A GEO audit inventories technical eligibility, visible content and evidence, entity clarity, and measured presence across answer engines. It cannot prove that an engine will index, rank, recommend, or cite a page. For Google AI Overviews and AI Mode, normal Search indexing and snippet eligibility apply; no special AI schema or file is required.

How is a GEO audit different from an SEO audit?

The durable work overlaps heavily with SEO: crawlability, indexability, helpful evidence, links, reputation, and measurement. A GEO-labeled audit additionally separates operator-specific crawler policy and samples answer-engine presence. Optional llms.txt and internal passage scores are workflow diagnostics, not ranking signals.

How long does a full GEO audit take?

A technical snapshot can run in minutes, but an evidence-backed visibility audit also needs provider coverage, repeated prompts, ranking data, source review, and human verification. CiteFuel automates a 26-check snapshot; it does not replace the longer authority, content, and reputation work.

Which AI crawlers should I check?

Separate training, retrieval, user-triggered, and control tokens. For example, GPTBot and OAI-SearchBot serve different OpenAI purposes. Google-Extended controls certain Gemini Apps/Vertex AI uses and has no effect on Google Search or AI Overviews rankings; Googlebot governs Search crawling.

Is there a free GEO audit tool?

Yes. CiteFuel's AI crawlability checker runs 26 checks across all five layers — AI crawler access, llms.txt, structured data, passage citability, and technical foundations — and returns a scored gap list. The audit is free and takes under two minutes. Start at citefuel.com/ai-crawlability-checker.

Checklist · Updated June 2026

GEO Audit Checklist

A free, step-by-step checklist for auditing your site's visibility to AI answer engines — ChatGPT, Claude, Perplexity, and Google AI Overviews. Five layers, 26 checks, with what to look for, why it matters, and exactly how to fix each gap.

Most sites lose AI citation not because their content is bad, but because one of five configuration layers is broken. A GEO audit works through those layers in priority order — access first, content last — because no amount of passage rewriting helps if a crawler can't reach the page. This checklist mirrors CiteFuel's free 26-check audit and gives you the manual version you can run against any site.

Layer 1: AI Crawler Access

A blocked retrieval crawler cannot fetch the affected page directly. That is important technical evidence, but it does not prove the brand can never appear through a third-party index or another source.

1.1 Check your robots.txt for blocking rules

What to check: Fetch https://yourdomain.com/robots.txt and look for User-agent: * with Disallow: /, or any rule that explicitly blocks AI tokens: GPTBot, OAI-SearchBot, ClaudeBot, Claude-SearchBot, PerplexityBot, Google-Extended.

Why it matters: A legacy wildcard disallow written before AI crawlers existed blocks every one of them by fallback. An allowlist that names only Googlebot does the same.

How to fix: Choose a policy for each documented purpose, then add explicit rules only for the crawlers or product tokens you intend to allow. Example policy — review before deploying:

User-agent: GPTBot
Allow: /

User-agent: OAI-SearchBot
Allow: /

User-agent: ClaudeBot
Allow: /

User-agent: Claude-SearchBot
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: Google-Extended
Allow: /

Google-Extended is a control token, not a separate request user-agent. Allowing it does not improve Google Search or AI Overviews rankings; blocking it does not remove a site from Search.

1.2 Check your WAF and CDN bot settings

What to check: If you use Cloudflare (Bot Fight Mode), another WAF, or a bot manager, verify that AI user-agent strings are not being challenged or blocked before robots.txt is even read.

Why it matters: A WAF can return 403 to an AI crawler while your robots.txt correctly says "allowed" — the most common silent failure in GEO audits.

How to fix: Add AI crawler user-agents to your WAF's allowed-bot list, or disable Bot Fight Mode for verified AI crawlers. Cloudflare lets you create firewall rules that match known AI user-agent substrings and bypass the challenge.

1.3 Verify crawler access with real user-agent probes

What to check: Use a tool that sends real HTTP requests with each AI user-agent string and checks the response code — not just robots.txt parsing. CiteFuel's checker does this for 14 AI agents.

Layer 2: llms.txt

llms.txt is a voluntary proposal for publishing a curated Markdown site map. Only around 6.9% of the sampled domains published one. That is an adoption statistic, not evidence of a visibility advantage. Google Search explicitly ignores the file.

2.1 Check for presence and location

What to check: Fetch https://yourdomain.com/llms.txt. It should return HTTP 200 with Content-Type: text/plain — not a 404, not an HTML page, not a redirect.

2.2 Validate the format

What to check: The file must open with an H1 site name (# Site Name), followed immediately by a blockquote description (> One sentence about the site.), then one or more ## sections containing - [Title](URL): description entries. Each part is required.

How to fix: The state of AI crawlability research shows that malformed llms.txt is nearly as common as missing ones. Use the free validator at CiteFuel to check all 11 spec requirements, or see the full format guide.

2.3 Check URL consistency with your sitemap

What to check: Every URL listed in llms.txt should exist in your sitemap and return 200. Dead links make the optional document less useful to compatible services, but CiteFuel has not established that a malformed file causes a ranking or citation penalty.

Layer 3: Structured Data / JSON-LD

Structured data can help search engines understand visible page content and support eligibility for supported rich-result features. Google requires no special schema for AI features, and valid markup does not guarantee ranking or citation.

3.1 Organization and WebSite schema on every page

What to check: Where applicable, your root domain can serve Organization and WebSite JSON-LD with stable @id anchors (e.g., https://yourdomain.com/#organization). Every property must match visible, current information; absence alone is not proof that an engine cannot identify the organization.

How to fix: Add both schemas to your site's base layout so they appear on every page. Add sameAs only for verified identity profiles you control or can substantiate. CiteFuel treats sameAs as informational; it does not prove entity recognition or earn score credit.

3.2 Keep structured data consistent with visible content

What to check: Mark up only content that readers can see, use the most applicable supported type, and remove hidden, irrelevant, duplicated, or misleading properties. Google no longer shows FAQ rich results and HowTo rich results are deprecated; neither is an AI-ranking tactic.

3.3 Article / BlogPosting schema on content pages

What to check: Content pages should declare Article or BlogPosting with datePublished, author, and publisher when those facts are visible and truthful. These fields improve metadata completeness; they do not manufacture expertise or guarantee recency-based ranking.

3.4 Validate schema parses cleanly

How to fix: Run your pages through Google's Rich Results Test or the Schema Markup Validator. JSON-LD that contains syntax errors is silently skipped — no warning, just no signal.

Layer 4: Citable Passage Structure

Even with all configuration correct, content that is structurally hostile to extraction will not be cited. AI retrieval pulls individual passages, not full pages — so each paragraph either stands alone or it doesn't.

4.1 Definition-first paragraphs

What to check: Your most important pages should answer the question in the first sentence of each section — not build toward an answer over three paragraphs. Ask: if only this paragraph were quoted, would it make sense?

How to fix: Rewrite key passages in definition-first order. Name the subject in the first sentence when that serves the reader. CiteFuel uses 40–150 words as an editorial heuristic, not a universal extraction window; preserve enough context and evidence for the claim.

4.2 Entity naming in each passage

What to check: Passages that rely on "it" and "they" can become ambiguous when excerpted. Name the entity where doing so improves clarity; this is an editorial aid, not a citation guarantee.

4.3 Avoid generic preamble

What to check: Phrases like "In today's digital landscape…" and "It's more important than ever to…" signal filler to AI verification systems. They dilute the signal-to-noise ratio of your content and compete with your actual citable claims.

4.4 Internal link concentration

What to check: Your most important topic pages should receive internal links from multiple pages on the same domain. Isolated pages lack the internal authority signal that helps AI systems identify your highest-confidence content on each topic.

Layer 5: Technical Foundations

AI retrieval inherits search infrastructure. A broken technical foundation depresses everything above it.

5.1 HTTPS with no mixed-content errors

What to check: All pages should serve over HTTPS with a valid certificate and no mixed-content warnings. HTTP pages are increasingly deprioritized or skipped entirely by AI crawlers.

5.2 Canonical tags

What to check: Each page should have a self-referencing canonical tag (<link rel="canonical" href="https://yourdomain.com/page" />) or correct cross-page canonicals for paginated or duplicate content. Without them, AI crawlers may index the wrong URL variant or treat duplicates as separate low-authority pages.

5.3 Sitemap discoverability

What to check: Your sitemap should be declared in robots.txt (Sitemap: https://yourdomain.com/sitemap.xml), return 200, and include all canonical URLs you want indexed. AI crawlers use sitemaps as a discovery layer.

5.4 Core Web Vitals

What to check: LCP under 2.5s, CLS under 0.1, INP under 200ms. AI systems that actively crawl for live retrieval (Perplexity, Bing Copilot, SearchGPT) deprioritize slow pages. The biggest wins are usually LCP — image optimization, defer non-critical JS, avoid render-blocking resources.

Run the full 26-check audit automatically

This manual checklist covers the same ground as CiteFuel's automated audit, but takes considerably longer and misses WAF-level probes you can't replicate by hand. The free 26-check audit runs all five layers — including real HTTP probes with AI user-agent strings — and returns a scored, severity-ranked gap list in under two minutes. For the research behind the scoring model, see the State of AI Crawlability 2026 report.

Whatever approach you use: fix genuine access and indexability failures first, then reconcile claims and visible evidence, then improve useful content and independent authority. Treat llms.txt as optional file hygiene and structured data as accurate description — never as shortcuts to ranking or citation.