What is GEO, or generative engine optimization?
GEO is generative engine optimization: writing and publishing a page so that ChatGPT, Claude, Gemini, Perplexity or Google AI Mode picks it as a source when they write an answer. The name comes from a 2023 research paper by Aggarwal and colleagues. It isn't the geostationary orbit that shares the abbreviation, a ring 35,786 km above the equator.

| Term | What it means | Who uses it |
|---|---|---|
| GEO | Getting a page cited in an answer an AI writes | The 2023 paper; Bing's documentation, per Wikipedia |
| AEO | The same goal, often described as structuring content so an engine can lift a clear answer | Microsoft Advertising and HubSpot, per Wikipedia |
| SEO | Ranking in a list of links | Google, whose AI features run on the same index |
Nobody owns the definition yet. Wikipedia's article on the subject says there is no consensus and that usage varies by practitioner and vendor. Profound, an AI visibility company, says GEO and AEO are the same practice and prefers AEO because GEO can be confused with geography.
What is answer engine optimization (AEO)?
Answer engine optimization is the same job under a second name, and the two terms differ more by vendor than by method. Microsoft Advertising uses AEO for optimizing toward assistants and agents such as Copilot, and GEO for generative AI search. HubSpot uses AEO.
The agency New Traffic Tail, currently the second Google result for this query, defines AEO as structuring content so an engine can pull a clear, self-contained answer to a specific question. It defines GEO as getting retrieved and cited when an engine writes its own answer. That split is handy for planning, but the tactics overlap almost entirely. This page says GEO and means both.
What did the Princeton GEO paper measure?
The paper is "GEO: Generative Engine Optimization" by Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan and Ameet Deshpande. It was posted to arXiv on 16 November 2023 and accepted to KDD 2024. Three authors list Princeton University, one lists IIT Delhi and two list themselves as independent.
The team built GEO-bench, 10,000 queries across 25 domains, each paired with the top five Google results. A simulated engine fed those five pages to a GPT-3.5 model and asked for a cited answer. The authors then rewrote one source page with each of nine tactics and measured how much of the answer came from it, discounted by where the citation sat.
Those are the Overall column of the paper's Table 1, the position adjusted word count, which weighs how much of the answer comes from your page and how early it appears. Quotation Addition and Statistics Addition won, Fluency Optimization (24.7) and Cite Sources followed, and Keyword Stuffing landed below the baseline. Technical Terms, Easy to Understand and Authoritative sat in the low 20s, and Unique Words stayed near baseline at 20.5. The authors call the best result a 41% gain over baseline.
The second finding matters for small sites. Cite Sources cut the visibility of the top ranked page by 30.3% and raised the fifth ranked page by 115.1%. Quotation Addition raised fifth ranked pages by 99.7%. The authors conclude that lower ranked sites gain more from GEO than leaders do.
The authors also ran 200 test queries through Perplexity, uploading the source text as files. They report Quotation Addition about 22% above baseline there, Statistics Addition up to 37% on the impression score, and Keyword Stuffing about 10% worse than baseline. The direction matched, though the test was small and ran in 2023. The paper also found that combining tactics helped: Fluency Optimization plus Statistics Addition beat any single tactic by more than 5.5%.
Why does GEO matter for a website?
GEO matters because many questions now end in a written answer that names only a few pages. If yours is one of them, the reader sees your name before any click. If it isn't, the answer can satisfy them without you.
The evidence for how big that shift is varies by source. Our own AI search statistics page collects the figures with their dates. What the primary docs do settle is the plumbing. Google says visits from AI features are counted in the normal Web report in Search Console, so the traffic isn't hidden. OpenAI, Anthropic and Perplexity each run a separate search crawler, so your robots.txt decides whether their answers can reach you.
A small site has the most to gain. The paper's fifth ranked pages gained far more from added citations than the top ranked pages did. That makes GEO one of the few edits where the underdog starts with an edge.
How do AI answers pick which sources to cite?
Each engine finds pages through its own search index or crawler, retrieves a handful, and lets a model write the answer from them. Google calls the retrieval step query fan-out: it issues several related searches across subtopics and data sources, then links a wider set of supporting pages. Being in the index comes first.
The crawler names matter because each company splits search, training and user requests into separate bots. The table is compiled from each company's own documentation, read on 22 November 2026.
| Engine | To appear in answers | Training crawler | Fetch when a user asks |
|---|---|---|---|
| Google AI Overviews and AI Mode | Indexed and eligible for a snippet, with no additional requirements | Not covered on the page read | Not covered on the page read |
| ChatGPT | OAI-SearchBot | GPTBot, an independent setting | ChatGPT-User; robots.txt rules may not apply |
| Claude | Claude-SearchBot | ClaudeBot | Claude-User |
| Perplexity | PerplexityBot | Not used for training, per its docs | Perplexity-User; generally ignores robots.txt |
The consequence is concrete. OpenAI says sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers, and that blocking GPTBot doesn't change that. A site can refuse training and still be cited.
Our own check shows the pull toward the open web. In the 144 answers behind which AI app do chatbots recommend, 140 cited at least one web page, Reddit appeared in 33 and YouTube in 26. That was one 30 question set on one day in November 2026, so read it as a sample.
How is GEO different from SEO, and does it replace it?
GEO changes the goal from ranking in a list to being quoted in an answer, but it doesn't replace SEO, because the same index feeds the answer. Google's documentation says "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary." A page that can't be indexed can't be cited.
Google's May 2025 guidance from John Mueller reads like old SEO advice with one emphasis. Make unique, non commodity content. Keep the page fast and crawlable, and keep structured data truthful. Mind your preview controls too, because Google says more restrictive permissions limit how your content is featured in AI experiences. There is no special schema to add.
The real differences are in what you write and what you can measure.
| SEO | GEO | |
|---|---|---|
| Win condition | A high place in a list of links | A citation or mention inside a written answer |
| What the paper found helps | Keywords in the title and body | Quotations, statistics and cited sources |
| Keyword stuffing | Old tactic | Scored 17.7 against a 19.3 baseline |
| Where you see it | Search Console | AI features are folded into the Web report, with no separate view, per Google |
Forrester's Nikhil Lai calls AEO and related terms significantly, but not fundamentally, different from SEO, as quoted by Wikipedia. That view fits the evidence above. Treat GEO as SEO with a sharper edit pass.
How do you measure GEO, and how long does it take?
Measure GEO by asking the engines your target questions on a schedule and logging which pages they cite. For Google, AI Overviews and AI Mode traffic is counted in the Performance report under the Web search type, with no separate view, per Google Search Central.
No source read for this page gives a reliable timeline for results. The one hard number is OpenAI's: robots.txt changes can take about 24 hours to be picked up by its search system. The agency New Traffic Tail says GEO often moves slower than AEO because it depends on signals outside your own pages, such as reviews, directories and communities. That is its claim, not a measured result. The paper tested only edits to the source page itself.
You don't need a paid tool to start. The checklist below is free, and the authors published their prompts and code at github.com/GEO-optim/GEO if you want to rerun their method.
What is a GEO checklist you can use today?
Nine changes cover what the sources above support. Each one is cheap, and none needs a tool.
- Allow OAI-SearchBot, Claude-SearchBot and PerplexityBot in robots.txt if you want to be cited, and set GPTBot and ClaudeBot on their own merits.
- Make sure Googlebot can index the page with a snippet, and don't add nosnippet.
- Answer the page's question in the first sentence, with a name and a number.
- Add a real quotation from a primary source, attributed by name.
- Add statistics with the year and the study or company behind each.
- Cite sources in the text and link them, as the paper's Cite Sources tactic did.
- Write short, plain sentences, since Fluency Optimization ranked fourth in the paper.
- Keep markup accurate to the visible page, with no special schema required.
- Ask two or three engines your target question each month and log who is cited.
The ninth step is a sample, not a rank. Answers change between runs, so a log of five runs tells you more than one screenshot.
What do real cases of GEO look like?
Three cases from the sources show the tactics doing, or not doing, their job.
- A fifth ranked page adds citations. In the paper, Cite Sources raised the visibility of the fifth ranked page by 115.1% and lowered the top ranked page by 30.3%. The lesson: a small site that links its sources gains the most, and a leader gains nothing from the same edit.
- A page repeats its keyword. Keyword Stuffing scored 17.7 against the 19.3 baseline in the main test and came out about 10% worse on Perplexity. The lesson: the old habit costs you here.
- A publisher blocks OAI-SearchBot. Per OpenAI, that site will not be shown in ChatGPT search answers, though it can still appear as a navigational link. The lesson: robots.txt is a GEO setting, and OpenAI says changes can take about 24 hours to apply.
What can you ask an AI about GEO?
These six prompts work in Claude, ChatGPT, Gemini or Whizi. Replace the brackets with your own details.
- Audit my page: "Here is my page: [paste]. Which sentences could an AI answer lift whole, and which claims have no source? List both."
- Answer first: "Rewrite the first paragraph so the answer to [question] is in sentence one. Use only these facts: [facts]."
- Source check: "List every number in this draft and mark the ones with no primary source I can link: [paste]."
- Robots file: "Write robots.txt rules that allow OAI-SearchBot, Claude-SearchBot and PerplexityBot and block GPTBot and ClaudeBot. Explain each line in one sentence."
- Compare to a rival: "Compare my page with this competitor page on facts, quotations and cited sources. Give me three edits: [paste both]."
- Citation log: "What pages would you cite for [question], and why those? Search the web first."
Prompt 6 asks the model about its own choices, so treat the reasons as a guess and the cited links as the evidence.
What mistakes ruin a GEO effort?
Four mistakes cause most wasted work.
- Blocking every AI bot out of habit. The fix: allow the search crawlers, then choose on training separately.
- Repeating keywords. The fix: add one quotation or one sourced number per section instead.
- Quoting a statistic that only one site states. The fix: find the primary source or label it as that site's claim. Fabricated numbers are the fastest way to get a page quoted wrongly.
- Judging results from one run. The fix: log several runs a month, in more than one engine.
Where does Whizi fit when you test GEO?
Whizi isn't a GEO tool and doesn't track citations, but it is one app for running the same question through several models. Web search is on by default for a new chat, so you can ask which pages a model would cite and read the links.
Pro carries Claude, GPT and Gemini in one app, and its first week is $1.99. Compare, on Pro and Powerhouse, puts two models side by side on one prompt, which is a fast way to see where their sources differ. The free plan gives 10 messages a day on Whizi AI, Auto and Claude Haiku, with no card.
What can GEO not do?
GEO can't guarantee a citation, and nobody outside the engines can see why one page was picked. The paper used a simulated engine built on a GPT-3.5 model and one Perplexity check, and today's engines have changed.
Most advice online comes from companies that sell GEO services, so ask for the primary source behind any number. How to fact check AI answers shows a two minute routine, and what is RAG explains the retrieval step that GEO is trying to reach.
- Allow the search crawlers you want citations from, and set training bots separately
- Put the answer, with a name and a number, in the first sentence
- Add one attributed quotation and one sourced statistic per section
- Cite and link your sources in the text
- Log citations from two or three engines each month
Frequently asked questions
What is GEO in SEO?
GEO is generative engine optimization, the practice of making a page the source an AI answer cites. In SEO it sits next to ranking: the same index feeds both, and Google says no special optimization is needed for AI Overviews, as of November 2026.
Is GEO the same as AEO?
Mostly yes. Both aim to get a page used in an AI written answer, and the split is by vendor: Microsoft Advertising uses AEO for assistants and GEO for generative search, per Wikipedia. The tactics overlap almost entirely.
Does keyword stuffing help GEO?
No. In the 2023 GEO paper, Keyword Stuffing scored 17.7 against a 19.3 baseline, and it came out about 10% worse on Perplexity. Quotations and statistics scored 27.2 and 25.2.
Can a small site compete in AI answers?
Yes, and it may have an edge. In the 2023 paper, Cite Sources raised the visibility of fifth ranked pages by 115.1% and lowered the top ranked page by 30.3%. That was a lab test on a GPT-3.5 model, so treat it as a direction.
How long does GEO take to work?
No primary source gives a timeline. OpenAI says robots.txt changes can take about 24 hours to apply, and the paper tested only page edits, not how long engines take to notice them.
Should I block GPTBot?
Blocking GPTBot stops training use only. OpenAI says each setting is independent, and sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers, so allow that one if you want citations.