Generative engine optimization: how brands get cited by AI

Ask ChatGPT to recommend vendors in your category and check whether you are in the answer. That is the problem generative engine optimization exists to solve: getting your brand named and cited when AI engines like ChatGPT, Gemini, Perplexity, and Google’s AI Overviews answer your buyers’ questions. SEO wins you a position on a results page, and GEO wins you a place inside the answer.

That definition is written to be lifted whole, which is the first lesson. I have spent ten years in SEO and growth, and I was putting AI into marketing workflows before it was standard practice. Today this is my job as Lead of Innovation and AI at Tandem Interactive, a digital agency in Fort Lauderdale. What follows is the playbook I run.

How AI engines pick sources

Every major engine answers in one of two modes. Parametric mode is the model talking from training data: whatever it absorbed about your brand from years of crawled reviews, forums, lists, and press. Retrieval mode is the model running live searches and writing an answer from the top results, with citations. Most commercial queries, the ones with money attached, trigger retrieval.

The retrieval pipeline is the part you can influence this quarter. The engine rewrites the user’s question into several sub-queries and pulls candidate pages from a search index. As I write this, ChatGPT uses Bing plus its own crawling, Gemini and AI Overviews sit on Google, and Perplexity runs its own index. The engine then reads passages, not whole pages, quotes whichever passage answers a sub-question most cleanly, and links a handful of sources.

You still have to be findable in a search index, because GEO sits on top of SEO. The unit of competition also shrank from the page to the passage. A 2,000-word page with the answer buried in paragraph fourteen loses the citation to a page that answers in the first fifty words, even when the long page outranks it.

Engines also carry priors that predate anything you launch this quarter. When a model already “knows” your brand from training data, retrieval confirms what it knows. That is why the off-site half of GEO, the roundups and communities that feed both retrieval and the next training run, takes months to appear in retrieval or later training data.

GEO vs SEO vs AEO

The acronyms are multiplying faster than the underlying practice is changing, so here is the short version:

AEO, answer engine optimization, came first. Chasing featured snippets taught the industry to write liftable passages years before LLMs arrived, and GEO extends the same instinct to engines that synthesize and cite. Do not spend much time on the labels, since a well-structured page can win on all three at once.

The on-page patterns that get quoted

I build measurement systems that sample AI answers at volume, so I read a lot of what engines lift. The patterns are consistent across every category I have tested.

Lead with the answer. Put a two-to-three sentence definition directly under the heading, then elaborate below it. If your page opens with three paragraphs of preamble, the citation goes to whoever answered first.

Ask the question, then answer it. Question-shaped H2s map neatly onto the sub-queries engines generate. Follow each with an answer-first paragraph and put the nuance after.

Number your steps. For how-to intent, engines reproduce numbered lists nearly verbatim. Give them a clean list with one action per step.

Publish numbers worth citing. Engines gravitate toward specific, attributed figures, so never invent one and always name the source. Publishing your own original data is stronger still, because original numbers make other pages cite you.

Add a real FAQ. Ask short questions and answer each one in two or three direct sentences. FAQ blocks map almost one to one onto the fan-out queries engines run, which is why this essay ends with one.

Use tables for comparisons. Structured comparisons come through into generated answers close to intact, formatting included.

Keep every passage self-contained. Retrieval systems chunk your page, and any section that depends on the three sections above it dies in transit. Write each block so it survives being lifted alone.

In my testing across client categories, the change with the highest return is the first one, a plain, liftable definition at the top of the page. It is also the cheapest one on the list.

Entity signals engines can resolve

Engines cite entities they can resolve, not strings. The string “Acme” is ambiguous. Acme the software company, with a consistent description and a web of corroborating profiles, is something a model can name with confidence. What that takes:

• Schema that resolves. Organization and Person markup with a stable @id, repeated identically across the site, carrying sameAs links to every profile you control.
• Consistency everywhere. Same name, same one-liner, same facts on your site and on every profile that mentions you. Contradictions cost confidence.
• Third-party corroboration. For “best X” questions, assistants synthesize from roundups and review sites far more than from anyone’s homepage. Ask the engines your buyers’ questions, read what they cite, then earn placements on those exact pages.
• A knowledge graph path. Where you can justify it, Wikidata and the graphs downstream of it give engines a canonical node to hang your name on.

I run this playbook on myself: a schema.org Person graph with a stable @id and reconciled sameAs links, plus a patient path toward Wikidata. The mechanics work the same for a brand, minus the part where I share a name with a retired ballplayer.

How to measure whether GEO is working

This is where most GEO advice falls apart, and where I spend most of my working time. Engines are nondeterministic by design. Ask the same question five times and you will get different brands in the answer. A screenshot of one response is a sample of one, and a dangerous thing to put in a client deck.

The method I run in production is to fix a prompt set that matches real buyer questions, then ask each prompt several times for each engine in each run. Record mention and citation rates, wrap a Wilson confidence interval around every rate, and compare intervals only. If last month’s interval overlaps this month’s, nothing provably moved, and the report should say so. I wrote up the full statistical method in Measuring AI visibility with confidence intervals.

Alongside the sampled rates, watch the corroborating signals in GA4: referral sessions from chatgpt.com and perplexity.ai, plus branded search trends moving in the same direction. None of these is precise on its own, but together they triangulate.

What to stop doing

Stop treating GEO as a replacement for SEO. Retrieval engines pull from search indexes, so crawlability and rankings still gate everything. The fundamentals still pay, and the same technical hygiene that lifts organic traffic is what makes a site quotable in the first place.

Stop reporting screenshots. A single AI answer is one sample from a distribution. If a tool or an agency hands you an “AI visibility score” with no sample size and no interval, the number has no basis you can check.

Stop publishing pages with no author and no point of view. Engines favor sources that look like sources: named authors with real credentials and something original to say. Faceless filler is exactly what these systems keep getting better at ignoring.

Stop blocking AI crawlers by accident. Plenty of sites blocked GPTBot in 2023 on principle and forgot about it. That is a legitimate call for a publisher. For a brand that wants to be recommended, it works against the goal, so decide either way on purpose.

The GEO checklist

Ten steps, in the order I’d run them:

1. Write down 20 questions your buyers ask an assistant, from “what is” through “best tool for.”
2. Baseline yourself: ask each question several times in every engine you care about, and log each brand mentioned and each URL cited.
3. Add a liftable two-to-three sentence definition to the top of every page that matters.
4. Rewrite key H2s as questions with answer-first paragraphs under them.
5. Convert how-to content into numbered steps and comparison content into tables.
6. Publish one asset with original numbers or an original position. That is what other pages will cite.
7. Ship Organization and Person schema with a stable @id and sameAs links, then fix every inconsistency across your profiles.
8. Earn placements on the third-party pages your baseline shows the engines already citing.
9. Audit robots.txt and decide your AI crawler policy deliberately.
10. Re-run the sampled measurement monthly, and only celebrate when the confidence intervals separate.

If you’d rather hand this list to someone else, my capabilities page covers what that looks like.

FAQ

How do I show up in ChatGPT answers?

The short answer is to be retrievable first and liftable second. ChatGPT’s browsing pulls from a search index, so pages that rank for the question get read first, and pages that answer within the first fifty words get quoted. Build the third-party footprint (reviews, roundups, communities) that makes the model comfortable naming you.

What’s the difference between GEO, AEO, and SEO?

SEO targets rankings on results pages, and AEO targets answer boxes like featured snippets. GEO targets citations and brand mentions inside generated answers from engines like ChatGPT and Perplexity. Underneath the labels, it is one discipline: structured, trustworthy content that machines can parse and quote.

How do I optimize for Google AI Overviews?

Roughly the way you’d chase a featured snippet, with more attention to passage structure. AI Overviews draw heavily from pages that already rank for the query and its related sub-questions, so standard technical SEO still gates entry. Answer-first formatting and self-contained sections do the rest of the lifting.

How long does GEO take to work?

On-page changes can register within weeks on retrieval engines, since they re-fetch pages constantly. Entity work and third-party placements move on a scale of months, and anything riding on model training data moves slower still. Sample monthly and be suspicious of anyone promising day-seven results.

Should I block AI crawlers like GPTBot?

If you monetize the content itself, maybe, and that is a business decision worth making carefully. If you are a brand that wants assistants to recommend you, blocking the crawlers that fetch your pages works against the goal. Audit robots.txt and make that choice deliberately.