What is GEO?
GEO (generative engine optimization) means making your content more likely to appear in the answers of AI search tools such as ChatGPT, Perplexity, and Claude. The term comes from the November 2023 paper "GEO: Generative Engine Optimization" (arXiv 2311.09735) by Aggarwal et al.: the authors are from the Indian Institute of Technology Delhi (IIT Delhi), Princeton University, and two independent researchers, and the paper was later published at KDD 2024. It proposes a set of methods and releases GEO-bench, a test set of 10,000 queries.
GEO is a new term, not an industry standard: different people use it for different things. Below we only cover what the paper tested and what kagentx evaluations can measure.
Where exactly do GEO and SEO differ?
| SEO | GEO | |
|---|---|---|
| What the user sees | A list of links, ranked | An answer written by an AI, with a few cited sources |
| What success looks like | A high rank and many clicks | The answer mentions you and cites your page |
| Main metrics | Impressions, clicks, average position | Asked N times, mentioned how many times; whether you appear in the cited pages |
| How to measure | The Search Console Performance report | Ask the same buyer questions several times on each AI, and record who is recommended and what is cited |
| Is the result stable? | The rank for one query is relatively stable | No: in SparkToro's 2026-01 study, asking the same question repeatedly produced the same recommendation list less than 1% of the time |
| Relationship to Google rank | The rank is the result | Limited: in Ahrefs' 2025-08 analysis, only 8% of the URLs ChatGPT cited ranked in Google's top 10 |
| Keyword stuffing | Used to be a way to raise relevance | Almost no improvement in the paper |
| What matters more in the content | Relevance, backlinks, page experience | Sentences, numbers, quotations, and sources that can be lifted straight into an answer |
It is not either/or. AI answers mostly find pages through search, so a page that cannot be read or is not indexed loses on both sides. The difference is in measurement: SEO looks at rank, while GEO has to be counted over repeated asks.
Which tactics did the paper actually test?
On an engine the authors built, the paper tested 9 ways of rewriting web content, and checked several of them on Perplexity.ai. The abstract's conclusion: GEO can boost visibility in answers by up to 40%, and the effect varies by domain.
- Adding cited sources (Cite Sources), quotations (Quotation Addition), and statistics (Statistics Addition) worked best.
- Keyword stuffing barely helped. On Perplexity.ai, pages with stuffed keywords did about 10% worse than unchanged ones.
- Lower-ranked sites gained more: in a simulation where every source was optimized at once, Cite Sources raised the visibility of the fifth-ranked site by 115.1%, while the top-ranked site lost 30.3% on average.
These are results measured on the paper's own test set in 2023–2024. They are not from Google's AI Overviews and cannot be applied directly to your site. Treat them as a direction: putting checkable numbers and sources on the page helps more than repeating keywords.
A real example: the search basics pass, and the AI still does not mention the site
On 2026-10-26 kagentx evaluated GammaSlide: all 4 technical checks, crawlability, AI readability, multilingual setup, and directive conflicts, came back green (4/4; crawlability means 30 requests from the homepage and 2 inner pages across 10 identities all returned 200, the same as Googlebot). That is the SEO foundation, and it had it.
The GEO side tells a different story: we put buyer questions to an AI (Claude Sonnet 5.5, web search on) 21 times and it mentioned the site 0 times; the most recommended were Gamma, Plus AI, Beautiful.ai (Gamma was named 12 times). The AI read 28 cited pages, 19 of which listed similar products, and 0 mentioned the site.
Same site: the SEO checklist passes, and the GEO count is 0/21. The difference is that who the AI recommends depends mostly on whether you are in the pages it cites. Full data in the GammaSlide case study.
Do both: where do I start?
1. Fix the SEO basics first
The page can be crawled, is indexed, serves its text in the HTML, and has a sitemap. Use the AI crawler check to see whether robots.txt blocks AI crawlers, and go through the AI visibility audit checklist item by item.
2. List buyer questions and measure GEO by counts
Without your brand name, write 10–20 questions a buyer would ask, ask each AI 3 times, and record how many times you are mentioned and which pages are cited.
3. Fill the gaps in the pages that get cited
If the list articles do not include you, pitch them; if comparison pages are missing, write your own; add numbers with sources and a sentence or two that can be quoted directly.
4. Run the same questions again
A few weeks after the changes, ask the same questions the same number of times and compare counts, not any single answer. For how to troubleshoot, see Why doesn't ChatGPT recommend my product?
FAQ
Who came up with GEO?
The November 2023 paper "GEO: Generative Engine Optimization" (arXiv 2311.09735). The authors are Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan, and Ameet Deshpande, from IIT Delhi, Princeton University, and two independent researchers. It was later published at KDD 2024.
Will GEO replace SEO?
There is no evidence that it will. AI answers mostly search first and then cite pages, and a page that cannot be read or is not indexed fails on both. A better view is that SEO is the foundation and GEO adds one more layer to measure on top of it: whether you are in the AI's answer.
Is there a GEO tactic that is guaranteed to work?
No. On the paper's test set, adding cited sources, quotations, and statistics raised visibility by up to 40%, and the effect varied by domain; that is an experimental result, not a promise for your site. Keyword stuffing barely helped in the paper.
How do I measure the effect of GEO?
Ask the same buyer questions several times on each AI, record how many times you are mentioned, who is recommended, and which pages are cited, then run it again a few weeks later and compare. Because the answer to one question differs every time, asking once means nothing.
Can a small site do anything with GEO?
In the paper's simulation, lower-ranked sites gained more, but that was an experimental setup. In practice, first check whether you appear in the pages the AI cites, then fill in list articles, comparison pages, and pages that can be quoted directly. That is the most realistic path for a small site.
Sources
- Aggarwal et al. (IIT Delhi, Princeton University) · GEO: Generative Engine Optimization （First submitted 2023-11-16, KDD 2024; abstract and text checked 2026-11-04）
- Ahrefs · Only 12% of AI Cited URLs Rank in Google’s Top 10 for the Original Prompt （published 2025-08-11, 15,000 prompts）
- SparkToro × Gumshoe · AIs are highly inconsistent when recommending brands or products （2026-01, 12 prompts run 60–100 times on each platform）
- kagentx evaluations: the data on the case study pages; each example gives its date.