GEO 101: What Generative Engine Optimization Is and How It Works

Last updated: November 2026

Generative engine optimization (GEO) is the work of getting your brand named, cited and described correctly in the answers that AI assistants write. ChatGPT, Gemini, Perplexity, Claude and Google’s AI Overviews now answer the questions buyers used to type into a search bar. GEO decides whether your company is in those answers.

This guide covers the basics: how the assistants build an answer, what gets a page cited, which popular tactics have no evidence behind them, and how to measure where you stand. It draws on a review of 68 published studies on AI citations (2023 to 2026), our own controlled experiments, and the GEO work we run with insurers and financial services companies.

What is GEO?

GEO (generative engine optimization) is the practice of improving how often and how accurately AI assistants mention your brand when buyers ask them questions. It covers three outcomes: being named in the answer, having your own pages cited as a source, and being described with the right facts.

The term comes from a 2023 research paper by a team at Princeton and IIT Delhi, “GEO: Generative Engine Optimization”. Since then it has become the common name for this work, alongside AEO (answer engine optimization) and LLMO (LLM optimization).

How is GEO different from SEO?

SEO competes for a position in a list of links, and the prize is a click. GEO competes for a place inside a written answer, and often there is no click at all. The buyer reads what the assistant says, makes a decision, and types your brand into Google later (or never).

The foundations overlap. A site that is fast, crawlable and well organized does better in both. But a good Google rank does not prove AI visibility: about half of the sites cited by Google’s AI Overviews are outside Google’s top 10, and for ChatGPT the overlap is smaller still.

How do AI assistants learn about your brand?

An assistant knows your brand through two separate channels, and you influence each one in a different way.

Training memory. Each model learns from a large snapshot of the web, frozen at a cutoff date. What it absorbs about you is blurry: it may know you exist and roughly what you sell, if you appeared often and consistently enough. You cannot edit this memory. You can only be present and consistent across reputable sites before the next model is trained, and that takes months.

Live search (grounding). For most buying questions, the assistant runs web searches before it answers, reads the results, and builds the answer from them. This is where most GEO work pays off, and changes can show up within weeks.

Well-known brands get named from memory even without a page to cite. Smaller brands depend almost entirely on live search, which means they depend on being found on the pages the assistant reads.

How does an AI assistant build an answer?

Here is what happens between a buyer’s question and a cited answer. You can win or lose at every step.

1. The question becomes several searches. The assistant rewrites a question like “best home insurance for a family renting in Austin” into short background searches, often called fan-out queries. ChatGPT runs about four per question, and they change from one run to the next.
2. Each assistant searches its own index. ChatGPT draws partly on Bing and partly on OpenAI’s own fast-growing index. Claude’s web search uses Brave Search. AI Overviews and Gemini use Google. Perplexity runs its own. If your page does not come up in that engine’s search, nothing else on the page matters.
3. Pages are cut into passages. The system does not read whole pages. It splits them into short chunks, often 40 to 300 words, and matches each chunk to the search by meaning, not by exact keywords.
4. The best passages are picked. The system ranks the chunks by how directly they answer the search, and gives weight to recent, specific and trusted sources.
5. The model writes the answer and attaches citations. This is the moment your brand is named, named wrongly, or left out.

The search step carries most of the weight. In one study, whether a page appeared in ChatGPT’s background searches explained 32% of the difference in citations between pages. All on-page features together (length, structure, schema) explained 3%. Position also matters: a page in first place in that search was cited 58% of the time, and a page in tenth place 14%.

What gets a page cited by AI?

Six tactics have evidence behind them. Here they are in the order we apply them with clients.

1. Let AI crawlers read your pages

AI crawlers mostly read the raw HTML your server sends and do not run JavaScript. If your prices, coverage details or FAQ answers only appear after a script runs, many assistants see an empty page. Pages that need JavaScript lose about 11% of their Gemini citations.

Check three things:

• robots.txt. OpenAI runs separate crawlers: GPTBot collects training data, OAI-SearchBot builds ChatGPT search, and ChatGPT-User fetches pages when a user asks. Blocking GPTBot does not remove you from ChatGPT search, but blocking OAI-SearchBot does. Sites that block Google-Extended get cited less in AI Overviews and Gemini.
• Your CDN. Since July 2025, Cloudflare blocks AI crawlers by default on new sites. Check that the crawlers you want are not blocked without your knowledge.
• The raw page. Fetch each key page with curl. The main answer has to be in the result.

2. Give each buyer question its own page or section

AI assistants cite the page whose title and paragraphs answer the search most directly. In a study of 158,000 AI answers, a title that matched the question brought about 14% more citations, and a matching paragraph about 15% more. An analysis of 50,553 ChatGPT answers by AirOps found that narrow pages get more citations than broad guides.

Progressive’s Answers section is a good model: one article per question, with the question as the title (“Does car insurance cover water damage?”).

How to write one of these pages:

• Use the buyer’s question, in their words, as the title or heading.
• Put a two- or three-sentence answer directly below it that makes sense on its own.
• Give each close sub-question its own heading.
• Put your brand, product, category and location names in the title and headings.
• Keep the page between 500 and 2,000 words, with short paragraphs.

3. Get mentioned on the sites AI already cites

Most citations go to other people’s sites. Across the studies we reviewed, about 75% of citations point to third-party pages (reviews, comparisons, media, forums) and the brand’s own site gets about 3%. A brand that appears on 11 or more different cited pages is named in 60% of answers. On one page, it is named in 2%.

“Best X” lists, comparison pages and review roundups hold about 90% of third-party brand mentions. Start with the profiles you control (Google Business Profile, BBB, Trustpilot, partner directories), because you can fix those in weeks. Editorial lists and reviews take months of PR and outreach, and some are paid. For AI Overviews, Gemini and Perplexity, YouTube videos and LinkedIn articles also get cited often.

4. Describe your brand the same way everywhere

If your website calls you one thing and review sites call you another, the assistant gets several stories and a smaller brand can drop out of the answer. Use the same name, the same category words and the same core facts on your site, social profiles, review platforms and partner listings.

5. Show dates and prices, and keep them current

For time-sensitive questions such as prices or “best X in 2026”, assistants prefer recent pages. A page is cited about half as often once it is around 40 days old. Show an “Updated on” date, update these pages every one to three months with real changes, and write prices as text on the page (not only inside images).

6. Use short paragraphs under clear headings

Splitting long text into short paragraphs roughly doubled how often pages appeared in AI search results in one set of tests, and headings and lists helped too. The effect is small next to the first five tactics, so treat it as hygiene.

Which GEO tactics have no proven effect?

A lot of GEO advice repeats a few early findings that later tests did not confirm. When we compared pages on the same site, these tactics did not bring more AI citations:

One more caution: no published study has yet proved that GEO work increases traffic or leads. Promise and report visibility, and connect it to revenue with your own conversion data.

How do you measure GEO?

You measure GEO by asking the assistants the questions your buyers ask, many times, and recording the answers. Your web analytics will not show it, because most AI influence never produces a click. Click-through from Google is around 40%; from ChatGPT it is closer to 1%.

AI answers are noisy. About 65% of cited sources change from one day to the next, so a single run tells you very little. Our checklist for numbers you can defend:

• Run each prompt 7 to 8 times per engine and use the average.
• Report on two- to four-week periods, not single days.
• Report each engine separately. ChatGPT, Gemini, Perplexity and AI Overviews cite very different pages.
• Separate branded from unbranded prompts. “Is [brand] good for X?” names the brand in 94 to 100% of answers. “Best provider for X” names it in 12 to 52%.
• Separate “named” from “cited”. ChatGPT names brands without citing any page about them in 46% of cases. A mention with no cited page is not a stable win.
• Use real questions where you can. In our research on real chat data versus synthetic prompts, the wording of a prompt changed which sources the assistants cited. Prompts invented from keywords can track a different world from the one your buyers live in.

Prompt tracking is one of four data sources. The other three are server logs (which AI crawlers visit which pages, and how many hit errors), AI referral traffic in your analytics, and asking new customers where they heard about you. None of them gives the full picture alone.

Change is measurable. In our own pre-registered experiment, we published content built for AI citation and fired the same 19 prompts every day for 33 days. ChatGPT’s citation rate for the test site went from 0% to 42% by week five, and Perplexity’s went from 0% to 27%.

Is GEO only about visibility?

Visibility is the top of a longer funnel. At Waniwani we split GEO into three layers:

1. GEO visibility. Is your brand named and cited for the questions buyers ask, compared with your competitors?
2. Performance GEO. When a buyer is ready to act, does the assistant suggest your own AI app inside ChatGPT or Claude?
3. Conversion. Once the buyer is in your app, do they get a price, a quote or a booked call?

For insurers and other companies with quote-based products, the first layer gets you into the conversation and the next two decide whether the assistant can sell you. A carrier that only works on visibility gets described. A carrier with an AI app can also quote. We cover the second layer in our guide to ChatGPT app directory optimization.

The layers also feed each other. The questions buyers ask inside your AI app are the best prompts to track for visibility, because they are real.

Where do you start with GEO?

A first month of GEO work, in the order that unblocks the most:

1. Collect 30 to 50 real buyer questions from your sales calls, support tickets, site search and Search Console. Tag each one by intent: learn, compare, choose, or check a fact.
2. Add 8 to 10 fact questions whose correct answer is on your own site (licensed states, products, phone number, office addresses). They prove whether assistants can read your pages.
3. Track the questions across engines for two to four weeks before you change anything, so you have a baseline.
4. Fix crawler access first: robots.txt, CDN settings, and content that only loads with JavaScript.
5. Read the cited sources for each question. If competitors’ pages are cited, write a better page for that question. If lists, reviews and media are cited, the decision happens on other sites, so work on getting listed there.
6. Re-measure every two weeks and compare changed questions with unchanged ones.

Frequently asked questions

What does GEO stand for?

GEO stands for generative engine optimization. It is the practice of improving how often and how accurately AI assistants such as ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews name and cite your brand in their answers.

Is GEO the same as AEO?

They overlap. AEO (answer engine optimization) started with Google’s featured snippets and voice assistants and focuses on being the direct answer to a question. GEO is broader: it covers how generative AI assistants compare, recommend and describe brands across a whole conversation.

Does GEO replace SEO?

No. Many AI assistants pull their sources from search engines, so a crawlable, well-organized site helps both. But a good Google rank does not guarantee AI visibility, so you need to track AI answers separately.

How long does GEO take to show results?

Changes that affect live search (crawler access, new question pages, refreshed content) can show up within two to six weeks. Third-party mentions on editorial lists and review sites take months. Changes to what a model knows from training only arrive with the next model version.

Does schema markup help with AI citations?

Studies that compare pages on the same site found no citation gain from schema markup or FAQ blocks. Keep them for traditional SEO and for readers, but do not count on them for AI visibility.

Should I block AI crawlers?

Block only what you mean to block. Blocking a training crawler such as GPTBot does not remove you from ChatGPT search, but blocking OAI-SearchBot or ChatGPT-User does. Check each crawler separately and check your CDN’s default settings.

How do I know if ChatGPT mentions my brand?

Ask it the questions your buyers ask, several times each, and record whether your brand is named, whether your pages are cited, and whether the facts are correct. Do the same on each engine separately, because they read different sources.

Is GEO worth it for insurers and banks?

Yes, and more so than for most industries, because buyers already ask AI assistants to compare coverage, prices and providers. Every change to a page or listing still has to pass compliance review, so plan for that in the timeline.