What is GEO (Generative Engine Optimization)?

GEO covers the techniques that optimize a site to be understood, extracted and cited by generative AI models (ChatGPT, Perplexity, Gemini, Claude) as they build their answer, rather than to appear in a classic list of results.

What is the difference between GEO and AEO?

GEO aims at being cited as a source by a generative AI model inside a conversational answer. AEO aims more broadly at becoming the direct answer, including Google featured snippets and voice assistants. Both overlap heavily on the technical side.

Is classic SEO still useful in 2026?

Yes, and it remains the foundation. Generative AI models rely largely on the same indexes as classic search engines: a poorly indexed or poorly structured site has very little chance of being cited by an AI.

What is an llms.txt file?

A text file placed at the domain root that gives AI crawlers structured context about the site's key pages, much like robots.txt does for traditional search engines.

How long does it take to improve a GEO score?

Technical fixes (schema, llms.txt, SSR) can ship within a few weeks. AI citation gains usually follow with a 2-to-3-month lag, the time models need to recrawl and integrate the new content.

How do I know whether ChatGPT or Perplexity already cites me?

By manually testing a set of prompts representative of your sector on each platform in a logged-out session, or with an automated tool such as OMNIRK's AI Share of Voice, which tests several LLMs and prompts at once.

What score should I aim for to be cited by AI?

A score of 16/20 or above means every blocking signal is in place: clean indexing, green Core Web Vitals, llms.txt, documented E-E-A-T, SSR rendering and FAQPage/Speakable schemas. Below 10/20, AI citations are statistically very rare.

01Method

SEO / GEO / AEO audit: the 20/20 checklist to be found on Google AND cited by ChatGPT, Perplexity, Gemini and Claude

2026-08-09·14 min read

A complete visibility audit in 2026 covers three dimensions: SEO for classic indexing and ranking, GEO (Generative Engine Optimization) to be cited as a source by generative AI, and AEO (Answer Engine Optimization) to become the direct answer. This article gives the scoring grid out of 20 — SEO /8, GEO /6, AEO /6 — and the 50-point checklist to fill it. For twenty years, being visible meant ranking on Google. In 2026, a growing share of searches ends without a click: the user asks ChatGPT, Perplexity, Gemini or Claude and gets a synthesized answer. If your brand is not mentioned there, you do not exist in that channel, even if you rank first on Google.

Classic SEO: the foundations of a 20/20 (/8)

A poorly indexed site will never be cited by an AI, because generative engines rely largely on the same indexes and crawlers as Google. Indexing and crawl: up-to-date XML sitemap submitted in Search Console and Bing Webmaster Tools, robots.txt excluding no strategic page, no accidental noindex, no orphan pages, no redirect chains. Structure and markup: unique title and meta description per page, consistent H1 → H2 → H3 hierarchy, canonical declared everywhere, correct hreflang if multilingual. Schema.org: Organization with sameAs to social profiles and Wikidata, BreadcrumbList, Article or BlogPosting with author, datePublished and dateModified. Performance: green Core Web Vitals on mobile (LCP < 2.5 s, INP < 200 ms, CLS < 0.1), valid HTTPS with no mixed content, responsive tested. Each subsection is worth 2 points when fully met, 1 when partial, 0 otherwise.

GEO: becoming a citable source (/6)

GEO does not replace SEO, it builds on it. The goal is no longer just to be indexed but to be understood, extracted and cited while the model builds its answer. Five blocks: 1) an llms.txt file at the domain root listing your key pages and giving AI crawlers structured context — still rare, therefore differentiating; 2) entities and semantic consistency: brand, founders and products recognizable as entities (Wikidata item, consistent mentions across the web), vocabulary aligned with AI prompts and not only Google queries, self-contained definitions extractable from a single paragraph; 3) reinforced E-E-A-T: identified authors with verifiable bios, primary sources cited, complete legal notice and About page, authentic dated reviews; 4) crawler-accessible rendering: essential content server-rendered rather than client-JavaScript only, nothing key hidden behind an accordion, complete Open Graph; 5) freshness: a real last-updated date displayed and content refreshed after any factual change (prices, figures, regulation).

What it looks like in practice

Three examples from sites supported by OMNIRK. OrChair, a news outlet, shipped full SSR rendering and GEO-optimized journalism verified in real time: Search Console shows a solid average position and steadily growing impressions over the first months. BookAfterWork, a booking platform, was reviewed within Station F's Fighter program: the feedback explicitly validated GEO citation evidence as a strength, with commercial traction remaining a separate topic. Mac Mahon Avocats, a business law firm, went through GEO infrastructure work including a dedicated Wikidata item for the firm — an entity signal that directly strengthens recognition by AI models.

AEO: structuring for the direct answer (/6)

AEO targets one precise outcome: your content being reused nearly verbatim as the direct answer, in a Google featured snippet, a voice answer or the excerpt cited by a conversational agent. Four levers: 1) question-and-answer structure, with a direct 40-to-60-word answer in the first two sentences of each section, plus bullet lists and tables for comparisons, easier to extract than a dense paragraph; 2) FAQPage markup on every page with a genuine FAQ, using questions taken from real queries (People Also Ask, suggestion tools); 3) speakable markup on summary paragraphs, relevant for voice assistants and AI agents that favor short declarative excerpts; 4) HowTo markup on any step-by-step procedural content, which increases rich result eligibility.

Your score out of 20: how to read it

Classic SEO /8 — what matters most: clean indexing and Core Web Vitals. GEO /6 — llms.txt, E-E-A-T and SSR rendering. AEO /6 — Q/A structure and FAQPage / Speakable schema. Total /20. Reading: 16 to 20 means you are ready to capture traffic on all three channels; 10 to 15 means solid foundations but gaps that cost you AI citations; below 10, the site is likely invisible to generative engines even if it still performs on Google.

SEO checklist: 22 points

Indexing and crawl (blocking): 1) XML sitemap submitted and current in Search Console and Bing Webmaster; 2) robots.txt verified, no strategic page blocked; 3) no accidental noindex; 4) zero orphan pages; 5) zero cascading 301 redirect chains; 6) strategic 404s fixed or redirected. Structure and markup (important): 7) unique optimized title; 8) unique compelling meta description; 9) one H1 per page and logical H2/H3 hierarchy; 10) canonical declared everywhere; 11) hreflang configured if multilingual; 12) short readable URLs without useless parameters. Schema.org (important): 13) Organization with sameAs to social profiles and Wikidata; 14) BreadcrumbList on deep pages; 15) Article/BlogPosting with author and dates; 16) Product or Service if commercial offer. Performance (blocking): 17) LCP < 2.5 s on mobile; 18) INP < 200 ms; 19) CLS < 0.1; 20) valid HTTPS with no mixed content; 21) responsive tested on mobile, tablet, desktop; 22) image compression and lazy loading enabled.

GEO checklist: 18 points

Files and AI signals (important): 23) llms.txt present and current; 24) api.json or equivalent if structured data is exposed; 25) complete Open Graph on every page (title, description, image); 26) AI-specific meta tags where applicable. Entities and E-E-A-T (blocking): 27) Wikidata item created or in progress for the brand; 28) consistent brand name and data across the web (NAP consistency); 29) identified authors with verifiable bio and expertise; 30) complete up-to-date About page; 31) verifiable legal notice and contact details; 32) primary sources cited (studies, figures, real cases); 33) authentic, dated, attributable testimonials and reviews. Rendering and freshness (blocking): 34) essential content server-rendered; 35) no key content hidden behind a required interaction; 36) real last-updated date displayed; 37) content refreshed after factual change. Extractable content (important): 38) self-contained definitions at the start of each section; 39) vocabulary aligned with user prompts; 40) short paragraphs, one idea each.

AEO checklist: 10 points

Q/A structure (important): 41) sections phrased as a question followed by a direct 40-60 word answer; 42) bullet lists and tables for comparisons; 43) TL;DR summary at the top of long content. Dedicated schema (blocking): 44) FAQPage on every page with a genuine FAQ section; 45) FAQ questions based on real queries (People Also Ask, suggestion tools); 46) HowTo on any step-by-step procedural content; 47) speakable on key summary paragraphs. Tracking and iteration (bonus): 48) monitoring of featured snippets won and lost; 49) regular citation testing on ChatGPT, Perplexity, Gemini and Claude; 50) content adjusted based on observed AI answers. Blocking means direct immediate impact on visibility, important means significant mid-term gain, bonus means fine-tuning once the basics are in place.

Test your score for free

Rather than eyeballing your score, OMNIRK offers two free tools with no signup. GEO Score: automated technical audit across 13 weighted checks (llms.txt, schemas, SSR, AI robots, Open Graph), results in under 10 seconds. AI Citation — Share of Voice: a test across 3 LLMs × 3 sector prompts to see whether ChatGPT, Gemini and Perplexity already cite you, and which competitors appear in AI answers instead. Both are available at omnirk.com/en/free-ai-tools. If the audit reveals more work than you have time for, our SEO + GEO Full Stack packages are public at omnirk.com/en/pricing.

FAQ

What is GEO (Generative Engine Optimization)?: GEO covers the techniques that optimize a site to be understood, extracted and cited by generative AI models (ChatGPT, Perplexity, Gemini, Claude) as they build their answer, rather than to appear in a classic list of results.
What is the difference between GEO and AEO?: GEO aims at being cited as a source by a generative AI model inside a conversational answer. AEO aims more broadly at becoming the direct answer, including Google featured snippets and voice assistants. Both overlap heavily on the technical side.
Is classic SEO still useful in 2026?: Yes, and it remains the foundation. Generative AI models rely largely on the same indexes as classic search engines: a poorly indexed or poorly structured site has very little chance of being cited by an AI.
What is an llms.txt file?: A text file placed at the domain root that gives AI crawlers structured context about the site's key pages, much like robots.txt does for traditional search engines.
How long does it take to improve a GEO score?: Technical fixes (schema, llms.txt, SSR) can ship within a few weeks. AI citation gains usually follow with a 2-to-3-month lag, the time models need to recrawl and integrate the new content.
How do I know whether ChatGPT or Perplexity already cites me?: By manually testing a set of prompts representative of your sector on each platform in a logged-out session, or with an automated tool such as OMNIRK's AI Share of Voice, which tests several LLMs and prompts at once.
What score should I aim for to be cited by AI?: A score of 16/20 or above means every blocking signal is in place: clean indexing, green Core Web Vitals, llms.txt, documented E-E-A-T, SSR rendering and FAQPage/Speakable schemas. Below 10/20, AI citations are statistically very rare.