Generative Engine Optimization (GEO): What It Is and What Works
Generative engine optimization (GEO) helps AI answers cite and name you. See how it differs from SEO and AEO, what research says works and how to measure it.
Generative engine optimization (GEO) is the practice of making your content and brand the source AI engines such as ChatGPT, Gemini, Perplexity, and Google AI Mode retrieve, cite and name when they answer a question. SEO earns a position in a list of links. GEO earns a place inside the answer, and it still depends on the crawling, indexing and authority SEO builds.
That changes what a win looks like. A buyer can get a definition, a shortlist or a recommendation from an AI answer without opening a single result, so the brands inside the answer are the ones that get considered.
GEO sits inside the AI search optimization baseline, next to answer engine optimization and the technical work that gets pages crawled and indexed. The evidence on what moves GEO is thinner than most advice suggests. The clearest lever it points to isn’t a page edit. It’s how much of the topic you cover.
Generative Engine Optimization Key Takeaways
- GEO is the practice of earning a place in AI answers: being retrieved, cited, mentioned, or recommended by engines such as ChatGPT, Gemini, Perplexity, and Google AI Mode.
- SEO competes for a position in a list of links. AEO targets answers inside search results. GEO targets the answers standalone generative engines write, and all three share the same foundation.
- In the original GEO paper, adding citations, quotations and statistics raised a source’s visibility by 30-40% in a benchmark. A 2026 survey of 45 studies found those gains hold only once a source is already in the engine’s context.
- In Floyi’s study of 354,955 AI citations, 40.2%-71.9% went to websites outside the top 20 for the query being answered. Websites ranking across more of a topical map were cited more often.
- Off-site pages count. When engines cite reviews, comparison articles and community threads, your standing on those pages decides whether your name reaches the answer.
- AI answers change between runs, so measure the same prompt set on a fixed cadence and read the trend, not a single check.
What Is Generative Engine Optimization (GEO)?
GEO is the work of getting a generative engine to use your content and name your brand when it writes an answer. An answer can give you four different results, and they aren’t interchangeable:
- Retrieved: the engine fetched your page while it built the answer. You can’t see this step directly, but nothing below it happens without it.
- Cited: the answer links your page as a source, as a footnote, a link chip or a source card.
- Mentioned: the answer text names your brand, with or without a link.
- Recommended: the answer names you as a pick or a shortlist option for the reader’s need.
The four rarely arrive together. A page can be cited while the answer never names its brand. A brand can be recommended from what the model learned in training, with none of its pages cited.
The term comes from a 2023 arXiv paper, “GEO: Generative Engine Optimization,” by a Princeton-led team with researchers from IIT Delhi. It was later published at KDD 2024. The paper treated generative engines as a new kind of search engine and measured which content edits raised a source’s visibility in their answers.
One clarification for searchers: GEO here has nothing to do with geo-targeting or local SEO. The “G” stands for generative, not geographic.
How Is GEO Different From SEO and AEO?
The three labels describe the same job on different surfaces. Floyi uses this split:
| SEO | AEO | GEO | |
|---|---|---|---|
| Surface | Organic results in Google and Bing | Answers inside search results: AI Overviews, featured snippets, People Also Ask, Bing Copilot | Standalone generative engines: ChatGPT, Gemini, AI Mode, Perplexity |
| Unit of success | A ranking position and a click | A direct answer lifted from your page | A citation, mention or recommendation inside the generated answer |
| What earns it | Relevance, crawlability, links, and topical authority | An extractable answer under a matching question | Topical coverage, quotable evidence and a presence on the pages engines cite |
| How you measure it | Rankings, clicks and Search Console data | Presence in snippets and AI Overviews | Citation rate, mention rate, share of voice, and stance across repeat runs |
In practice, the labels overlap more than the acronyms suggest. The work under all three is the same: cover the topic, back claims with evidence and make it obvious which entity you are. Argue for that work, not for the acronym.
AEO adds its own method at the level of the single answer. The audit and build steps for those answers live in the answer-asset workflow behind AEO. GEO asks a wider question across every engine and prompt your buyers use: whose sources make it into the answer, and whose brand gets named.
What Does Research Say Actually Works?

Three bodies of evidence are worth knowing. Read together, they point the same way.
The Original GEO Paper Measured Content Edits
The 2023 paper built GEO-bench, a benchmark of 10,000 queries, and tested nine ways of rewriting a source, from adding statistics to keyword stuffing. Its main findings, from the GEO paper on arXiv:
- Evidence edits worked best. Cite Sources, Quotation Addition and Statistics Addition achieved “a relative improvement of 30-40% on the Position-Adjusted Word Count metric.” That metric weighs how much of the answer draws on a source by where in the answer it appears.
- The headline gain was up to 40%. The authors report that GEO methods “can boost visibility by up to 40% in generative engine responses.”
- A live engine moved too. On Perplexity.ai, the paper reports “visibility improvements up to 37%.”
- Keyword stuffing didn’t. It offered “little to no improvement on generative engine’s responses.”
- Lower-ranked sources gained most. Cite Sources raised visibility by 115.1% for websites ranked fifth in the search results and cut it by 30.3% on average for the top-ranked website.
Sources that cited their own sources, quoted people and added numbers got used more. Keyword tricks didn’t.
A 2026 Survey Narrowed What Those Gains Mean
A July 2026 critical survey by Olivier Martinez reviewed 45 studies selected from a November 2023 to July 2026 publication window. Its abstract, from the 2026 GEO survey on arXiv, makes four points that matter for practice:
- The original paper’s gains are “valid within its experimental setting but conditional on a source already being present in a fixed context.” They don’t show a page getting found in the first place.
- Topical relevance and context position are “the most reproducible levers.”
- Generic heuristics transfer poorly, and citation-oriented rewrites can impair retrieval.
- Within that corpus, “no reviewed technique shows a stable, longitudinal, cross-platform causal effect on organic discoverability or downstream behavior.”
Page edits help once you’re in the room. They don’t get you into it.
Floyi’s Data Shows Citations Follow Topical Coverage
Floyi’s Topical Authority Report studied 42 topical maps, 535,239 Google rankings and 354,955 AI citations from AI Overviews, AI Mode, ChatGPT, and Gemini. It asked where those citations went:
- Many citations land outside the query’s top 20. 40.2%-71.9% of AI citations went to websites outside the top 20 for the query being answered, from 40.2% on AI Overviews to 71.9% on ChatGPT.
- Most of those websites rank elsewhere in the map. 26.8%-43.7% of all citations went to websites that ranked in the same topical map, just not for that query.
- Broader coverage went with more citations. Websites ranking across more of a map were cited in more AI answers on all four engines, with correlations from 0.33 on ChatGPT to 0.51 on AI Overviews.
- Coverage beat Domain Rating. Ranked coverage tracked AI citations more closely than Domain Rating did, 0.51 against 0.09 on AI Overviews.
Outside the top 20 doesn’t mean unranked. Those websites may rank below 20th for the query or not at all. The study measures correlation, not cause, and its engine-by-engine tables show how citations follow topical map coverage.
Put the three together and GEO looks less like a page trick and more like a coverage problem. The survey’s most reproducible lever is topical relevance. Floyi’s data shows most websites cited off-query rank elsewhere in the same map, and broader coverage goes with more of those citations.
GEO isn’t won on one page. It’s won across a topic.
What GEO Isn’t
Three tactics keep showing up in GEO advice that the evidence doesn’t support:
- A file you add. Some advice treats llms.txt as a requirement. Google’s guide, updated July 10, 2026, says you don’t need “new machine readable files, AI text files, markup, or Markdown to appear in Google Search (including its generative AI capabilities), as Google Search itself doesn’t use them.”
- Keyword stuffing. The GEO paper tested it directly and found little to no gain.
- Chopping pages into fragments. Google’s guide says, “There’s no requirement to break your content into tiny pieces for AI to better understand it.”
Each one takes time that the coverage and evidence work needs more.
The Coverage-to-Citation Loop: 5 Steps
The Coverage-to-Citation Loop turns the research into a routine you can repeat. It starts with prompts, not pages, and it ends on a recheck:
buyer prompts -> cited and named baseline -> topic coverage -> liftable evidence -> third-party presence -> recheck
Each pass gives you a new baseline. The next pass then measures against it.
Step 1: Pick the Prompts Your Buyers Ask AI
GEO starts from the questions buyers type into AI engines, not from your existing pages. Many of them carry Generative intent: the buyer wants the AI to produce a comparison, a summary or a plan rather than a list of links. It’s one of the five search intents Floyi’s Authority Planner labels topics with, alongside Informational, Navigational, Commercial, and Transactional.
Start with the prompts closest to a buying decision:
- Shortlist prompts: “best content planning software for agencies”
- Comparison prompts: “topical map vs keyword list for a small SaaS site”
- Problem prompts: “how do I find which topics my competitors own in AI answers”
- Definition prompts: “what is topical authority”
Tie every prompt to one topic in your map. A prompt with no topic behind it has no page that could answer it, and that’s a gap worth knowing about before you measure anything.
Keep the set fixed once you start. Changing the prompts between runs makes every comparison meaningless.
Step 2: Record Who Gets Cited and Named Today
Run each prompt on several engines and keep the full answer, not a yes or no. For each engine, record four things:
- The URLs it cited, yours and everyone else’s
- The brands it named in the answer text
- The stance it took on each brand, from recommended to criticized
- Whether your brand was named, cited, both, or neither
AIRS Analyzer does this for one query on demand across 11 search engines: 9 AI engines plus Google and Bing SERPs. It highlights each mention inside the answer, records the stance and sentiment for every brand named and counts AI citations apart from organic rankings. Each report is a snapshot of that run.
Keep the cited-source list. The pages engines already cite for your prompts become the outreach list in Step 5.
Step 3: Cover the Whole Topic, Not Just the Page
A single strong page answers one prompt. Coverage is what puts you in contention for the prompts around it. In Floyi’s data, ranking across more of a map went with more citations on queries a website didn’t rank for.
Plan coverage around each priority prompt with the four-level hierarchy:
- Pillar: the broad subject, such as content strategy.
- Hub: the section of it the prompt belongs to, such as content planning tools.
- Branch: the page that answers the prompt itself.
- Resource: the deep supporting pages that answer the follow-up questions.
A topical map in Floyi lays the subject out as Pillar > Hub > Branch > Resource, so you can see which parts of the topic have no page yet. Fill the branches around a priority prompt before you polish the one page that targets it.
Step 4: Put Evidence Where Engines Can Lift It
Once a page is in an engine’s context, the paper’s findings apply. Make the answer easy to lift and easy to trust:
- An answer-first passage under a heading that matches the question
- Original data, or a specific fact the cited pages don’t have
- Quotations and sources placed next to the claim they support
- A clear byline and a visible publish or update date
Edit with care. The 2026 survey found citation-oriented rewrites can impair retrieval, so a page that already earns citations isn’t the place to restructure everything at once.
Schema markup helps machines read the page, and it can make a page eligible for rich results. It guarantees no citation. Google’s guide says, “Structured data isn’t required for generative AI search, and there’s no special schema.org markup you need to add.”
The signals behind source selection go deeper than any single page edit, and how AI engines weigh citations and provenance covers them in full.
Step 5: Earn a Place on the Pages Engines Already Cite
When the cited sources for a prompt are review sites, comparison articles and community threads, your own page may not be the shortest path into the answer. Your standing on those pages is.
Work through the cited-source list from Step 2:
- Reviews: keep your listings current on the review sites engines cite for your category.
- Comparison articles: correct outdated details, or offer the author data they can quote with your name attached.
- Community threads: take part openly, as your brand, in the threads where buyers ask for recommendations.
- Industry coverage: pitch original research to the publications that already show up as sources.
Earn these, don’t manufacture them. Google’s guide advises against seeking inauthentic mentions across the web.
How Do You Measure GEO?

Measure GEO as several results per engine, for the fixed prompt set from Step 1. Use four core metrics:
- Citation rate: the share of answers that cite a URL on your domain
- Mention rate: the share of answers that name your brand in the text
- Share of voice: your mentions and citations against the competitors in the same answers
- Stance: whether answers recommend, compare or criticize you
Keep citation rate and mention rate in separate columns. Floyi’s Part 2 research shows how rarely AI answers name cited sites: in the typical map, answers mentioned only 4.6%-7.6% of the ranking websites they cited.
Answers also shift between runs. The 2026 survey points to “substantial run-to-run variability” in commercial audits, so re-run the same prompts on a fixed cadence and read the trend. A single check is a sample, not a result.
Two more data sources round out the picture. AI referral visits show up in your analytics. Query and click data for the pages behind your citations sit in Search Console. The full set of metrics and baselines is in benchmarking AI citations, traffic and conversions.
Floyi splits the read across three tools:
| Question | Where to read it |
|---|---|
| Who gets cited and named for this one prompt right now? | AIRS Analyzer, on demand across 11 search engines |
| Are my AI mentions and citations rising across every topic in my map? | Topical Authority Scorecard, AI Authority |
| What do AI answers already say about the topic I’m about to brief? | Briefs, with AI answer data next to SERP competitors |
The Topical Authority Scorecard’s AI Authority pillar trends mentions and citations per map topic across AI Overviews, AI Mode, Gemini, and ChatGPT over time. Briefs can pull in the AI Overview plus AI Mode, ChatGPT and Gemini answers alongside the SERP competitors, so the draft starts from what engines already say.
That trend is how you know the loop worked. See where AI answers cite you with Topical Authority, then rerun the loop on the topics where competitors own the answer.
Generative Engine Optimization FAQs
1. Is GEO Just SEO With a New Name?
Partly. Google’s guide says that from its perspective, optimizing for generative AI search “is optimizing for the search experience, and thus still SEO.”
The foundation is shared. The unit of success isn’t: SEO wins a position in a list of links, and GEO wins a citation or a mention inside the answer. Measure the two separately.
2. Does GEO Replace SEO?
No. Google says a page must be “indexed and eligible to be shown in Google Search with a snippet” to appear as a supporting link in AI Overviews or AI Mode. Other engines also answer from pages they can fetch.
A page that search engines can’t find gives no engine anything to cite.
3. How Long Does GEO Take to Show Results?
There’s no reliable fixed timeline. A change has to be recrawled and reindexed, and then each engine has to retrieve the page for the prompts you care about, which varies by engine.
Treat any promised timeline with suspicion and judge progress against your own baseline.
4. Do You Need to Rank on Google to Get Cited by AI?
Not for the exact query. In Floyi’s Topical Authority Report, 40.2% of AI Overview citations went to websites outside the top 20 for the query being answered.
Most of those websites still ranked elsewhere in the same topical map, though, so ranking across the topic still matters.
About the author

Yoyao Hsueh
Yoyao Hsueh is the founder and CEO of Floyi, the topical authority platform, and the founder of Main Heading, an SEO and AI search agency. He is an expert in topical authority and topical mapping. He created Topical Maps Unlocked, a course studied by thousands of SEOs, content strategists and digital marketers, founded TopicalMap.com, a done-for-you topical mapping service for agencies and enterprise teams, and publishes the weekly Digital Surfer newsletter on SEO, content strategy and AI search.
About Floyi
Floyi is a closed loop system for strategic content. It connects brand foundations, audience insights, topical research, maps, briefs, and publishing so every new article builds real topical authority.
See the Floyi workflow