Generative engine optimization (GEO): what it is and what the research says works

Generative engine optimization (GEO) is the practice of making your content more likely to be cited, quoted and recommended when AI systems such as ChatGPT, Perplexity, Gemini, Claude and Google AI Mode generate an answer. SEO earns a ranking and a click. GEO earns a place inside the answer itself, often without a click.

The term comes from a 2023 research paper, GEO: Generative Engine Optimization by Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan and Ameet Deshpande, accepted at KDD 2024. Most GEO advice online paraphrases a headline from it. This guide goes through what the paper actually measured, what independent data has shown since, and the process we teach at CXL for getting cited.

Key takeaways

• Evidence beats keywords. In the original GEO study, adding quotations lifted a source’s visibility in AI answers by about 41%, adding statistics by about 31% and citing sources by about 28%. Keyword stuffing made visibility 8% worse.
• GEO helps smaller sites most. A source ranked fifth gained 115% visibility from citing sources, while the top-ranked source lost 30%.
• SEO is still the base. Ahrefs found 37.1% of pages cited in Google AI Overviews also rank in the organic top 10 for the query, and Google’s own guidance says AI search runs on its normal ranking systems.
• Placement matters. In CXL’s analysis of 100 AI Overview citations, 55% of cited passages came from the first 30% of the page.
• Being cited pays, but less than a classic ranking. Seer Interactive’s data across 5.5 million queries shows cited brands get roughly twice the clicks of uncited brands on AI Overview searches, but still about 38% fewer than on searches with no AI Overview.

What is generative engine optimization?

GEO is optimization for engines that write answers instead of listing links. A generative engine retrieves several sources, then uses a large language model to summarize them into one response with citations. Your goal is to be one of the sources it uses, and to have your words, numbers and brand appear prominently in what it writes.

The original paper measured this with two metrics. Position-adjusted word count counts how many words in the answer come from your source, weighted toward the top of the answer. Subjective impression uses a model to rate how relevant, influential and prominent your citation is. Both are better proxies for GEO success than rankings, because the answer is the new results page.

GEO vs SEO vs AEO

The three overlap heavily. The difference is where you want to show up and how you measure it.

In practice you do all three on the same pages. Our guide to answer engine optimization covers AEO in depth, and our look at whether AEO and GEO are just SEO hype covers how the three relate.

What did the original GEO research find?

The researchers built GEO-bench, a set of 10,000 queries across domains, split into 8,000 for training and 1,000 each for validation and testing. They rewrote source pages using nine different methods and measured how much each source then appeared in the generated answers. Every method used the same content, only rewritten, so the differences come from presentation, not from new information.

Three findings matter most for marketers:

1. Evidence is the lever. The top three methods all add verifiable substance: quotes, numbers and citations. The paper reports the best methods improved visibility by 30% to 40% on the word count metric and 15% to 30% on the impression metric.
2. Keyword stuffing backfires. The classic SEO tactic performed worst, cutting visibility by about 8%. Generative engines reward content that helps them write a credible answer, not content that repeats the query.
3. Lower-ranked sources gain the most. For a source in fifth position, citing sources raised visibility by 115.1%, adding quotations by 99.7% and adding statistics by 97.9%. For the source in first position, citing sources reduced visibility by 30.3%. The authors suggest this gives smaller publishers a way to compete with sites that win on backlinks and domain strength.

The team also tested on Perplexity, a live engine, with a 200-query subset. Quotation addition improved the word count metric by about 21%, statistics addition improved the impression metric by about 37%, and keyword stuffing again made results about 9% worse.

What the GEO study does not prove

Treat these numbers as direction, not a forecast. The experiments ran in 2023 on a research setup and a 200-query live sample, and AI search has changed a lot since then. The paper also measured visibility inside the answer, not clicks, leads or revenue. And the authors found that the best method varies by domain, so test on your own queries before rewriting your whole site.

How do AI search engines choose which sources to cite?

Most AI search products retrieve pages from a search index, then generate an answer from them. Google’s AI Overviews and AI Mode use Google’s existing index and ranking systems, and run several related searches around your question before writing the answer, a technique Google calls query fan-out. Google’s own guidance, which we review in what Google’s AI search optimization guide gets right, is that AI search is still SEO.

Ranking for the exact query is not the whole story, though. In Ahrefs’ analysis of 4 million AI Overview citation URLs, 37.1% also ranked in the organic top 10, 26.2% ranked between positions 11 and 100, and 36.7% did not rank in the top 100 at all. Fan-out explains much of the gap: your page can be cited for a sub-question the user never typed.

Position on the page matters too. CXL mapped where the cited passage sat inside the source page for 100 Google AI Overview citations: 55% came from the first 30% of the page, 24% from the middle and 21% from the bottom 40%. Kevin Indig’s study of 18,012 verified ChatGPT citations found the same pattern, with 44.2% coming from the first 30% of the document.

Off-site reputation is the third input. In Ahrefs’ brand study, branded web mentions correlated at 0.664 with visibility in AI Overviews, one of the strongest signals they measured. That is a correlation, not proof of cause, but it matches what practitioners see: AI engines repeat what the wider web says about you.

Does GEO actually drive traffic?

Yes, but less than a classic ranking did. AI answers reduce clicks overall, and being cited claws back part of that loss rather than all of it.

Seer Interactive’s 2026 analysis covers 53 brands, 5.47 million queries and 2.43 billion organic impressions from January 2025 to February 2026. In February 2026, organic click-through rate was 2.36% on searches with an AI Overview and 3.82% on searches without one. Being cited made a large difference:

So on a query with an AI Overview, being cited roughly doubles your clicks compared with not being cited. Pew Research Center’s study of real browsing behavior points the same way: users clicked a traditional result on 8% of visits when an AI summary appeared, against 15% when it did not.

The business case for GEO is therefore not traffic alone. It is presence in the place where buyers now form their shortlist. If a buyer asks ChatGPT for the best tools in your category and you are not named, you do not get the chance to win the click at all.

How to do generative engine optimization: 7 steps

1. Build a fixed prompt set. Write down 30 to 50 questions your buyers ask on the way to a purchase: category questions (“best B2B analytics tools”), problem questions (“how to reduce churn in SaaS”) and comparison questions (“X vs Y”). Use sales call notes and support tickets, not only keyword tools. Many of these questions show zero search volume and are still asked in chat every day.
2. Record a baseline. Run every prompt in ChatGPT, Perplexity, Gemini and Google AI Mode. Log whether you are mentioned, whether you are cited, which competitors appear and which third-party sites get cited. That last list is your map of where AI engines get their evidence.
3. Put the answer first. On each target page, answer the main question in the first two or three sentences, then support it. Make every H2 a real question with a direct answer underneath. Given that most citations come from the top third of a page, this is usually the fastest win.
4. Add evidence, not adjectives. This is where the GEO research is clearest. Replace vague claims with specific numbers, attribute them to named sources, and add short quotes from credible experts or your own customers. Original data you own, such as a survey, benchmark or analysis of your product data, is the strongest version, because nobody else can cite it first.
5. Earn mentions where AI engines look. Use the third-party sites from your baseline: review sites, Reddit and community threads, industry publications, comparison posts and podcasts. Real participation and customer reviews build the kind of consensus AI answers repeat. Manufactured mentions on low-quality sites do not, and Google explicitly warns against them.
6. Keep the technical door open. Make sure AI search crawlers can reach your pages. OpenAI uses OAI-SearchBot for ChatGPT search and GPTBot for model training, and you can allow one while blocking the other; see OpenAI’s crawler documentation. Google-Extended controls whether your content is used for Gemini training and grounding, and it does not affect whether you appear in Google Search or AI Overviews. Content hidden behind scripts, logins or tabs is easy for crawlers to miss.
7. Re-run the prompt set every week. Change one group of pages and leave a similar group alone. After four to six weeks, compare citation rates between the two groups. Treat each change as a hypothesis, the same way you would treat an A/B test.

Which GEO tactics don’t work?

A lot of GEO advice is untested. Three popular tactics have little evidence behind them:

• llms.txt files. SE Ranking analyzed nearly 300,000 domains and found no meaningful relationship between having an llms.txt file and how often ChatGPT cites a site. In OtterlyAI’s 90-day log study, only 84 of more than 62,100 AI crawler visits requested the file, about 0.1%. Google says it does not use llms.txt for Search.
• AI-specific schema. Ahrefs tracked 1,885 pages that added JSON-LD markup against a matched control group and found no citation lift in AI Overviews, AI Mode or ChatGPT. Structured data still helps rich results. It is not a citation switch.
• Chunking content for AI. Google says its systems handle long pages and you do not need to split content into artificial chunks. Clear sections with one job each help readers and retrieval alike, which is different from breaking pages apart.

We cover the evidence on each of these in detail in our review of Google’s AI search guidance.

How do you measure GEO?

Measure GEO at four levels, from closest to the answer to closest to revenue:

A spreadsheet of 30 prompts re-run weekly is enough to start. If you want to automate it, see how to build an AI visibility tracker with n8n and how to track AI visibility with Bing.

When GEO is not worth your time

• Your pages don’t rank for anything yet. AI search engines retrieve from search indexes. If your site has no organic visibility, fix your SEO basics first. GEO will not rescue a site search engines barely know.
• Your buyers don’t research with AI. Some local, impulse and low-consideration purchases rarely start in a chatbot. Check your own buyers before you invest, for example by asking new customers how they found you.
• You have nothing original to say. GEO rewards evidence. If your content repeats what the top ten results already say, AI engines have no reason to cite you over the sources you copied.
• You expect a tool to do it for you. Many GEO tools sell dashboards for a problem a spreadsheet and an hour a week can measure. Buy one when you know which prompts matter and need to track them at scale.

Generative engine optimization FAQ

What is GEO in marketing?

GEO, or generative engine optimization, is the work of getting your brand mentioned and your content cited in answers from AI tools like ChatGPT, Perplexity, Gemini and Google AI Mode. It focuses on the evidence on your pages and your reputation on other sites, because that is what AI engines draw on when they write an answer.

Is GEO the same as SEO?

No, but they overlap. SEO aims for rankings and clicks in search results. GEO aims for citations and mentions inside AI-generated answers. Most AI search engines retrieve from search indexes, so good SEO is the foundation GEO builds on.

Does GEO replace SEO?

No. Google’s AI features run on its normal ranking systems, and 37.1% of AI Overview citations come from pages that already rank in the top 10. GEO adds a layer on top of SEO: more evidence on the page, answers placed earlier and stronger presence on third-party sites.

What is the difference between GEO and AEO?

AEO, answer engine optimization, focuses on being the direct answer to a specific question, including featured snippets and voice assistants. GEO focuses on generative AI systems and covers your brand’s presence across the whole answer, not only one snippet. The tactics overlap heavily.

How long does GEO take to work?

On-page changes can show up in AI answers within weeks once pages are recrawled, especially for engines that search the web live, such as Perplexity and ChatGPT search. Changes that depend on off-site reputation, like reviews and mentions, take months. Re-run the same prompts weekly so you can see which changes moved what.

Does llms.txt help GEO?

There is no good evidence that it does. Studies of hundreds of thousands of domains found no link between having an llms.txt file and being cited, and Google says it does not use the file for Search.

Learn GEO by doing it

CXL is a B2B marketing training company focused on AI and conversion optimization. If you want to apply this to your own site, CXL’s Optimize Pages for AI Search with GEO/AEO course, taught by Steve Toth, walks through auditing your AI visibility and building the comparison content buyers use in AI research. Compare it with other options in our list of the best AI SEO and GEO courses.

Sources

• Aggarwal, P., Murahari, V., Rajpurohit, T., Kalyan, A., Narasimhan, K. and Deshpande, A. GEO: Generative Engine Optimization. KDD 2024. arXiv:2311.09735.
• Seer Interactive. AIO Impact on Google CTR: 2026 Update. April 2026.
• Ahrefs. AI Overview citations and the organic top 10. 2026.
• Pew Research Center. Google users are less likely to click on links when an AI summary appears in the results. July 2025.
• CXL. Where Google AI Overviews take their citations from. 2026.
• CXL. What Google’s AI search optimization guide gets right and what’s missing. 2026.