What Is Generative Engine Optimization (GEO) and How Does It Work?

Generative engine optimization (GEO) is the practice of shaping your pages so AI tools such as ChatGPT and Perplexity quote and link them in their answers. It builds on SEO and adds three jobs. Let AI crawlers in, answer at the top of each section and back claims with facts an engine can check.

Most of that work overlaps with SEO you may already pay for. The new part is narrower than the sales pitches suggest, and it rests on one 2023 study plus what AI companies publish about their own crawlers.

On this page

• What is GEO and where did the term come from?
• How ChatGPT, Gemini and Perplexity choose sources to cite
• Which AI crawlers need access to your site
• What the original GEO study found
• Lower-ranked pages gained the most from GEO methods
• How to get cited by AI on the pages you already have
• What one rewritten service page looks like
• Which GEO add-ons you can skip
• GEO vs SEO and AEO compared
• Is generative engine optimization worth a separate budget?
• How to measure generative engine optimization results
• Where to start with generative engine optimization this month

What is GEO and where did the term come from?

GEO is the work of earning a place in AI-written answers, and the name comes from a research paper. Pranjal Aggarwal and five co-authors posted "GEO: Generative Engine Optimization" on arXiv on November 16, 2023. The team, which includes researchers from Princeton University and IIT Delhi, presented it at the ACM KDD conference in Barcelona in August 2024.

Its authors built GEO-bench, a set of 10,000 search queries, and tested nine ways of rewriting a source page. The best methods raised a page's visibility in AI answers by up to 40%. They called the field generative engine optimization. The name fits, because these tools write a fresh answer from several sources instead of listing links.

The term now has company. Many agencies sell the same service as AI search optimization, which is what our AI development team calls it. Others say LLM SEO or AEO.

How ChatGPT, Gemini and Perplexity choose sources to cite

All three tools run the same loop. They decide whether to search, run one or more searches, read what comes back and write an answer with links. Your page has to survive every step to earn a citation.

First, the model decides whether a search would help. Google's Gemini docs say the model "analyzes the prompt and determines if Google Search can improve its response."

If it can answer from memory, you get no links at all. Questions about prices, this year's rules or local providers give it a reason to search, and those are the questions where a business can earn a citation.

Then one question can turn into several searches. ChatGPT searches on its own when a question "might benefit from information on the web," OpenAI says. It may also pass queries to providers such as Bing. Google calls its version query fan-out, a set of related queries the model writes to gather more facts.

Last, the engine builds its reply from passages. Perplexity says every answer has numbered footnotes that link to the original sources. Gemini's grounding data goes further and ties every source to a span of text in the reply. A page where one paragraph answers one question gives the engine a clean match.

Which AI crawlers need access to your site

Only pages a crawler can reach are in the running. OpenAI's crawler documentation says a site that blocks OAI-SearchBot won't appear in ChatGPT search answers. GPTBot gathers data to train OpenAI's models, so blocking it is a training decision, separate from search. OpenAI adds that robots.txt changes take about 24 hours to reach its systems.

Perplexity says PerplexityBot exists to surface and link websites in its results, and it doesn't crawl content to train AI models. Google's AI Overviews and AI Mode only use pages that Google has indexed and that qualify for a snippet.

So one blanket "block all AI bots" rule can drop you from these answers without anyone on your team noticing. Before you edit the file, read what robots.txt can and cannot control for AI crawlers. In October 2026, a Cloudflare and OpenAI pilot for ChatGPT search helped OpenAI's crawler find pages with few inbound links.

What the original GEO study found

Quotations and statistics gave the biggest lifts in the study. The researchers scored how many words of each AI answer came from each source. Adding quotations from credible sources took a page's score from 19.5 to 27.8, a rise of about 43%. Keyword stuffing, an early SEO trick, scored below the untouched pages.

Look at the tone row before you rewrite anything. A more confident tone moved the score from 19.5 to 21.8, less than a third of the gain from quotations. Evidence earns the citation. Confident wording on its own adds little.

The results also shifted by topic. Statistics helped most on law and government questions, the paper reports, and quotations worked best for people, society and history. Cited sources did best on factual questions. Match the method to the questions your buyers ask.

Lower-ranked pages gained the most from GEO methods

Pages lower in Google's results gained far more than the top result, which matters for smaller firms. The researchers grouped sources by where they ranked in Google. Citing sources lifted the visibility of fifth-ranked pages by 115.1%, while top-ranked pages lost 30.3% on average under the same method.

Quotations and statistics gave fifth-ranked pages lifts of 99.7% and 97.9%. Even a page that never reaches the top spot in Google can win a large share of an AI answer. It needs the quote or the number the answer rests on.

The team also tested its methods on Perplexity, a live engine, across 200 queries. Adding quotations did best there too, with a 22% gain over the baseline, and keyword stuffing did 10% worse than changing nothing.

Treat these numbers as a lab result. In the main tests, GPT-3.5 Turbo wrote the answers from the top five Google results for each query. Live tools in 2026 run newer models over bigger indexes. The pattern matches what the engines say about themselves, since they pair sentences with sources and a sentence with a checkable fact is the easiest to pair.

How to get cited by AI on the pages you already have

Six changes cover most of the gap. None of them needs new software or a new website, and whoever edits your site today can make them.

• Open every section with the answer. The first two or three sentences under a heading should answer it and still make sense if an engine lifts them out alone.
• Swap adjectives for numbers and name the source. "Fast response times" gives an engine nothing to repeat. "Call-outs within 90 minutes, seven days a week" gives it a fact.
• Quote named people and documents, such as an engineer, a regulator or a published standard, because added quotations did best in the GEO study.
• Cover the follow-up questions. With query fan-out, a search for "custom CRM cost" may also pull in setup time, monthly fees and data migration. A page with a short section on each has four chances to get picked instead of one.
• Publish what only you know, such as your price ranges, timelines and the questions buyers ask on sales calls. That's what Google's guide means by "unique, expert-led content."
• Put a visible "last updated" date on pages with prices or rules so readers and engines can tell your current figure from a two-year-old post.

What one rewritten service page looks like

Say a 12-person accounting firm has a services page that promises "expert tax planning for growing businesses" and little else. That line gives an engine nothing it can quote or check.

After the rewrite, the first lines say who the service is for: companies with 10 to 200 staff. They give the price as a fixed monthly fee, with the range stated. Then they state the onboarding time (three weeks) and the filing deadlines it covers, linked to the IRS tax calendar.

The page is no longer than before. It now holds four facts and a source, where the old version held none.

Which GEO add-ons you can skip

Google says its AI features don't need the popular add-ons. Google updated its guide to generative AI features in Search on July 10, 2026. It says you don't need new machine-readable files, AI text files, markup or Markdown to appear. There's no need to break content into tiny pieces or add special schema.org markup.

A good share of generative engine optimization selling rests on these items. Agencies sell llms.txt files, pages cut into 50-word fragments and schema packages billed as "AI visibility" upgrades. Google says Search itself doesn't use llms.txt files, and you don't need a special writing style for AI search either. The guide to how Google AI Mode picks the pages it links to covers that side in more depth.

Structured data keeps its normal uses. Accurate product, review or event markup can earn rich results in regular Google Search, so add it for that reason. Don't buy it as a shortcut into AI answers.

GEO vs SEO and AEO compared

GEO, SEO and AEO overlap far more than they differ. Google's July 2026 guide says work aimed at its generative AI features is still SEO. The real split is where you want to appear and what you count.

On GEO vs AEO, the line is even thinner, and many guides treat the two as one. AEO usually centers on direct answers, from featured snippets to chatbot replies. GEO, the term from the 2023 study, looks at how much of a written reply your content shapes. For the AEO side in full, read what answer engine optimization is and how it differs from SEO.

Is generative engine optimization worth a separate budget?

For most small firms, no, because answer-first pages on a sound SEO base capture most of the gain. The case for its own budget line is narrow. If buyers reach your sales calls with a shortlist from ChatGPT or Perplexity, give AI checks a named owner.

The way you count results does change. A Pew Research Center study of 900 US adults tracked their Google searches in March 2025 and found that 18% showed an AI summary. When one appeared, people clicked a normal result on 8% of visits, against 15% when none did.

They clicked a link inside the summary on 1% of visits. So a citation works more like a brand mention than a visit, and a report built only on clicks will miss it.

How to measure generative engine optimization results

Use two webmaster reports and one spreadsheet. Google began rolling out generative AI performance reports in Search Console in early June 2026 and says all websites had them by August 31, 2026.

The reports count impressions from AI Overviews, AI Mode and AI features in Discover, split by page, country and date. They had no click data when the worldwide rollout finished. Compare the pages listed there with the pages you rewrote.

Bing added its AI Performance report in Bing Webmaster Tools as a public preview on February 10, 2026. It shows how often Copilot and Bing's AI summaries cite you and which of your URLs they pick. It also lists sample grounding queries, the phrases the AI used to find your content.

For the chat tools, pick 25 questions your buyers ask before they contact you. Run each one in all three tools on the first working day of every month. Note whether it cited you, which competitor it cited instead and which of your pages it used. Stay logged out: OpenAI says ChatGPT with Memory on "may use relevant details to formulate a more useful search query."

A first month might show you cited on 3 of 25 questions in ChatGPT, 5 in Perplexity and 2 in Gemini. One rival might show up on 14. Open that rival's cited pages next to yours and list the facts they state that you don't. Give each change three months before you judge it, because content changes wait on the next crawl.

Where to start with generative engine optimization this month

Do the work in this order, because each step makes the next one easier to measure.

1. Open yourdomain.com/robots.txt and confirm it lets OAI-SearchBot, PerplexityBot and Googlebot in.
2. Pick the five pages that answer your buyers' most expensive questions, such as pricing, timelines and comparisons with other options.
3. Rewrite the opening lines under each heading so they answer it. Add a number and a source wherever you have one.
4. Add one fact to each page that only your business can state.
5. Turn on the Search Console and Bing reports, then run your first 25-question check as a baseline.

Run the same 25 questions three months later. You'll see whether generative engine optimization has moved your citations or the pages need another pass. If you'd rather hand the rewrites and tracking to a team, talk to our SEO team about AI search visibility.