What Is Generative Engine Optimization (GEO) and How Do You Get Cited by AI?

Generative engine optimization (GEO) is the work of getting your business quoted and linked inside the answers that AI tools such as ChatGPT, Gemini and Perplexity write. You get cited by letting their search crawlers reach your pages, by answering the exact question in plain sentences at the top of each section, and by backing your claims with numbers, named sources and facts that only your business can publish.

The name comes from a 2023 research paper that introduced GEO, later presented at the ACM KDD conference in Barcelona in August 2024. Its authors tested nine ways of rewriting web pages on GEO-bench, a set of 10,000 search queries they built for the purpose. The best methods raised a source's visibility in AI answers by up to 40%. GPT-3.5 Turbo wrote the answers in its main experiments, so the exact figure won't carry over to the tools you use today. The direction does: AI engines favour pages that are easy to quote and easy to check.

Most of the work overlaps with SEO you may already pay for. The new part is narrower than the sales pitches suggest. It sits in how you write each page, which crawlers you let in and how you check whether AI tools mention your business.

How ChatGPT, Gemini and Perplexity choose which sources to cite

All three tools run the same basic loop. They decide whether a question needs fresh web results, turn it into one or more search queries, read the pages that come back and write an answer with links to the sources they used. A page has to survive every step to earn a citation. Each step drops pages for a different reason.

The model decides whether to search at all. Google's documentation for grounding Gemini in Google Search sets out the order: the model analyses the prompt, judges whether a search would improve the answer, and only then writes and runs its own queries. A question the model can answer from what it already knows may come back with no links. Questions that depend on current facts give it a reason to search. Prices, this year's rules and who offers a service in a given city are all examples, and those are the questions where a business can be cited.

Then one question becomes several searches. OpenAI's help page for ChatGPT search says it partners with other search providers and rewrites a question into one or more targeted queries. Its own example turns "what's the latest on the development of drugs that target CCR8 for cancer?" into "CCR8 immunotherapy drug development 2025". After reading those results, ChatGPT may send narrower follow-up searches. Google calls the same idea query fan-out, and its AI Overviews and AI Mode run related searches across subtopics before writing a response. Every sub-question is a separate chance to be the source.

Only pages the crawler can reach are in the running. OpenAI uses its OAI-SearchBot crawler to surface websites in ChatGPT search, and its crawler documentation states that sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers. Perplexity says PerplexityBot exists to surface and link websites in its results and that it respects robots.txt. Google's AI Overviews and AI Mode only use pages that are indexed and eligible to show a snippet in normal Google Search. One blanket "block all AI bots" rule in robots.txt can remove you from these answers without anyone on your team noticing.

Last, the answer is written from passages, with a few links. Perplexity says each answer includes numbered citations that link to the original sources. Gemini's grounding API goes further and ties each citation to a specific span of text in the answer. The engine is matching sentences to sources, so a page where one paragraph answers one question is simple to cite. A claim buried in the fourth paragraph of your company history gets skipped.

What the original GEO study found about getting cited

The GEO researchers measured how much of each AI answer came from each source, giving more weight to text near the top of the answer. Adding statistics, adding quotations from credible sources and citing those sources gave the largest gains, between 30 and 40% on that measure. Keyword stuffing, a tactic from early SEO, scored below pages that were left alone.

Look at the authoritative-tone row before you rewrite anything. Making a page sound more confident moved its score from 19.5 to 21.8, a fraction of what quotations and statistics did. Evidence earns the citation. Tone alone adds little.

The second finding is the one smaller firms should read twice. The researchers grouped sources by where they ranked in normal search results. 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 similar lifts, at 99.7% and 97.9%. A page that will never take the top spot in Google can still win a large share of the AI answer.

Treat these numbers as a lab result rather than a forecast. Each test query came with the text of its top five Google results, and GPT-3.5 Turbo wrote the answers. Live tools in 2026 run newer models over bigger indexes with their own ranking. The pattern still matches what the engines say about themselves. They pair sentences with sources, so sentences that carry checkable facts are the ones worth pairing.

How to get cited by AI: what to change on your pages

AI tools cite pages they can reach, that answer the question in the first lines, and that back each claim with numbers, quotes and sources the engine can check. Six changes cover all of that. None of them needs new software or a new website, and most can be done by whoever already edits your site.

1. Open every section with the answer. The two or three sentences under a heading should answer the question in that heading, and they should still make sense if an engine lifts them with nothing around them. Background, history and caveats come after.
2. Swap adjectives for numbers and say where each number came from. "Fast response times" gives an engine nothing to repeat. "Call-outs within 90 minutes, seven days a week" gives it a fact. A link to the source of any outside statistic gives it a reason to trust the page. A visible "last updated" date on pages with prices or rules stops it choosing between your new figures and a two-year-old blog post.
3. Quote people and documents by name. A line from a named engineer, a regulator's guidance note or a published standard is the material the GEO study found most effective. It also shows a human reader that you checked.
4. Answer the questions that come after the first one. Because of query fan-out, a search for "custom CRM cost" may also trigger searches on setup time, monthly fees and data migration. A page that covers each follow-up in a short, clearly labelled section has four chances to be picked instead of one.
5. Publish what only you know. Google's guidance asks for "non-commodity" content with expert or experienced takes that go beyond common knowledge. Your real price ranges, your project timelines, the questions customers ask on sales calls and the mistakes you see repeated are facts an engine cannot find on any other site.
6. Check that the right crawlers can reach you. Open yourdomain.com/robots.txt and look for rules that block OAI-SearchBot, PerplexityBot or Googlebot. GPTBot is a separate OpenAI crawler used for model training, so blocking it is a training decision and not a ChatGPT search decision. OpenAI says robots.txt changes take about 24 hours to reach its systems.

Here is how that plays out on one page. A 12-person accounting firm has a services page that promises "expert tax planning for growing businesses". After the rewrite, the first lines under the heading say who the service is for (companies with 10 to 200 staff), how it is priced (a fixed monthly fee, with the range stated), how long onboarding takes (three weeks) and which filing deadlines it covers. A link to the tax authority's official calendar backs the deadlines. The page is no longer than before. It now holds five facts an engine can quote and check, where the old version held none.

What you can skip: llms.txt, content chunking and special schema

Google says the popular GEO add-ons are not needed for its AI features. Google's guide to generative AI search, last updated in July 2026, states that you don't need new machine-readable files, AI text files, markup or Markdown to appear in Google Search. It adds that there is no requirement to break content into tiny pieces and no special schema.org markup to add for generative AI search.

That matters because a good share of GEO selling is built on exactly these items. An llms.txt file, a site cut into hundreds of 50-word fragments, or a schema package sold as an "AI visibility" upgrade are simple to sell and look good in a monthly report. For Google, none of them decides if you appear. The same guide says you don't need to write in a special way for AI search either.

Structured data still has its ordinary uses. Accurate markup for products, reviews or events can earn rich results in normal Google Search, so add it for that reason. Don't buy it as a shortcut into AI answers.

GEO vs SEO vs AEO: three names for overlapping work

GEO, SEO and AEO overlap far more than they differ. Google's July 2026 guide says that, from Google Search's point of view, work aimed at its generative AI features is still SEO. The difference is the result you count. SEO counts rankings and clicks, while GEO and AEO count how many times your content becomes part of the answer itself.

The three terms describe one website from three angles. SEO is the base: pages that search engines can crawl, index and rank. AEO is about being the answer when someone asks a direct question, in a Google AI Overview or a chatbot reply. GEO, the term from the 2023 study, looks at the generative engines that write those replies and at how much of each reply your content shapes. The earlier Techno Mantraa guide explains how AEO differs from SEO in more detail, and nearly all of it applies here.

The counting difference is real. A Pew Research Center study of 900 US adults in March 2025 found that people clicked a traditional search result on 8% of visits when an AI summary appeared, against 15% when none did. They clicked a link inside the summary on just 1% of visits. A page can lose clicks to an AI Overview and still be one of the sources that overview names. That makes a citation closer to a brand impression than a visit, and an SEO report built only on clicks will miss it.

The case for treating GEO as a separate budget line is narrow. If your sales calls show buyers arriving with a shortlist they built in ChatGPT or Perplexity, give monthly AI checks a named owner. If they don't, answer-first pages on a sound SEO base will capture most of the gain.

How to measure generative engine optimization results

Measure GEO with three sources side by side: Google Search Console's generative AI performance reports for impressions in Google's AI features, Bing Webmaster Tools' AI Performance report for citations in Microsoft's AI answers, and a monthly manual check of 20 to 30 buyer questions in ChatGPT, Gemini and Perplexity.

Google announced its generative AI performance reports on 3 June 2026 and says they reached all websites worldwide by 31 August 2026. They count the impressions your URLs earned in generative AI features in Search and Discover, list the pages that appeared, and split the data by country, device and date. Compare the pages listed there with the pages you rewrote. If rewritten pages start to appear and untouched pages don't, the changes are working.

Bing Webmaster Tools added its AI Performance report as a public preview on 10 February 2026. It counts the times your content is cited as a source in Microsoft's AI answers, shows which pages were cited and tracks the trend over time.

The chat tools still need a spreadsheet. Pick 20 to 30 questions your buyers ask before they contact you. Run each one in all three tools on the first working day of every month, and record three things: if you were cited, which competitor was cited instead and which of your pages was used. Use a logged-out browser window, because OpenAI says ChatGPT can use saved memories when it rewrites a search query. Then check your analytics for referral visits from chatgpt.com, perplexity.ai and gemini.google.com. Add "an AI tool like ChatGPT" to the "how did you hear about us" options on your enquiry form as well.

A first month might show you cited on 3 of 25 questions in ChatGPT, 5 in Perplexity and 2 in Gemini, with one competitor named on 14. Open that competitor's cited pages next to yours. Look for the facts they state that you don't: a price range, a timeline, a named standard. Give every change three months of monthly checks before you judge it. Content changes wait on the next crawl, and one month is too few data points to call a trend.

Where to start this month

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

1. Read your robots.txt and make sure OAI-SearchBot, PerplexityBot and Googlebot are allowed.
2. Choose the five pages that answer your buyers' most expensive questions, such as pricing, timelines and comparisons with alternatives.
3. Rewrite the first two or three sentences under each heading on those pages so they answer the heading directly, with a number and a source wherever one exists.
4. Add one fact to each page that only your business can state.
5. Set up the Search Console and Bing reports, then run your first 25-question check so you have a baseline to compare against.

If you would rather hand the SEO and content work to a team, send those five pages to info@technomantraa.com or use the Get a Quote form on the Techno Mantraa contact page for a custom quote. Either way, the page that states the fact gets the citation.