What Is Generative Engine Optimization?

Generative engine optimization (GEO) is the work of making AI answer engines describe your brand correctly: ChatGPT, Google AI Overviews and AI Mode, Perplexity, Gemini and Claude. A buyer asks one of them "what's the best invoicing tool for a 20-person agency?" and gets a paragraph with three names in it. If yours isn't one of them, you lost that buyer before they saw your site. If yours is there with last year's price, you may have lost them anyway.

You never see that answer. There is no ranking report for it, often no click, and no referral line in your analytics. The answer is generated fresh, it can change between two runs of the same question, and it may come from a web search or from what the model memorized in training.

This lesson defines GEO, separates it from SEO honestly, shows where each of the six engines gets its facts, and says plainly what GEO can and cannot promise. Most of what is sold under the GEO label has not survived testing. Some of it has. By the end of this module you will know which is which.

What generative engine optimization means

An answer engine is any system that responds to a question with a written answer instead of a list of links. Generative engine optimization is the practice of making sure those systems can reach your pages, read your facts and repeat them correctly. You will also see the term answer engine optimization (AEO). Some people use AEO for the older work of winning featured snippets and voice answers, and GEO for chat-style AI. In practice both labels describe the same job, and this course treats them as synonyms.

GEO has three outcomes worth measuring:

• Mentioned. Does the answer name your brand at all, for the questions your buyers ask?
• Correct. When it names you, are the price, features, location, founding date and leadership right?
• Cited. Does the answer link to your site as a source, so a reader can click through and check?

Notice what is missing: position. There is no position one in a generated answer. The same question asked twice can return different brands in a different order. Google's Search Console report for its AI features shows impressions only, with no clicks, no click-through rate and no position, and Google's guidance on third-party SEO tools (June 2026) states that "third-party tools don't have access to our internal ranking data." Anyone selling you an "AI ranking" is selling a number with nothing underneath it.

So treat GEO as a fixing job, not a tracking job. You find what an engine gets wrong or cannot reach, you fix the cause on your side, and you check again.

GEO vs SEO: what carries over and what doesn't

GEO is not SEO with a new name. It is also not a separate discipline that makes SEO obsolete. Some SEO work carries straight over, some carries over weakly, and some does not apply at all.

What carries over: eligibility and crawlability. An engine can only use a page its crawler could fetch and store. For Google the rule is explicit. To be shown as a supporting link in AI Overviews or AI Mode, Google says "a page must be indexed and eligible to be shown in Google Search with a snippet," and "there are no additional technical requirements." A page with noindex or nosnippet is out of AI Overviews for the same reason it is out of the snippet. Every crawlability fix you ever made for Google still counts.

What carries over weakly: rank. Ranking well in Google helps you get cited in AI Overviews, but less than most guides claim. Ahrefs found 37.9% of URLs cited in AI Overviews also appeared in the first 10 results (March 2026), with roughly a third coming from positions 11 to 100 and a third from beyond 100. Treat organic rank as a moderate, decaying proxy, never a guarantee.

What doesn't carry over:

• One crawler. SEO mostly means Googlebot. GEO means a set of crawlers, each with its own robots.txt token and its own consequence. OAI-SearchBot decides whether you can appear in ChatGPT search. PerplexityBot feeds Perplexity's index. Claude-SearchBot serves Claude's search. Your Googlebot rules say nothing about any of them.
• JavaScript rendering. Googlebot runs JavaScript. Vercel's crawler study (December 2024) found that ChatGPT's and Claude's crawlers fetch JavaScript files but do not execute them, so anything that only appears after scripts run is invisible to them. A site that ranks fine on Google can look empty to an AI crawler.
• Keyword variants. Google's AI optimization guide (July 2026) says you don't have to worry about "long-tail" keywords or capturing every variation of a query for its generative features. Engines rewrite the question before they search, so matching the exact prompt is not the game.
• The click as the unit. Google AI Overviews send no referral signal at all. An answer can name you, shape the buyer's shortlist and send nobody to your site.
• Answers with no page behind them. SEO never had to deal with this. Many AI answers come from what the model learned in training, with no web search. No page you publish this quarter changes those answers.

The six engines 99Visibility measures, and where each gets its facts

Older GEO guides, including an earlier version of this course, split engines into two camps: "parametric" engines like ChatGPT that answer from training data, and "RAG" engines like Perplexity that search the web. That split is wrong. Every one of these engines can answer from memory, and most of them can search. On most of them the choice is made per query, by the model, the product or the user.

The useful labels describe the answer, not the engine:

• Retrieval-off: the model answers from what it learned in training. Its knowledge stops at a cutoff date.
• Retrieval-on: the system searches, pulls pages into the model's context, and writes the answer from them. These answers usually show links or numbered sources.
• Mixed: the engine decides per query whether to search.

Two consequences follow. First, any single "AI visibility" number, ours included, averages a mix of retrieval-on and retrieval-off answers. That is why 99Visibility reports every engine separately, marks every answer as searched or from memory, and why the AI visibility score lesson spells out how the columns combine. Second, the same brand can be right on one engine and wrong on another for completely different reasons. A wrong price in a retrieval-off answer and a wrong price in a cited answer have different causes and different fixes. The next lesson, Where AI Answers Come From, takes that apart stage by stage.

What GEO can honestly promise, and what it can't

That rules out a lot of promises. It does not leave you with nothing. Honest GEO work can deliver four things:

• Open the door. Make sure each AI retrieval crawler is allowed in robots.txt, is not blocked by your firewall or CDN, finds your content in the raw HTML without running JavaScript, and is not shut out by noindex or snippet controls. These gates are documented by the engines themselves. Failing one is a total, silent loss on that engine.
• Make your facts agree with themselves. If your pricing page, your about page, an old blog post and your footer state your price three different ways, an engine that reads them has no way to know which is right. Making every fact consistent on your own site is fully in your control, and it is the plausible mechanism behind most "AI gets our pricing wrong" complaints.
• Get corroborated. Correct the facts on third-party pages that describe you: review sites, directories, comparison articles. An engine that searches will often cite them instead of you.
• Measure honestly. Ask the same questions on a schedule, record whether each answer searched, and track mentions, accuracy and citations per engine.

And it cannot promise these:

• A citation, a position or a percentage lift on any engine.
• A change to retrieval-off answers this quarter. Those come from training data frozen at a cutoff.
• Visibility in proportion to effort. Well-known brands are cited far more often, and that advantage is not something markup can buy.

The course roadmap

The course runs in six modules. Every lesson ends with a win: something you have done, not something you have read.

1. How AI Answers Get Made (this module). What GEO is, the four stages behind every answer, and what the evidence says decides citations.
2. Measure What AI Says. Run your first AI answer audit, read your AI visibility score, and find what AI gets wrong about you.
3. Open the Door: Crawler Access. AI crawlers and robots.txt, bot by bot, firewalls that block what robots.txt allows, and rendering and snippet controls. The strongest evidence in the course lives here.
4. Fix the Facts AI Repeats. Content AI can quote, what schema still does, what E-E-A-T actually counts for, and off-site corroboration.
5. Engine Playbooks. Engine-specific detail for ChatGPT, Perplexity, Google AI Overviews, and Claude, Gemini and the rest.
6. Operate and Prove It. Competitor analysis, measuring results, and the capstone, your 30-day AI visibility fix plan.

Take them in order if you are new. If you already suspect that AI crawlers are blocked from your site, Module 3 pays off fastest.

Common mistakes

• Testing once, logged in, and calling it data. OpenAI says that if memory is enabled, ChatGPT "may use relevant saved memories when rewriting a search query." Your own history colors your results. Test in a private window or with memory off, and ask more than once.
• Buying additive tactics before checking access. An llms.txt file, FAQ schema or AI-style rewrites do nothing for you if OAI-SearchBot gets a 403 from your CDN. Check the gates first.
• Expecting this month's edits to change every answer. Retrieval-off answers move only when a new model ships. Judge your fixes on answers that show sources.
• Treating GEO as a replacement for SEO. Index and snippet eligibility on Google is still the gate for AI Overviews. Do not dismantle what already works.
• Chasing one number. A single cross-platform score hides which engine is wrong and why. Read the per-engine results first.