Generative engine optimization (GEO) is the work of getting your pages seen and cited inside AI-generated answers: the written summaries that AI search tools produce instead of, or above, a list of links. You will also see it searched as “GEO SEO” or “LLM SEO”. They mean the same goal. This page sits under our guide to the difference between People Also Ask and AI Overviews, and it covers where the term came from, what the research behind it found, and what Google says you should and should not do about it.
Where the term GEO came from
GEO started as an academic idea, not a Google term. The paper “GEO: Generative Engine Optimization” was first posted to arXiv on 16 November 2023 by Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan and Ameet Deshpande, a team that included researchers from IIT Delhi and Princeton University. The third version (28 June 2024) is marked “Accepted to KDD 2024”, the ACM SIGKDD Conference on Knowledge Discovery and Data Mining.
The paper describes “a new paradigm of search engines that use generative models to gather and summarize information to answer user queries,” which it calls generative engines. The authors’ worry was the people who write the pages. They wrote that content creators “have little to no control over when and how their content is displayed” once an AI writes the answer. GEO was their proposed fix: a way to measure how visible a source is inside an AI answer, and a set of edits to test against that measure.
For the wider naming debate (GEO, AEO, LLMO and the rest), see what SEO is called now with AI. The closely related answer engine optimization has its own page, including how the two terms differ.
What the GEO paper actually tested
The research setup matters, because most GEO advice online quotes the paper’s headline number without it. The authors built a benchmark called GEO-bench, “consisting of 10000 queries from diverse domains and sources.” For each query, they fetched the top 5 sources from Google, fed them to OpenAI’s gpt-3.5-turbo model, and had it write an answer citing those sources. Then they rewrote one source at a time using a given method and measured whether that source got more of the answer.
They used two measures of visibility. Position-Adjusted Word Count looks at how many words of the answer cite a source and how early they appear. Subjective Impression scores softer qualities, such as how relevant and influential the citation is.
Here are the nine methods and what the paper reports for each:
| Method | What it changes | What the paper found |
|---|---|---|
| Cite Sources | Adds relevant citations from credible sources | One of the three top performers |
| Quotation Addition | Adds quotations from credible sources | One of the three top performers |
| Statistics Addition | Swaps qualitative discussion for numbers where possible | One of the three top performers |
| Fluency Optimization | Makes the text read more smoothly | A 15 to 30% visibility boost |
| Easy-to-Understand | Simplifies the language | A 15 to 30% visibility boost |
| Authoritative | Makes the tone more persuasive and authoritative | No significant improvement overall |
| Keyword Stuffing | Adds more keywords from the query | Little to no improvement |
| Unique Words, Technical Terms | Adds unusual or technical words | Tested, but not among the methods the paper highlights |
The top three, the paper says, “achieved a relative improvement of 30-40% on the Position-Adjusted Word Count metric and 15-30% on the Subjective Impression metric.” That is where the widely quoted “up to 40%” comes from. The authors also ran the methods on Perplexity.ai, a live AI search engine, and reported “visibility improvements up to 37%.”
Two findings are more interesting than the headline.
- Keyword stuffing did not work. The authors wrote that it offers “little to no improvement on generative engine’s responses,” even though it was a classic search tactic.
- Lower-ranked pages gained the most. In the paper’s analysis by search position, the Cite Sources method “led to a substantial 115.1% increase in visibility for websites ranked fifth in SERP, while on average, the visibility of the top-ranked website decreased by 30.3%.”
Domain mattered too. The authors found Authoritative helped on debate-style and history questions, Cite Sources helped most on factual questions, and Statistics Addition helped in areas like law and government. Their conclusion was that GEO needs “domain-specific optimization methods.”
The limits of the research
Read the paper as a lab study, not a description of Google. It tested a simulated engine built on GPT-3.5 Turbo and five Google results per query, plus Perplexity. The authors note in their limitations section that methods “may need to adapt over time as GEs evolve,” and that “we didn’t evaluate how GEO methods affect search rankings.” Google’s AI Overviews and AI Mode were not what was measured. None of this means the findings are wrong. It means the numbers describe that setup, not any one commercial engine today.
Is GEO different from SEO?
Google’s answer is no, at least for Google Search. Its guide to optimizing your website for generative AI features on Google Search (last updated 10 July 2026) names both terms directly: “‘AEO’ stands for ‘answer engine optimization’ and ‘GEO’ for ‘generative engine optimization’.” It then says: “From Google Search’s perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO.”
The reason Google gives is that its AI features “are rooted in our core Search ranking and quality systems.” It describes them using retrieval-augmented generation (grounding), which relies on “our core Search ranking systems to retrieve relevant, up-to-date web pages from our Search index.” In other words, the AI writes from pages that Search has already found and ranked.
That still leaves real differences in what you aim for and how you check it:
| Classic SEO | GEO | |
|---|---|---|
| The goal | A page that ranks in a list of results | A page the AI answer cites or draws on |
| What the reader sees | Your title and snippet | A written answer, with your link among the sources |
| Where it happens | Search results pages | AI Overviews, AI Mode, ChatGPT, Perplexity and similar tools |
| How to measure it in Google | Search Console Performance report | Search Console generative AI performance report (impressions) |
| What Google says to do | SEO best practices | The same SEO best practices |
For a page on whether the job itself is going away, see is SEO dead now with AI?
What Google says you do not need to do for GEO
The same guide has a section called “Mythbusting generative AI search: what you don’t need to do”. It says that while AEO and GEO are common terms, “many suggested ‘hacks’ aren’t effective or supported by how Google Search actually works.” The list:
- llms.txt and other special files. “You don’t need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search.” Google says creating them “will neither harm nor help” your visibility in Google Search, though other services may use them.
- Chunking content. “There’s no requirement to break your content into tiny pieces for AI to better understand it.” Google adds that “There’s no ideal page length.”
- Rewriting for AI. “You don’t need to write in a specific way just for generative AI search.”
- Buying or faking mentions. Seeking inauthentic mentions “isn’t as helpful as it might seem.”
- Overdoing structured data. “Structured data isn’t required for generative AI search, and there’s no special schema.org markup you need to add.” Google still recommends it for rich result eligibility.
There is also a warning that anyone working from question lists should read twice. Google says that creating separate content for every variation of a search, “for example, by focusing on other queries that people have asked, or fan-out queries,” mainly to manipulate rankings or AI responses, violates its scaled content abuse policy. Google’s spam policies define that as “when many pages are generated for the primary purpose of manipulating search rankings and not helping users.”
What Google says does help
Google’s positive advice falls under three headings in the guide. Here they are, with the parts most relevant to GEO.
Create non-commodity content
Google says unique, useful content “will likely influence your website’s presence in generative AI search in the long run more than any of the other suggestions in this guide.” It contrasts commodity content, such as “7 Tips for First-Time Homebuyers”, with first-hand pieces that offer “unique expert or experienced takes that go beyond common knowledge.” Its reasoning is specific to AI: “Our AI systems take a look at a variety of sources, so it can be helpful to have a unique viewpoint that stands out.”
This lines up with the GEO paper more than it first appears. Citing sources, quoting people and adding real numbers are all ways of bringing something a summary of other pages would not have. The difference is that Google wants them because they are true and useful, not as a formatting trick.
Keep the technical basics in order
Google’s AI features and your website page (last updated 10 December 2025) sets the bar: to be shown as a supporting link in AI Overviews or AI Mode, “a page must be indexed and eligible to be shown in Google Search with a snippet”, and “There are no additional technical requirements.” The newer guide adds that a site must also be “included in Search generative AI features in Search Console”. That means Search Console’s Search generative AI control: inclusion is the default, Google rolled it out to all websites worldwide as of 31 August 2026, and a site that chooses to be excluded “won’t receive any traffic or impressions from these features.” The guide’s checklist covers allowing crawling in robots.txt and at your CDN, internal links, page experience, important content in text form, and structured data that matches the visible page.
The AI features page names the page-level controls if you want less exposure: nosnippet, data-nosnippet, max-snippet and noindex limit what Google shows from your pages, including in AI features.
Keep business details current
For shops and local businesses, Google says AI responses can include product listings and business information, and points to Merchant Center feeds and Business Profiles.
GEO beyond Google
Google’s guide only speaks for Google. GEO, as the paper defined it, covers any generative engine. We cover the main ones on their own pages:
- Google AI Overviews: how do you rank in AI Overviews?, and for the feature itself, what is a Google AI Overview?
- ChatGPT: how do you appear on ChatGPT?
- Perplexity: how does Perplexity SEO work?
How do you measure GEO?
For Google, Search Console now has a generative AI performance report. Its help page says: “As of August 31, 2026, we’ve rolled out these insights to all websites worldwide.” It covers AI Overviews and AI Mode, and shows impressions, which it defines as “how many times links to your site were shown to a user in a generative AI feature on Google Search.” You can group them by page, country, device and date.
Outside Google you are mostly relying on third-party trackers. Google’s guide is blunt about those: “No third-party tool has access to our internal ranking or AI systems.” Use them if they help, and check their advice against Google’s. We cover the options in how AI Overview tracking works and in our roundup of ai seo tools.
A practical GEO routine
Pulling the research and Google’s guidance together, a sensible routine looks like this:
- Check eligibility first. Is the page indexed and allowed a snippet? Is crawling allowed in robots.txt and by your CDN? Nothing else matters until these are true.
- Answer the question early and plainly. Google says people appreciate pages organized “by paragraphs and sections, along with headings that provide a clear structure.”
- Add what a summary cannot. First-hand experience, your own data, named sources and exact quotes. This is the overlap between Google’s non-commodity advice and the GEO paper’s top methods.
- Skip the hacks. For Google, that means no llms.txt expectations, no chunking, no rewriting for machines and no fake mentions.
- Cover the questions around the topic on one strong page. Do not spin up a thin page for every variation. Google calls that out by name.
- Measure. Watch the generative AI performance report in Search Console, and check AI answers for your key topics by hand.
Step 5 is where People Also Ask helps. PAA shows you the related questions Google links to a topic, which is a good map of what a complete page should cover. PAA Box is our Chrome extension for collecting those questions: it opens the People Also Ask box up to four levels deep and exports the questions to CSV. It only reads the People Also Ask block. It does not read AI Overviews, AI Mode, ChatGPT or Perplexity, and it does not track whether you are cited. Use the questions to plan sections on pages that deserve to exist, not as a list of pages to mass-produce.