Ask ChatGPT or Google's AI Mode which accounting software suits a small design agency, and you get a short answer that names three products and links to five pages. Your comparison page ranks on the first page of Google for that exact question, yet it is not among the five. In the next marketing meeting someone says you need "GEO", and someone else says it is just SEO with a new name.
This post explains what generative engine optimization means, how AI answers choose the pages they cite and how GEO differs from classic SEO. It also covers what has evidence behind it, which tricks count as spam, how to measure results and why answers change from one country to another.
What is generative engine optimization (GEO)?
Generative engine optimization is the practice of creating content that generative AI systems use, and credit, when they answer questions. A generative engine is a search product that writes an answer instead of only listing links, such as Google's AI Overviews and AI Mode, ChatGPT search, Copilot or Perplexity.
The term comes from a 2023 research paper by Pranjal Aggarwal and five co-authors, accepted at the KDD 2024 conference, which framed GEO as a way for content creators to raise their visibility in AI-generated answers. In SEO (search engine optimization), success means a high position in a list of links. In GEO, it means your page is one of the sources the answer draws on and links to.
You will meet other names for the same work. AEO (answer engine optimization) stresses answering questions directly, while "AI SEO" and "LLM SEO" are looser labels. Google's guide to its generative AI features mentions AEO and GEO by name and says that, from Google Search's point of view, optimizing for generative AI search is still SEO.
How do AI answers choose which pages to cite?
AI search products find pages through a search index, keep the ones their ranking systems trust, write an answer from them and link to some as sources. Google calls this grounding, or retrieval-augmented generation (RAG): the answer rests on pages retrieved at the moment of the question, not only on what the model learned in training.
- The question becomes several searches. Google calls this query fan-out: related searches on subtopics run at the same time, so a question about running shoes for flat feet may also search for arch support and recent shoe tests.
- The searches run against an index. Google's AI features use the Google Search index; ChatGPT search relies on OpenAI's OAI-SearchBot crawler and Perplexity on PerplexityBot. A page that is not indexed cannot be retrieved.
- Ranking systems pick the candidates. Google says its AI features rely on its core ranking and quality systems, and a page must be indexed and eligible to appear with a snippet before it can be shown as a supporting link.
- The model reads and writes. A large language model (LLM) reads passages from the chosen pages and writes one answer that combines them.
- Some sources become citations. Links appear beside sentences or below the answer; many pages read along the way are never shown.
No vendor publishes why one of ten good pages gets the link and another does not. Answers also vary from run to run, and Google says AI Overviews appear only when its systems judge that they add something beyond the normal results.
Training data or live search: which one can you influence?
You can influence live search far more than training data. Training data is text collected before a model was built; it shapes the answers given without a search and changes only with a new model version, so you cannot edit it.
Live retrieval reads your current pages at the moment of the question, so a change reaches these answers once the page is recrawled. OpenAI says ChatGPT search reflects a robots.txt change in about 24 hours, while Google says recrawling can take from several days to several months.
Each company uses separate names for its search crawler and for training, which matters when you write robots.txt:
| Company | Crawler for AI search | Token for training |
|---|---|---|
| Googlebot | Google-Extended | |
| OpenAI | OAI-SearchBot | GPTBot |
| Anthropic | Claude-SearchBot | ClaudeBot |
| Perplexity | PerplexityBot | None listed |
Google-Extended is a robots.txt token rather than a crawler: it controls use of your content for Gemini training and for grounding in the Gemini apps, and Google says it has no effect on Search. Blocking a search crawler takes you out of that company's AI answers, while blocking a training token only keeps your pages out of future training; our robots.txt guide shows how to write a rule group for each bot.
GEO vs SEO: what is different?
GEO changes the target and the scoreboard, not the foundations:
| Classic SEO | GEO | |
|---|---|---|
| Goal | A high position in the list of links | A source the answer uses and links |
| What wins | A whole page | A passage, fact or figure on the page |
| Success signal | Rankings, clicks, traffic | Citations, mentions, AI impressions |
| Where it shows | Search results pages | AI Overviews, AI Mode, ChatGPT, Copilot |
| Foundation | Crawling, indexing, useful content | The same |
The last row matters most. Google's AI features pick candidates with the same systems as its normal results, and ChatGPT and Perplexity send their own search crawlers, so a site that is hard to crawl or thin on substance has no GEO shortcut.
What changes is the reader. A person scans a results page and clicks one link; an AI system reads several pages, takes one sentence from yours and summarizes the rest. Facts stated plainly, with the evidence beside them, are easier to quote correctly.
What actually helps your pages get cited?
The practices with real support are the ones Google and Microsoft publish and the GEO study measured: be reachable, be clear, be original, show your evidence and stay current.
- Let the search crawlers in. Check that robots.txt, your CDN and any bot-protection rules allow Googlebot, Bingbot, OAI-SearchBot, PerplexityBot and Claude-SearchBot, and that important pages are indexed.
- Answer near the top. Put the direct answer in the first sentence of a section, under a heading that says what the section covers. Google adds that there is no need to cut content into tiny chunks or write in a special style for AI.
- Publish what others cannot copy. Google asks for original, first-hand content: your own tests, prices, photos and data rather than a rewrite of the top results.
- Show sources, figures and quotes. Microsoft's AI Performance announcement recommends backing claims with examples, data and cited sources, and the GEO study found these changes raised visibility most.
- Say who wrote it and how. Google's guidance on helpful content asks for accurate bylines and, where readers would wonder, a note on how the page was made, including any use of AI.
- Keep facts consistent and current. Keep prices, dates and specifications up to date, and keep the same facts in your Google Business Profile, Merchant Center feed and Bing Places listing. Microsoft also suggests IndexNow, a protocol that tells participating search engines when a page is added, updated or removed.
- Use structured data where it fits. Structured data is code, usually JSON-LD, that labels facts such as a product's price. Google says AI features do not require it and no special AI markup exists, but it still powers rich results and must match the visible text.
What did the GEO research paper find?
The paper that named GEO tested how rewriting a source page changes its share of an AI-generated answer. For each of 10,000 queries across 25 topic areas, the top five Google results went to GPT-3.5, which wrote an answer from them, and the team measured how much of the answer came from each source.
They rewrote the pages in nine ways. Adding citations to credible sources, quotations and statistics worked best, raising visibility by about 30 to 40 percent on their main measure; keyword stuffing did not help. Pages that started lower in the top five gained the most.
Read it as a lab result, not a ranking formula: a 2023 model answering from five fixed pages, and the authors expect the methods to change as engines change. The full paper on arXiv describes the setup. Its direction matches Google's and Microsoft's advice: evidence on the page helps, repeated keywords do not.
What does not help, or counts as spam?
Anything built to fool the system rather than inform the reader is either useless or against the rules. Google's spam policies list attempts to manipulate generative AI responses in Search as spam, and the older rules apply to AI features too.
- Hidden instructions for AI. Text people cannot see, such as white-on-white lines or text moved off screen, is hidden-text abuse. Instructions planted for an AI model to follow are prompt injection: an attack, not marketing.
- Keyword stuffing. It breaks a spam policy, and the GEO study found it did not raise visibility.
- Mass-produced pages. Generating many near-identical pages, with or without AI, mainly to get into answers is scaled content abuse.
- Fake reviews and mentions. Markup must match what visitors see, and Google says chasing inauthentic mentions across the web does little, because its spam systems also feed its AI features.
- Files that promise AI rankings. Google Search ignores llms.txt, so it does nothing for visibility in Google; our llms.txt guide covers where the file is useful.
- Tools that claim insider data. Google states that no third-party tool has access to its internal ranking or AI systems.
How do you measure GEO results?
You measure GEO with the search engines' own reports and your analytics, not with a single test question. Search Console's Generative AI performance report counts impressions of your pages in AI Overviews and AI Mode, Bing Webmaster Tools' AI Performance report counts citations in Copilot and Bing's AI summaries, and referral traffic shows visits from assistants such as ChatGPT and Perplexity. Which numbers to track, how to build a set of test questions and how to read these reports is the subject of What Is AI Visibility?.
Why do AI answers differ from country to country?
AI answers differ by country because they are built on search results, and search results depend on where the searcher is and which language they use. Google says it uses your current location to give more helpful results even when the query names no place, and Search Console's AI report splits impressions by country. A software company can be cited for a question asked in the United States and missing from the same question asked in Germany.
Checks from the office show only the office's country. For a few manual spot checks in another market, a browser connected through a Residential Proxy with an IP address in the target country and city receives the answers a local user would get.
Keep those checks manual and few. Google's spam policies count automated queries and rank-checking scrapers as machine-generated traffic, so do not automate Google Search or the AI assistants' apps; data at scale comes from Search Console, Bing Webmaster Tools and official APIs. Our SEO proxy page shows where location checks fit into that workflow.
Where does GEO matter most?
- Software and service companies whose buyers ask an assistant for options before visiting any website.
- Online shops, whose prices and stock reach AI answers through product pages and Merchant Center data.
- Local businesses, whose opening hours and addresses also come from business listings.
- Publishers and documentation sites, whose how-to answers are often summarized on the results page.
- Any brand with a reputation question: an assistant repeats what the most-cited pages say, accurate or not.
Common mistakes
- Treating GEO as separate from SEO. If pages are not crawled and indexed, no AI feature can cite them.
- Blocking the wrong bot. A rule meant to stop AI training that also blocks OAI-SearchBot or PerplexityBot removes you from those search answers.
- Forgetting the Search Console setting. Under Settings > Search generative AI, the Exclude option keeps your site out of AI Overviews and AI Mode; check which option your property uses.
- Rewriting every page into question-and-answer fragments. Google says its systems do not need chunked text, and readers rarely enjoy it.
- Judging from one prompt. A single answer on one day, from one account in one country, tells you almost nothing.
- Paying for guaranteed "AI rankings". No outside tool sees Google's internal signals, and nobody controls which source an answer cites.
Decision guide
| Your situation | What to do |
|---|---|
| Your pages rank but are never cited | Check snippet eligibility, then add first-hand facts and sources |
| You never appear in ChatGPT search | Allow OAI-SearchBot in robots.txt and in your CDN rules |
| You want out of AI training, not AI search | Block GPTBot, ClaudeBot and Google-Extended; allow search crawlers |
| You want out of Google's AI features | Choose Exclude under Settings > Search generative AI |
| You sell in several countries | Compare AI impressions by country; spot-check from each market |
| An agency guarantees AI citations | Ask for the method; nobody can guarantee them |
Frequently asked questions
Is GEO replacing SEO?
No. GEO builds on SEO: AI search features retrieve pages from search indexes and rank them with search systems, so crawling, indexing and useful content remain the base. Google's own guide says optimizing for generative AI search is still SEO. What GEO adds is attention to how your facts read when an AI quotes them.
What is the difference between GEO, AEO and AI SEO?
They name the same goal with different emphasis: GEO comes from research on AI-written answers, AEO stresses answering questions directly, and AI SEO is a broad marketing label. All three mean being found and cited by AI search.
How do I get my website cited in ChatGPT search?
Allow OpenAI's OAI-SearchBot in robots.txt and in any firewall or CDN rules, because OpenAI uses it to surface sites in ChatGPT search. Then make sure the page answers the question plainly, with first-hand facts and sources. OpenAI says robots.txt changes take about 24 hours to show; nobody outside OpenAI can promise a citation.
Does schema markup help AI answers?
Not directly in Google's case. Google says structured data is not required for its generative AI features and there is no special markup for them, although it still helps pages qualify for rich results. Keep it accurate and matching the page, and treat it as useful structure rather than an AI ranking lever.
How long does GEO take to work?
Changes reach AI answers only after the page is recrawled and reindexed. For AI search that ranges from about a day, OpenAI's figure for robots.txt changes, to several months, the upper end of Google's recrawl range; what a model learned in training changes only with a new model version. Judge results over months, not days.
Can I keep my site out of AI Overviews and AI Mode?
Yes. In Search Console, Settings > Search generative AI offers an Exclude option that keeps your links and content out of AI Overviews, AI Mode and the AI features in Discover, usually within one or two days. Google had rolled it out to all sites by 31 August 2026 and, as the Search Console help page explains, it does not affect ranking in the rest of Search. It does not stop AI training; that is the job of the Google-Extended token.
Summary
GEO is SEO aimed at AI-written answers: the goal is to be one of the sources that Google's AI features, ChatGPT search, Copilot and Perplexity read and cite. These systems search an index, rank candidates and quote clear, evidence-backed passages, so crawlable pages with original facts, visible sources and current details do the work, while hidden prompts and mass-produced pages count as spam. Measure with Search Console and Bing Webmaster Tools, and judge each market from that market.




