What Is Generative Engine Optimization? A Practical Guide

TL;DR: Generative engine optimization (GEO) is the practice of making your content easy for AI search tools such as Google AI Overviews, ChatGPT search and Perplexity to find, trust and cite in their answers. In practice it is mostly good SEO applied to a new format: crawlable pages, original content with clear answers and real evidence, and honest measurement. No tactic guarantees a citation.

Last reviewed: 29 October 2026

Generative engine optimization, usually shortened to GEO, went from an academic term to a line item in agency proposals in about three years. If you run a website, someone has probably told you that you need a GEO strategy, a special AI file, or a tool that tracks your “rank” inside ChatGPT. Some of that advice is useful. Much of it is guesswork sold with confidence.

This guide separates the two. It covers what GEO means, how AI search engines choose sources and what the original research actually measured. It then sets out what Google and Bing now say officially, and a practical workflow that needs no new tools.

The guide is written by Janak Pokharel, an SEO practitioner and co-founder of Limi Creatives, whose work is mostly organic SEO and AEO. Where the evidence is strong, the guide says so. Where it is thin, experimental or comes from a vendor, it says that too.

What is generative engine optimization?

Generative engine optimization (GEO) is the practice of improving a website’s content and technical setup so that AI search tools, such as Google AI Overviews, ChatGPT search and Perplexity, can retrieve, understand and cite it in their answers. GEO aims for inclusion in an answer rather than a position in a list.

The term was formalised in a research paper called GEO: Generative Engine Optimization. Its six authors, four from Princeton University and IIT Delhi plus two independent researchers, first posted it in November 2023 and presented it at the KDD 2024 conference. The paper defines a generative engine as a system that fetches web sources and uses a large language model to write one combined answer citing some of them.

A classic results page shows ranked links and lets the searcher choose. A generative answer does the reading for the searcher and shows a handful of supporting links, so your page competes to be one of the few sources it draws on.

You will also see this work called AEO (answer engine optimization), LLMO or AI SEO. The labels overlap heavily; the mechanics underneath matter more.

How do AI search engines choose which pages to cite?

Most AI search engines follow the same pipeline: crawl, index, retrieve, then write an answer with links. A crawler collects pages into an index, a retrieval system picks the pages most relevant to the question, and a language model writes an answer citing some of them. A page that is never crawled or retrieved cannot be cited.

Google describes its version in its guide to optimising for generative AI features on Google Search. AI Overviews and AI Mode use retrieval-augmented generation, also called grounding. Google’s core ranking systems retrieve relevant pages from the Search index, and the model writes a response that links to supporting pages. They also use query fan-out, running several related searches at once. Google’s example turns “how to fix a lawn that’s full of weeds” into follow-up searches about herbicides, chemical-free removal and prevention.

If your page does not rank for the question, or for the related questions the model fans out to, it is unlikely to be in the pool the answer is written from.

Each engine also has its own crawler and controls:

AI search engine	What must be true for a page to be shown	Main control for site owners
Google AI Overviews and AI Mode	Page is indexed and eligible to show with a snippet, and the site is included in Search generative AI features	Googlebot rules in robots.txt, snippet controls, and the Search generative AI setting in Search Console
ChatGPT search	Page can be crawled by OAI-SearchBot	robots.txt rules for OAI-SearchBot (GPTBot controls training, not search)
Perplexity	Page can be crawled by PerplexityBot	robots.txt rules for PerplexityBot

Google’s rules come from its AI features and your website page and the Search generative AI control, rolled out to all sites by 31 August 2026 and set to include sites by default. OpenAI’s crawler documentation separates OAI-SearchBot, used for ChatGPT search, from GPTBot, used for model training. Perplexity’s crawler documentation says PerplexityBot surfaces and links websites in its results. Our guide to robots.txt rules for search and AI crawlers covers what to block and allow.

A useful way to think about GEO is in two layers. The retrieval layer decides whether your page is in the candidate set at all. The synthesis layer decides how much of your page the answer uses and whether it links to you. For most small sites, the biggest wins are still in the retrieval layer, which is ordinary SEO.

What did the original GEO study actually find?

The 2023 GEO study by Aggarwal and colleagues found that adding quotations, statistics or citations to a source page increased how much of an AI-generated answer drew on that page, by up to about 40% in a simulated engine. Keyword stuffing slightly reduced visibility. The study did not test retrieval or rankings.

The researchers built GEO-bench, a set of 10,000 queries from nine sources across 25 subject areas. For each query they fetched the top five Google results and asked GPT-3.5 Turbo to write a cited answer from those pages. They then rewrote one of the five pages with nine different methods and measured how visible it became. Their main metric, Position-Adjusted Word Count, counts the answer words attributed to the page, weighted so that earlier citations count more.

The headline results from Table 1 of the paper, with percentage changes calculated from its scores:

Method applied to the page	Position-Adjusted Word Count	Change versus the unmodified page
No optimisation (baseline)	19.3	None
Quotation Addition	27.2	About +41%
Statistics Addition	25.2	About +31%
Fluency Optimization	24.7	About +28%
Cite Sources	24.6	About +27%
Keyword Stuffing	17.7	About -8%

On a 200-query test with Perplexity, Quotation Addition improved the same metric by 22%, and keyword stuffing performed 10% worse than the baseline. Gains were also largest for lower-placed pages: adding citations raised a fifth-placed page’s visibility by about 115%, but lowered it for the top-placed page by about 30%.

The limits, several of which the authors state themselves:

• Only two engines were tested, and the Perplexity test used uploaded source files, not live web search.
• Retrieval was fixed. The five sources were chosen in advance, and the authors did not evaluate effects on search rankings.
• The rewrites were produced with a language model. On a real site, statistics and quotations must be genuine and sourced.
• The metric is share of answer text, not clicks, traffic or sales.
• Engines have changed since 2023, and the authors note the methods may need to adapt.

The fair reading: once a page is retrieved, evidence-rich writing appears to help it contribute more to the answer, and keyword stuffing does not. It is not proof that adding quotes gets you cited by today’s ChatGPT or Google, and 40% is not a forecast for your site.

What Google and Bing now say about GEO

Google’s July 2026 guide says that, for Google Search, optimising for generative AI search is optimising for search, “and thus still SEO”. The guide lists things you can ignore for Google: AI text files such as llms.txt, breaking content into tiny chunks, rewriting content in a special style for AI, and chasing inauthentic mentions. It warns that creating pages mainly to target fan-out queries can breach the scaled content abuse rule in Google’s spam policies. It also says no third-party tool has access to its internal ranking or AI systems.

What Google says matters most is unique, non-commodity content: first-hand experience and expert perspective rather than a summary of what is already online.

Bing took a slightly different route. In February 2026, Bing added GEO to its Webmaster Guidelines, framing it as making content eligible for grounding and reference in AI answers, with no guarantee of citation. The same update widened its keyword stuffing rules to cover language engineered to trigger AI citations and added a section on prompt injection.

Side by side, the research and official guidance look like this:

GEO tactic	What the evidence says	Verdict
Original, first-hand content	Google says unique, non-commodity content will likely matter most in the long run	Prioritise
Real statistics, quotations and citations	Raised answer share in the 2023 study; matches Google’s focus on original information	Use, with genuine sources
Crawl access for search crawlers	Required by Google, OpenAI and Perplexity documentation	Check first
llms.txt and other AI text files	Google says Search ignores them	Optional, no Google benefit
Splitting content into tiny chunks	Google says it is not required	Skip
Many thin pages for fan-out queries	Can breach Google’s scaled content abuse policy	Avoid
Keyword stuffing or citation-bait wording	Lowered visibility in the study; covered by Google and Bing spam rules	Avoid
Hidden instructions aimed at AI systems	Bing now lists prompt injection as abuse	Avoid

For more on why repetition backfires, see our post on whether keyword density matters.

A practical GEO workflow you can start this month

No paid GEO platform is needed, and the cheapest checks come first.

1. Confirm that AI search can reach you. Check robots.txt for Googlebot, OAI-SearchBot and PerplexityBot, and make sure your CDN or bot protection is not blocking them. In Search Console, confirm the Search generative AI setting still includes your site. Use a technical SEO checklist to catch indexing problems, and keep an XML sitemap submitted to Google.
2. List the questions your buyers actually ask. Start with 10 to 20. Add the follow-up questions a fan-out search might ask, and answer them on the most relevant page rather than creating a thin page for each.
3. Answer first, then explain. Put a clear, self-contained answer near the top of each section. Google says no special style is needed for AI, so treat this as good writing for people that is also easy to quote.
4. Add evidence you genuinely have. Use dated statistics with named sources, verified quotes, original photos and first-hand observations. This is where the research and Google’s guidance agree, and only real numbers count.
5. Make your entities unambiguous. Use one consistent business name, show who wrote each article, and say clearly what you do and where. Google says any structured data must match the visible text.
6. Earn real mentions. Google says its AI features can reflect what blogs, videos and forums say about a business, but that chasing inauthentic mentions is less helpful than it seems. Earn genuine coverage instead.
7. Measure visibility, not “AI rank”. Use the Generative AI performance report in Search Console, rolled out to all sites by 31 August 2026, for impressions in AI Overviews and AI Mode. Check AI assistant referrals in your analytics. For manual checks, run a fixed set of prompts many times and record how often you appear.

That last point matters because AI answers are unstable. A January 2026 study by SparkToro and Gumshoe ran 12 prompts 2,961 times across ChatGPT, Claude and Google’s AI. It found less than a 1 in 100 chance of getting the same list of brands twice, and roughly a 1 in 1,000 chance of the same order. The authors call the findings preliminary, but the lesson holds: track how often you appear across many runs, not a single position.

Conclusion

Generative engine optimization is real, but it is not a secret separate discipline. AI search engines can only cite what they can crawl, index and retrieve, so the foundation is the technical and content work good SEO has always needed. The research adds one useful nudge: pages with genuine evidence, such as real statistics, verified quotes and named sources, tend to contribute more to answers once retrieved.

The most important step this week is to confirm that Google, ChatGPT and Perplexity can reach your key pages. To check quickly, run the free SEO audit from Nomor AI Marketing. It crawls a sample of your pages, checks technical SEO, on-page SEO, structured data and AI search readiness, and gives you a score and a PDF report, with no signup.

Frequently asked questions

Is generative engine optimization different from SEO?

Partly. Google says optimising for its AI features is still SEO, because AI Overviews and AI Mode rely on its core ranking systems. The differences are in output and measurement. AI answers cite a few sources rather than listing ten, and other engines use their own crawlers. Visibility is also tracked across repeated prompts, not as one ranking.

Can GEO guarantee that ChatGPT or Google will cite my website?

No. Google says meeting every requirement does not guarantee a page will be crawled, indexed or shown, and Bing’s guidelines say GEO does not guarantee citations. AI answers also vary between runs. GEO improves your eligibility and the quality of your evidence, not your certainty of being cited.

Do I need an llms.txt file or special schema for GEO?

Not for Google. Google says Search does not use llms.txt or other AI text files, and no special schema.org markup is needed for AI features. Structured data is still worth using for rich results if it matches the visible content. Other AI services may treat llms.txt differently, so check each service’s documentation.

Can I stop Google from using my content in AI Overviews?

Yes. Search Console has a Search generative AI setting, rolled out to all sites by 31 August 2026. It excludes your site’s links and content from AI Overviews, AI Mode and generative AI features in Discover. Snippet controls such as nosnippet or max-snippet also limit what is shown. Excluding your site means no impressions or traffic from those features.

How long does GEO take to show results?

There is no reliable timeline. Changes only count once engines recrawl your pages, and Google says recrawling can take from several days to several months. Gains from better content and reputation build gradually, so measure over months with the Search Console Generative AI performance report and repeated prompt checks.