visibility, while the one ranked first lost 30.3% on average.
Read it for what it is: an experiment with its own benchmark, its own metrics and GPT-3.5 Turbo as the model. It does not describe how ChatGPT or AI Overviews choose their sources today, and its "up to 40%" is not a forecast for your website. The lesson it leaves is sensible: content with sources, data and quotations, written clearly, does better; keyword stuffing does not.
How is GEO different from SEO?
At its core, hardly at all. Google puts it this way in its guide to generative AI features: "From Google Search's perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO." And in AI features and your website: "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary." If your Google is set to Spanish, AI Overviews appear as vistas creadas con IA and AI Mode as modo IA.
The reason is technical. According to Google, its AI features rest on its usual ranking systems: they retrieve pages from its index and answer from them, which is called RAG or grounding. On top of that, the model runs several related searches at once to complete the answer (Google calls this query fan-out). Bing says the same of Copilot: it relies on the same crawling, indexing and ranking foundation as its search engine.
The difference at a glance:
| SEO | GEO | |
|---|---|---|
| Where you want to appear | In the list of results | Inside the answer, as a link or cited source |
| Who reads you | Googlebot and Bingbot | Also each AI's search crawlers and the agents that open pages when someone asks |
| How your page is chosen | Query by query | From several related searches and several sources the system combines |
| What you measure | Clicks, impressions and position | Impressions in AI features, citations and visits from AI crawlers |
| Guarantees | None | None |
How does each AI search engine find your website?
Each company splits its bots by purpose: search (what gets you shown), opening a page because a user asked, and training models. They are independent. OpenAI gives the example on its crawlers page: you can allow OAI-SearchBot to appear in search results and disallow GPTBot so your content is not used to train its models.
| Where you want to appear | Token for search | Other tokens and what they control |
|---|---|---|
| Google AI Overviews and AI Mode | Googlebot | Google-Extended: training of Gemini and grounding in the Gemini app and Vertex AI; does not affect Google Search |
| ChatGPT search | OAI-SearchBot | GPTBot: training. ChatGPT-User: visits a user asks for; robots.txt may not apply and it does not decide whether you appear in search |
| Perplexity | PerplexityBot (not used for training) | Perplexity-User: visits a user asks for; generally ignores robots.txt |
| Claude | Claude-SearchBot | ClaudeBot: training. Claude-User: visits a user asks for |
| Copilot and Bing | Bingbot | The NOARCHIVE directive stops content being used in Copilot; NOCACHE limits it to the URL, title and snippet |
Sources for the table: Google, OpenAI, Perplexity, Anthropic and Bing.
Three more details:
- ChatGPT. If you block OAI-SearchBot you will not be shown in its search answers, though you can still appear as a navigational link, and a robots.txt change takes about 24 hours to take effect. According to its help centre, ChatGPT sometimes sends your question, rewritten, to other search providers, and among them it links to Microsoft's privacy statement. Being well indexed in Bing does no harm.
- Perplexity. Changes take up to 24 hours. If you use a web application firewall (WAF), its documentation explains how to allow its bots in Cloudflare and AWS.
- Claude. Anthropic says its bots respect robots.txt and that blocking their IP addresses may not work, because it stops them reading that file.
Google now also talks about AI agents that visit websites to carry out tasks for a person, such as booking or comparing products, and that read screenshots, the page structure and the accessibility tree. We explain what an agent is in AI agents for business.
What should go in your robots.txt?
If you want to appear everywhere and do not mind training, block nothing. If you want to appear in AI search without your content training models, here is an example:
# Search engines, AI ones included: welcome
User-agent: *
Allow: /
# Model training: no (optional)
User-agent: GPTBot
Disallow: /
User-agent: ClaudeBot
Disallow: /
Sitemap: https://www.example.com/sitemap.xml
If you add Google-Extended, bear in mind that, according to Google, you leave more than training: your content will not be used to ground answers in the Gemini app either.
Three checks:
- That you are not blocking by mistake OAI-SearchBot, PerplexityBot, Claude-SearchBot or Bingbot, for example with a blanket rule against "AI bots".
- Your CDN and firewall. Google asks that crawling be allowed by your CDN and hosting as well; OpenAI, that you accept its published IP addresses. An anti-bot rule can keep you out even when robots.txt says yes.
- What robots.txt does not do. It does not stop visits a user asks for (ChatGPT-User, Perplexity-User), and it does not remove a page from Google: under its technical requirements, the URL may still appear. That is what noindex is for.
What should you change on your website? A checklist for a small business
- Make sure Google can index the page. To be shown as a link in AI Overviews or AI Mode, a page must be indexed and eligible to appear with a snippet. The minimum: Googlebot not blocked, an HTTP 200 response and indexable content. Check it with the URL Inspection tool in Search Console. Do not put nosnippet on pages you want cited: it is the control for limiting what Google shows, and Bing warns it can weaken Copilot citations.
- Text in the HTML your server sends. Google can process JavaScript, but its JavaScript SEO guide says server-side rendering or pre-rendering is still a great idea, because the site is faster and not all bots can run JavaScript. Bing asks you not to hide critical content behind client-side rendering. Quick test: view the page source; if the text is not there, some crawlers may not see it.
- The answer near the top. Bing recommends putting essential information first, avoiding long introductions, making facts and definitions explicit and keeping each URL to one topic. Google asks for paragraphs and sections with clear headings, while making clear there is no need to chop content into pieces for AI.
- What only you know. Google insists on "non-commodity" content: a point of view of your own, first-hand experience, not recycling what is already online. For a small business that means your prices with a date, how you work, how long things take and the mistakes you see clients make.
- Who wrote it and where the data comes from. Google recommends making it clear who wrote each piece, with a byline that leads to information about the author, and warns that inventing authors is deception (helpful, reliable content). Bing favours pages that can be verified on their own. And in the original study, citing sources was among the most effective changes.
- Structured data that Google still uses. It is not required for AI features and there is no special markup for them, Google says, but it still helps with rich results. Its list of supported markup still includes Organization, Local business, Article and Breadcrumb, among others. FAQ markup (FAQPage) no longer counts: Google stopped showing that rich result on 7 May 2026 and removed its documentation in June. Whatever markup you use must match the visible text.
- An up-to-date business listing. Google names Business Profile and Merchant Center as ways to help your products and services appear in AI answers.
- Search Console, Bing Webmaster Tools and IndexNow. Register your site in both tools and submit a sitemap listing only canonical URLs with their last-modified date (lastmod). With IndexNow you notify search engines the moment you add, change or delete a URL; Bing says this reduces outdated references in Copilot. The notification is shared among participating search engines, which the IndexNow FAQ lists as Bing, Naver, Seznam.cz, Yandex and Yep. Google is not on that list: there, your sitemap and normal crawling still rule, and you can ask Google to recrawl a page from Search Console.
- Keep it current. Bing asks you to update content when facts change and remove what is out of date, so wrong information does not surface in its answers.
How do you know whether you appear in AI answers?
- Search Console. The Generative AI performance report shows your impressions in AI Overviews and AI Mode, by page, country, date and device. Google rolled it out to all websites worldwide as of 31 August 2026. If you cannot see it, the usual reason is that you do not yet have enough impressions. These appearances also count in the general Performance report, under the "Web" search type.
- Bing Webmaster Tools. Bing warns that your content can surface in Copilot as an impression, citation or grounding reference without anyone clicking, so fewer clicks do not always mean less visibility. Its Search performance report gives clicks and impressions per page and per keyword.
- Manual checks. Write down the 10 or 20 questions your clients ask you. Every month, ask them in ChatGPT with search, Perplexity, Copilot and Google, and open the sources of each answer. If your clients are a mix of Spanish and English speakers, ask in both languages. Note the date, the tool and whether you or a competitor appear. Answers change: Google warns that AI Overviews and AI Mode may use different models and show different links, and OpenAI that results and citations can be incomplete, outdated or incorrect.
- Your server logs. Visits from OAI-SearchBot, PerplexityBot or Claude-SearchBot mean they are crawling you for their search. Visits from ChatGPT-User, Perplexity-User or Claude-User mean someone asked a question and the assistant opened your page. Because a bot's name can be faked, OpenAI and Perplexity publish their IP ranges so you can check.
And be wary of tools that promise more. Google's guidance on third-party SEO tools and advice says no third-party tool has access to its internal ranking data and none can guarantee performance.
What does not work or is unproven?
- Guarantees. Google says indexing and serving are not guaranteed; OpenAI, that placement in ChatGPT is not guaranteed; Bing, that GEO does not guarantee citations. If someone promises to get you into ChatGPT, be wary.
- llms.txt as a ranking factor. llms.txt is a proposal by Jeremy Howard, published in September 2024 and revised in August 2026: a Markdown file that summarises a website for AI agents. It is not a standard. Google says its search engine ignores it, that it neither helps nor harms, and that you can keep one if other services use it. None of the OpenAI, Perplexity, Anthropic or Bing crawler pages we cite says they use it to find or cite pages.
- Buying links or mentions. Bing treats link buying, private networks and artificial promotion as violations. Google says seeking inauthentic mentions is not as helpful as it might seem, because its spam systems also apply to AI answers.
- A page for every variation. Creating content for every possible way of asking, primarily to manipulate AI answers, breaches Google's scaled content abuse policy.
- Keyword stuffing and "writing for AI". In the original study it barely helped. Bing penalises language engineered to trigger citations, and Google says there is no need to rewrite content just for AI.
- Hidden instructions for AI. Bing warns that content designed to manipulate its models can cost you visibility or removal from its index. It is a form of prompt injection.
What if you do not want AI to use your content?
There are controls, and each has a cost:
| You want to | How | What you lose |
|---|---|---|
| Leave AI Overviews, AI Mode and AI features in Discover | In Search Console, Settings > Search generative AI > exclude (Ajustes > IA generativa en la Búsqueda if your account is in Spanish; Google help) | Links, impressions and traffic from those features. It does not affect the rest of Search and takes a few days |
| Limit what Google shows from a page | nosnippet, data-nosnippet, max-snippet or noindex | Snippets in normal results; with noindex, the whole page |
| Stop models training on your content | Block GPTBot, ClaudeBot or Google-Extended in robots.txt | With Google-Extended, grounding in the Gemini app as well |
| Stop Copilot using your content | NOARCHIVE | Copilot answers and citations |
Two details from Google's help page: if you exclude your site, content from other sites similar to yours may still appear in those features, and the Search Console control does not affect training. That is what Google-Extended is for.
Is GEO worth it for a small business?
If your clients look for your services on Google or ask ChatGPT, yes. But most of the work is good SEO: being crawlable, clear content of your own, and measuring. The GEO-specific part is a handful of cheap tasks: checking robots.txt and your firewall, registering with Bing, switching on IndexNow and running your manual checks every month.
If you would rather go through it with someone, at Quantum Nova we handle technical and content SEO together with GEO: technical audits, AI crawler permissions and tracking how your site appears in Google, ChatGPT and Perplexity. We do not promise rankings or citations, because nobody can. See our SEO + GEO service.
Frequently asked questions
What does GEO mean in digital marketing?
It stands for Generative Engine Optimization: the work of getting AI tools that answer with search, such as ChatGPT, Perplexity, Copilot or Google's AI Overviews, to use and cite your website. The term comes from a 2023 academic study. It is not the same as geo-targeting or local SEO, although your business listing also counts in Google's AI answers.
How do I get ChatGPT to cite my website?
OpenAI asks for two things: allow OAI-SearchBot in robots.txt and check that your host or CDN accepts its published IP addresses. Beyond that, ChatGPT ranks results using several factors for relevance and reliability, and placement is not guaranteed. What helps is the same as on Google: original, clear content with sources.
If I block GPTBot, do I disappear from ChatGPT?
No. According to OpenAI, GPTBot is for training its models, and ChatGPT search depends on OAI-SearchBot. The two settings are independent: you can block one and allow the other.
Does llms.txt help you appear in AI answers?
On Google, no: it says its search engine ignores llms.txt and that the file neither helps nor harms. It is a 2024 proposal, not a standard, meant to help agents find information on a website. Having one does no harm, but do not expect it to improve your rankings.
How long does a change take to show?
It depends on the system. OpenAI and Perplexity mention about 24 hours to reflect a robots.txt change. Google says recrawling a page can take anywhere from several days to several months, and that its Search Console generative AI control takes a few days to apply.
Sources
- Google Search Central: Optimizing your website for generative AI features on Google Search
- Google Search Central: AI features and your website
- Google: List of Google's common crawlers (Googlebot and Google-Extended)
- Search Console Help: Search generative AI control
- Search Console Help: Generative AI performance report (Search)
- Google Search Central: Google Search technical requirements
- Google Search Central: Ask Google to recrawl your URLs
- Google Search Central: Understand JavaScript SEO basics
- Google Search Central: Creating helpful, reliable, people-first content
- Google Search Central: Structured data markup that Google Search supports
- Google Search Central: Latest documentation updates (FAQ rich result removed, 2026)
- Google Search Central: Guidance on using third-party SEO tools, services, and advice
- OpenAI: Overview of OpenAI Crawlers
- OpenAI Help Center: Searching the web with ChatGPT
- Perplexity: Perplexity Crawlers
- Anthropic (Claude Help Center): Does Anthropic crawl data from the web, and how can site owners block the crawler?
- Microsoft Bing: Bing Webmaster Guidelines
- Microsoft Bing: Refreshed Webmaster Tools (Search performance report)
- Microsoft Bing: How to add IndexNow to your website
- IndexNow: FAQ (participating search engines)
- Aggarwal et al.: GEO: Generative Engine Optimization (arXiv 2311.09735, KDD 2024)
- llmstxt.org: The /llms.txt file (proposal)