What Is Generative Engine Optimization (GEO)?

What is generative engine optimization?

Generative engine optimization (GEO) is the practice of shaping your content, and what the web says about you, so AI search tools cite or name you in the answers they write. A generative engine is a search tool that writes an answer from several sources instead of returning a list. ChatGPT search, Perplexity, Gemini and Google AI Mode are examples.

The term was introduced in a 2023 paper by Pranjal Aggarwal and colleagues, which was accepted to the KDD 2024 conference ¹. Wikipedia lists GEO, AEO and AIO as names for the same practice and notes there is no agreed definition ⁵. Google calls GEO and AEO third-party labels and says optimizing for generative AI search is still SEO ².

In this library, GEO is the credibility side of AI SEO: will an AI trust your content enough to cite it? Its sibling, AEO, asks whether an AI can read and lift your answer.

What did the original GEO research test?

It compared nine ways of rewriting a web page on a bench of 10,000 queries from 25 areas such as arts, health and games. Each query came with the text of the top 5 Google results, and GPT-3.5 Turbo wrote the answer from them, in a simple two-step setup ³. The authors called the bench GEO-bench.

Visibility was scored by how many words of the answer came from your source and how early they appeared, plus a judged impression score ³. So the result measures your share inside one answer, not traffic, clicks or sales.

The authors also ran 200 test queries through Perplexity.ai, but they supplied the source texts as uploaded files ³. The paper reports gains of up to 37% on Perplexity.ai on one metric, but with the source texts supplied as uploaded files, so it still does not show how live ranking behaves.

What worked in the study, and what did not?

Adding evidence worked best, and keyword stuffing did not work. The table shows each method's score on the paper's main metric against a baseline of 19.3. The percentage changes are our own arithmetic from the paper's Table 1 ³.

The authors' text says a more persuasive tone gave no significant gain, although the table shows about +10% for it on this metric. The text also rounds the top three results to '30 to 40%', while the table gives 27% to 41% ³. Treat the exact ranking as soft and the broad pattern as the finding.

What can the research not tell you?

It cannot tell you what will happen in ChatGPT, Gemini or Google AI Mode, because it tested a simulated engine. Real assistants differ in ways that matter. For example, Ahrefs found that responses contained fan-out (search) queries in only 14.4% of 66.6 million free, logged-out responses ⁶.

The paper also reports that lower-ranked sites gained most. The Cite Sources method raised the fifth-ranked result's visibility by 115.1% and cut the top result's by 30.3% ³. That was a scenario in which every source applied the method, so it is a hint about fairness, not a promise for small sites.

• Queries. 80% were informational and 10% each transactional and navigational, so buying questions were a small share ³.
• Topic matters. The abstract says the efficacy of the strategies varies across domains ¹.
• The authors' own limits. They say methods may need to adapt as engines evolve, and they did not test the effect on search rankings ³.

What else is known to move AI citations?

Three industry studies add to the paper, and all three are correlations or vendor data. None proves cause. The table says how much weight each deserves.

What did two AI tools search for when we asked about GEO?

On 20 October 2026 we asked 'what is generative engine optimization GEO' to Gemini 2.5 Flash, with Google Search on, and to a Claude agent with a web search tool. Gemini chose 2 searches. The Claude agent had been told to run 6 to 10 and recorded 6 ⁹. The Claude agent searched for the research paper by name and for criticism, which is why this page covers both. The two runs listed 17 different websites, none in both lists, and only arxiv.org was the original research. One run per tool on one day is an illustration only.

Is GEO worth it? What do critics say?

The cheap parts are worth doing, and the expensive promises deserve doubt. Google tells site owners to put effective SEO ahead of 'AEO/GEO hacks', and it says llms.txt files, chopping text into tiny chunks and seeking inauthentic mentions can be ignored for Google Search ². Google adds that outside services cannot guarantee performance ¹⁰.

Forrester analyst Nikhil Lai argues that vendors exaggerate how different these practices are from SEO ¹¹. Digiday reported in May 2026 that marketers doubt AI visibility tools because different tools gave different answers to the same prompts. It listed entry prices of about $99 to $449 a month for plans from Profound and Ahrefs ¹².

Our view: the writing habits in the paper, which are sourced numbers, named quotes and clear prose, are ordinary good editing. Do them first. Then test any paid claim before you buy.

• 'GEO means gaming the AI.' The tactics that worked in the paper make a page more useful to a reader. Keyword stuffing, the gaming tactic, scored below the baseline ³.
• 'You will get the 40% lift.' That was a best case in a test engine ³.
• 'GEO replaces SEO.' Google says it is SEO ².

What does a GEO edit look like on one paragraph?

It swaps a vague claim for a sourced number and names the source. The example below uses a real, checked figure, and the edit follows two of the paper's top methods: adding a statistic and citing a source. Only add figures you can source, because a wrong number does more harm than none.

How can you apply GEO this week?

Edit three pages and build a baseline. Choose the three pages that answer your most valuable questions.

• Replace each vague claim with a specific number and name where it comes from.
• Use plain, smooth sentences. The paper's readability edits did better than a persuasive tone.
• Check that your name, services and prices match across your site, profiles and listings, since outside mentions seem to matter ⁴.
• Ask five buyer questions in ChatGPT, Gemini and Perplexity, and log whether you are named, then repeat monthly.
• Skip keyword stuffing, hidden text and fake reviews. Google warns that mass-produced pages and inauthentic mentions are poor tactics ².

What an AI assistant searches when it answers this

We gave the question below to AI search tools and recorded the web searches they ran behind the scenes. This is called query fan-out. Each phrase is a page you could publish to be found and cited.

Frequently asked questions

About the author

Sources

1. GEO: Generative Engine Optimization · arXiv (Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan, Deshpande); accepted to KDD 2024Opened October 20, 2026. Supports: Introduces Generative Engine Optimization and GEO-bench; reports that GEO can boost visibility by up to 40% in generative engine responses; says efficacy varies across domains. First submitted 16 November 2023; version 3 dated 28 June 2024; comment 'Accepted to KDD 2024'.
2. Google's Guide to Optimizing for Generative AI Features on Google Search · Google Search CentralOpened October 20, 2026. Supports: AEO and GEO are named as third-party terms; optimizing for generative AI search is still SEO; next steps include prioritizing effective SEO over 'AEO/GEO hacks' and ignoring llms.txt files, chunking and inauthentic mentions for Google Search. Page footer: last updated 2026-07-10.
3. GEO: Generative Engine Optimization (full paper, Sections 3, 4, 5, 6, 9 and Table 1) · arXiv (Aggarwal et al.), KDD 2024Opened October 20, 2026. Supports: GEO-bench of 10K queries from 25 domains (80% informational, 10% each transactional and navigational) with the top 5 Google results and gpt-3.5-turbo; metrics; Table 1 scores for nine methods; the tone and keyword-stuffing findings; Table 2 rank effects (Cite Sources +115.1% at rank 5, -30.3% at rank 1); the Perplexity.ai test on 200 samples with source texts uploaded as files; the limitations section.
4. Top Brand Visibility Factors in ChatGPT, AI Mode, and AI Overviews (75k Brands Studied) · Ahrefs, 12 December 2025Opened October 20, 2026. Supports: In a study of 75,000 brands, YouTube mentions (about 0.74) and branded web mentions (0.66 to 0.71) correlated most with AI visibility; backlinks and URL rating correlated weakly; site page count about 0.19; the authors say correlation is not causation.
5. Generative engine optimization · WikipediaOpened October 20, 2026. Supports: GEO is also called AEO and AIO; the term was introduced in a 2023 paper presented at KDD 2024; there is no consensus definition and the efficacy of third-party GEO tools is debated.
6. How ChatGPT Does Query Fan-Out (15 Million Queries Studied) · Ahrefs (Louise Linehan), 12 October 2026Opened October 20, 2026. Supports: Of 66.6 million free, logged-out ChatGPT responses (GPT-5.6 Luna, August and September 2026), 14.4% triggered a web search; the study is a snapshot of one model.
7. Do AI Assistants Prefer to Cite Fresh Content? (17 Million Citations Studied) · Ahrefs (Ryan Law), 28 July 2025Opened October 20, 2026. Supports: Across 16.975 million cited URLs, AI assistants cited URLs averaging 1,064 days since publication versus 1,432 for organic results (Ahrefs calls this 25.7% fresher); AI Overviews were slightly older than organic; cited content still averaged about 2.9 years old; the study reports correlation, not cause.
8. The half-life of an AI citation is 11 days · Profound (Davis McCain), 30 September 2026Opened October 20, 2026. Supports: Vendor study of 883,000 cited pages across seven answer engines from 8 September 2025 to 8 September 2026 (1-in-128 sample of cited URLs): the median page lost half its citation share within 11 days of peaking. Profound sells AI visibility tools and the metric is proprietary.
9. Query Fan-Out Explorer (observed examples) · Chalk LabsOpened October 20, 2026. Supports: The searches and sites recorded on 20 October 2026 for the prompt 'what is generative engine optimization GEO': Gemini 2.5 Flash with Google Search grounding (2 searches, 9 sites) and a Claude agent with a web search tool told to run 6 to 10 searches (6 recorded, 8 sites). No site appeared in both lists.
10. Google Search's Guidance on Third-Party SEO Tools & Advice · Google Search CentralOpened October 20, 2026. Supports: Google does not evaluate third-party services and they cannot guarantee performance; services promising improvements for AI experiences are listed among those to think critically about. Page footer: last updated 2026-06-05.
11. SEO's Hype-Fueled Move To The Center Of The Marketing Mix · Forrester (Nikhil Lai), 24 November 2025Opened October 20, 2026. Supports: An analyst's view that newer terms such as AEO, GEO and LLMO borrow SEO's reputation and that vendors exaggerate how different the practices are. The post also promotes Forrester's own services.
12. Marketers question expensive AI visibility tools as inconsistent results fuel skepticism · Digiday (Kimeko McCoy), 1 May 2026Opened October 20, 2026. Supports: Marketers and agency executives say AI visibility tools return different answers for the same prompts and often give point-in-time snapshots; reported plans from Profound start at $99 and $399 a month and from Ahrefs at $129, $249 and $449 a month (as listed in May 2026).
13. Architecting and evaluating an AI-first Search API · Perplexity Research, 25 September 2025Opened October 20, 2026. Supports: Perplexity decomposes documents into self-contained spans individually retrieved and ranked, merges lexical and semantic retrieval into a hybrid candidate set, and uses cross-encoder rerankers.