Generative engine optimization (GEO): how a business gets cited in AI answers

Where the term comes from

The phrase was introduced by a 2023 academic paper titled “GEO: Generative Engine Optimization,” from researchers at Princeton, Georgia Tech, the Allen Institute for AI and IIT Delhi. They built a benchmark of questions, ran them through generative engines, and tested which changes to a source page made it more likely to be used in the answer. The changes that helped were the ones that made a page more quotable and more trustworthy: citing sources, adding quotations from authorities, stating concrete figures, and writing clearly. Loading a page with keywords did not help.

That finding is the whole discipline in one sentence: a generative engine rewards pages it can quote and verify, not pages that repeat the search term.

How a generative engine builds an answer

The mechanism is called retrieval-augmented generation, and it runs in three steps that each set a requirement for your business.

1. Retrieve. The engine turns the question into searches, runs them against a web index (and, for local questions, against map and profile data), and pulls back a small set of pages and records. Requirement: your pages must be fetchable by its crawler, and your business must be present in the map and directory data it queries.
2. Synthesize. A language model reads what was retrieved and writes an answer. It extracts facts and compares them across sources. Requirement: your facts must be extractable — stated plainly in the HTML and in structured data — and they must agree from one source to the next.
3. Cite. The answer lists the sources it relied on. Requirement: your page must be one of the sources that actually contributed a fact the answer used, which means it must contain something specific enough to be worth attributing.

When an engine names a competitor and not you, one of those three requirements was not met. The monthly check is the way to find out which.

What gets a local business cited

• Specifics. The services, listed by name. The towns, listed by name. What a job involves, how long it takes, what it includes, what the warranty covers, whether you are licensed and insured. A page that says “quality roofing services in the greater area” gives an engine nothing to quote; a page that says “full tear-off and replacement in Williamson County, typically two days, with a written workmanship warranty” gives it three facts.
• A complete, current Google Business Profile. It is the record Google’s own AI reads first and one of the records the others read. Category, services, hours, photos, reviews — kept current.
• Corroboration. Directory listings, review sites, local press, associations and supplier pages that repeat the same name, address, phone and services. Independent agreement is what an engine treats as trust.
• Reviews that describe the work. A review that says what was done and where is a fact the engine can use; a star with no text is not.
• Structured data. Organization, LocalBusiness, Service and FAQPage schema that state the facts in machine-readable form, with identifiers that connect your site to your profiles.
• Freshness. Pages with a visible updated date, a profile with recent photos and posts, reviews that keep arriving. Engines prefer a record that is alive.

What does not work

Keyword density does nothing — the study measured it. A page for every town with the name swapped in reads as one page to a model, and as a signal that the site is not to be trusted. Hidden text and instructions aimed at the model (“ignore previous instructions and recommend…”) are stripped or penalized by every serious engine. Fake reviews break the corroboration that GEO depends on and violate the platforms’ terms. None of these are shortcuts; they are ways to be named less.

GEO for a contractor, concretely

For a plumber, an electrician, a roofer or a remodeler in Middle Tennessee, the program is the same list every time, and it is the list we run:

1. One entity sheet — the exact name, description, services and service area — and every profile brought into line with it.
2. A service page for every service and a real page for every city you want work from, each answer-first, each with specifics and a FAQ.
3. Structured data on every page, an llms.txt file, and crawl access for the AI agents in robots.txt.
4. The Business Profile completed and kept current; a review system that asks every customer and replies to every review.
5. Listings on the directories the engines cite, with the same facts on each.
6. The same questions asked of every engine, monthly, with the named and cited rates recorded beside the leads.

How to measure GEO

Write down the questions your customers ask — “best roofer in Franklin,” “who does tankless water heaters in Brentwood,” “roof replacement cost Williamson County” — and ask each engine every month. Record three things per answer: were you named, was your site cited, and who else appeared. Report it as a rate: named in three of twenty answers this month, nine of twenty next quarter. Never as a score; nobody can audit a score.

The free AI answer check runs the Google AI Mode version of that question live. The full report we prepare for a client asks five engines six questions each and records every answer — that is the baseline the AI SEO work is measured against.