Is your business found by AI? A founder's guide to AEO & GEO in 2026

AI visibility is the measurable degree to which a business gets cited by large language models — ChatGPT, Gemini, Perplexity, Claude — when those LLMs answer buyer-intent questions in its category. It is a discrete signal from Google rankings: a site can hold position #1 in classic SERPs and be invisible inside the AI answer that increasingly arrives before the user ever scrolls. This guide explains the two disciplines that govern it (AEO and GEO), the empirical evidence that they work, the precise mechanism by which grounded LLMs choose what to cite, and how to instrument your own domain so you stop guessing and start measuring.

The data: AI search is the new front door

Five numbers worth knowing before you build:

• +37% citation lift from structured statistics — Princeton's 2023 paper GEO: Generative Engine Optimization (Aggarwal et al., NeurIPS 2024) ran 10,000+ controlled tests across major generative engines. Inserting precise statistics raised the probability of being cited by 37%.
• +40% from authoritative citations. The same paper showed adding citations to authoritative sources lifted the probability of being chosen as a source by 40%. LLMs preferentially cite content that itself cites.
• +28-30% from direct quotations. Quoting experts or primary sources directly increased visibility by 28-30% in the same tests. Block quotes are an under-deployed AEO lever.
• Search market share is shifting. Gartner's 2024 forecast projects traditional search volume will drop ~25% by 2026 as users migrate to AI assistants — and BrightEdge data already shows AI Overviews touching 60%+ of informational queries.
• The cost asymmetry is brutal. Each ChatGPT or Gemini answer that surfaces a competitor instead of you represents a free placement they earned and you didn't. There is no paid-ads equivalent inside generative answers — yet.

AEO vs GEO vs SEO — three disciplines, one stack

The acronyms have proliferated because the search interface has split. Here's the clean mental model:

• SEO — Search Engine Optimization. Optimizes for the blue-link ranking on Google's 10 organic positions. Levers: backlinks, on-page keywords, Core Web Vitals, internal linking.
• AEO — Answer Engine Optimization. Optimizes for being extracted into the answer surface: AI Overviews, featured snippets, voice answers, ChatGPT replies. Levers: FAQPage schema, entity consistency, Q&A formatting, unambiguous one-sentence facts.
• GEO — Generative Engine Optimization. The empirical sibling of AEO, formalized in the 2023 Princeton paper. Optimizes for being chosen as a citation by the LLM's retrieval step. Levers: structured statistics, authoritative citations, direct quotations, source diversity.

How a grounded LLM actually chooses what to cite

When a user asks Gemini "best AI voice agent for dental clinics Vancouver," the model runs a real-time Google Search internally (this is what "grounded" means), pulls 5–15 candidate pages into context, then synthesises an answer with inline citations. Three observations from running thousands of grounded queries:

1. Citation probability is not proportional to ranking. A page that ranks #6 with crisp FAQ schema often outpicks a page that ranks #1 with a wall of marketing copy. Format wins over position.
2. Entity ambiguity kills citation. If your business name resolves to multiple entities (a salon called "Aesthetic" in Toronto and an "Aesthetic Tree" in Vancouver), the LLM tends to drop both rather than guess. Disambiguation via Organization schema and consistent NAP across the open web is the highest-ROI fix most local businesses skip.
3. Citation density compounds. Pages that already cite authoritative sources get cited more often themselves. The Princeton +40% finding rhymes with a deeper truth: LLMs treat "cites" as a trust signal. Refusing to cite anyone reads to the model as isolation.

A worked example: running an audit on your own site

Here is the exact methodology we use across the Zealous portfolio (and that we put behind a one-click form at frankyao.com/ai-visibility):

1. List 3–5 seed terms — what should buyers find you for? "custom home builder Vancouver," "AEO audit Vancouver," etc. Stay bottom-of-funnel.
2. Expand into buyer questions — "best custom home builder Vancouver reviews," "laneway home builder Vancouver," etc. 6–8 queries is the sweet spot.
3. Run each through a grounded LLM — Gemini 2.5 Flash with the google_search tool is the cheapest production-grade option and returns citation URLs in the response metadata.
4. Count cited / missed — is your domain in the citation list? List the competitor domains that are.
5. Diagnose the gap — for the queries where you missed, what does the cited competitor have? Usually it's a named-entity FAQ block, a granular service-area page, a long-tail comparison post, or an organization schema with samesAs links to social.

The five fixes that move the needle (ranked by effort:impact)

1. FAQPage schema on every BOFU page. 5+ named questions, real answers (not marketing fluff), full JSON-LD. The single highest-ROI lever; takes one afternoon for a 10-page site.
2. Author bylines with E-E-A-T credentials. Real name, real credentials, schema.org/Person markup, "reviewed by" for medical/legal/financial. Closes the trust gap for YMYL queries.
3. Statistics with citations. Every claim → a stat → a linked authoritative source. This is the Princeton finding made concrete: 5+ cited statistics per BOFU page lift visibility ~37%.
4. Organization schema with sameAs links to LinkedIn, Crunchbase, Wikipedia (if applicable), GMB. Disambiguates the entity for the LLM's retrieval step.
5. An llms.txt at the domain root. An emerging convention (analogous to robots.txt for crawlers) that tells LLMs which pages to prioritize. Cheap to ship, asymmetric upside.

Common mistakes that nuke citation rate

• Marketing-voice copy with no extractable facts. LLMs can't cite adjectives.
• Inconsistent business name / address across the site, GBP, social, citations. Every variant fragments the entity in the LLM's knowledge graph.
• FAQ sections without FAQPage schema. The questions are great; the LLM can't parse them as Q&A pairs without the markup.
• Single-source content (only your own marketing, no outbound cites). Reads as isolation to the LLM's trust step.
• Auto-generated content with no human author byline. AI Overviews are increasingly demoting AI-generated content with no demonstrable human expertise behind it.

What changed in 2026 — the new ranking is the new front door

For 25 years, getting found meant ranking on Google. The mechanism was stable: build backlinks, write good copy, ship fast pages. The 10 blue links were the front door, and SEO told you exactly how to get there. That mechanism still works — for the queries that still flow through classic search.

But the queries that matter most for SMBs — informational lead captures, "best X near me" comparisons, "how do I choose Y" pre-purchase research — are migrating fast. Gartner projects a 25% drop in classic search volume by 2026. BrightEdge sees AI Overviews on the majority of informational queries already. The new front door is the generated answer. If you're not in it, you may as well not exist.

The good news: it's still early. Most SMBs have no AI visibility instrumentation, no AEO/GEO program, no FAQPage schema, no llms.txt. The gap between "invisible" and "cited" is often a single afternoon of structured-data work. The first step is finding out where you stand.

Frequently asked questions

Written by Frank Yao, AI automation architect & AEO/GEO strategist based in Vancouver, BC. For a custom AEO/GEO program on your site, book a free consult. For the free tool, try the AI Visibility Audit.