GEO audit checklist: 50-point assessment for AI visibility
A 50-point GEO audit checklist scores a site's readiness to be cited by AI search engines across eight pillars — crawlability, content structure, entity coverage, schema, citation worthiness, internal links, SEO foundations, and tracking. Each item is worth one point; the total score maps to a Severity band that prioritizes rewrite work.
TL;DR. Use this 50-point checklist to score any page or site for Generative Engine Optimization (GEO). Award one point per item that fully passes (no half points). Sum the score, map it to the Severity table, and tackle the lowest-scoring pillars first. The whole audit takes 60-90 minutes per page once you have the tooling in place.
How to use this checklist
- Pick the page (or template) you want to audit.
- Walk through all 50 items in order. Award one point only when an item is fully satisfied.
- Total your score out of 50 and convert to a percentage.
- Map the percentage to the Severity band below to prioritize fixes.
| Score (out of 50) | Percentage | Severity | Action |
|---|---|---|---|
| 45-50 | 90-100% | Low | Keep monitoring; minor polish only |
| 38-44 | 76-89% | Medium | Targeted fixes on weak pillars |
| 30-37 | 60-75% | High | Schedule rewrite within the sprint |
| 0-29 | < 60% | Critical | Rewrite or unpublish before AI indexing |
Pillar 1 — Crawlability & technical readiness (1-8)
- [ ] 1. robots.txt does not block major AI crawlers (GPTBot, PerplexityBot, ClaudeBot, Google-Extended).
- [ ] 2. No firewall, CDN, or bot-protection rule returns 403/429 to AI user agents.
- [ ] 3. Page renders fully without client-side JavaScript (server-side or static HTML).
- [ ] 4. Canonical tag points to a single, indexable URL (no canonical loops or chains).
- [ ] 5. XML sitemap includes the page and is referenced from robots.txt.
- [ ] 6. HTTPS is enforced; mixed-content warnings are zero.
- [ ] 7. Largest Contentful Paint < 2.5s on mobile (Core Web Vitals pass).
- [ ] 8. llms.txt exists or is intentionally omitted with a documented decision (impact is currently unproven, so either choice is acceptable).
Pillar 2 — Content structure & answer-first format (9-16)
- [ ] 9. A single, descriptive
matches the page title.
- [ ] 10. An AI summary block (callout or blockquote) appears within the first 200 words.
- [ ] 11. A TL;DR paragraph (2-3 sentences) is snippet-ready and self-contained.
- [ ] 12. Heading hierarchy is logical (h2 → h3), with no skipped levels.
- [ ] 13. Each h2 section answers one extractable sub-question.
- [ ] 14. Lists, tables, and short paragraphs dominate over long prose blocks.
- [ ] 15. A dedicated FAQ section (3-5 Q/A pairs) appears near the bottom.
- [ ] 16. Word count fits the content-type range (definition 600-1400, guide 1200-3500, tutorial 1500-4000, comparison 800-2000, framework 1000-2500, checklist 500-1500).
Pillar 3 — Entity coverage & disambiguation (17-22)
- [ ] 17. The page's primary entity is named explicitly in the first paragraph.
- [ ] 18. All key sub-entities (people, products, standards, acronyms) are mentioned by their canonical name at least once.
- [ ] 19. Acronyms are expanded on first use (e.g., "Generative Engine Optimization (GEO)").
- [ ] 20. A canonical_concept_id (kebab-case) exists in frontmatter and is unique across the corpus.
- [ ] 21. entities[] and aliases[] arrays in frontmatter cover all reasonable LLM phrasings.
- [ ] 22. Brand or organization name is consistent across the page, schema, and external profiles (Wikipedia/Wikidata when applicable).
Pillar 4 — Schema & structured data (23-28)
- [ ] 23. Valid JSON-LD is present and passes Schema.org Validator.
- [ ] 24. Article, TechArticle, or HowTo schema matches the page's content_type.
- [ ] 25. FAQPage schema mirrors the on-page FAQ (questions and answers match exactly).
- [ ] 26. Organization and WebSite schema are referenced site-wide (with @id linking).
- [ ] 27. Author is marked up with Person schema and links to a real bio page.
- [ ] 28. BreadcrumbList schema reflects the URL path and on-page breadcrumbs.
Pillar 5 — Citation worthiness & evidence (29-34)
- [ ] 29. Every strong claim links to a primary source (official docs, peer-reviewed paper, or first-party data).
- [ ] 30. Statistics include the source, year, and sample size when relevant.
- [ ] 31. No outdated facts (anything older than review_cycle_days is re-verified).
- [ ] 32. First-party data, original research, or expert quotes appear at least once.
- [ ] 33. last_reviewed_at and version fields in frontmatter are current.
- [ ] 34. No unverified AI-generated claims ("hallucinations") survive editing.
Pillar 6 — Internal linking & topical authority (35-39)
- [ ] 35. At least one link points to the section's hub or pillar page.
- [ ] 36. Two to three contextual links point to sibling articles in the same cluster.
- [ ] 37. Anchor text is descriptive (not "click here" or raw URLs).
- [ ] 38. related_articles[] in frontmatter lists up to five relevant slugs.
- [ ] 39. No broken internal links (4xx/5xx) anywhere on the page.
Pillar 7 — SEO foundations (40-44)
- [ ] 40.
tag is 50-60 characters and includes the focus keyword. - [ ] 41. Meta description is 120-160 characters and reads naturally.
- [ ] 42. Open Graph and Twitter Card metadata are complete.
- [ ] 43. Image alt text is descriptive and entity-aware.
- [ ] 44. URL slug is short, lowercase, hyphen-separated, and matches frontmatter.slug.
Pillar 8 — Tracking & monitoring (45-50)
- [ ] 45. Analytics segments referral traffic from chatgpt.com, perplexity.ai, and gemini.google.com.
- [ ] 46. A monthly prompt set (20-30 buyer-style questions) is logged across ChatGPT, Perplexity, and Gemini.
- [ ] 47. Citations are tracked over time — appearance, position, and competitor share.
- [ ] 48. Server logs are reviewed for AI crawler activity (GPTBot, PerplexityBot, ClaudeBot).
- [ ] 49. A re-audit is scheduled at published_at + review_cycle_days.
- [ ] 50. The audit score, severity, and date are stored in your CMS or content database.
Scoring example
A new long-form guide scores 41/50 (82%). Pillars 1 (crawlability) and 4 (schema) hit full marks, but Pillar 5 loses three points (claims without sources) and Pillar 8 loses three points (no AI-referrer tracking). Severity is Medium — the content is solid, but a focused fix on evidence and tracking should ship before the next sprint.
FAQ
Q: How is a GEO audit different from a traditional SEO audit?
A traditional SEO audit grades crawlability, on-page tags, and link signals against a ranking algorithm. A GEO audit additionally scores citation worthiness, entity clarity, and structured data because AI engines synthesize answers and cite sources rather than rank ten blue links.
Q: How often should I run this 50-point audit?
Run it on every new article before publishing, and re-run it at the cadence set by review_cycle_days in frontmatter (90 days is a safe default for fast-moving AI search topics).
Q: Do I really need llms.txt to pass the audit?
No. Independent studies — including a Search Engine Land tracking experiment across ten sites and SE Ranking's analysis of 300,000 domains — found no measurable lift in AI citations from llms.txt. Item 8 awards the point if the file exists or is intentionally omitted with a documented rationale.
Q: What score is "good enough" to publish?
Aim for 38/50 (76%) or higher before publishing. Below that threshold, the page is unlikely to be cited consistently by ChatGPT, Perplexity, Claude, or Google AI Overviews.
Q: Can I automate this audit?
Partly. Crawlability, schema validation, and Core Web Vitals are easy to automate (Screaming Frog, Schema Validator, PageSpeed Insights). Entity coverage, citation worthiness, and AI-referrer tracking still need human review or specialized GEO tools.
Related Articles
AI Platform Citation Mix Strategy
Portfolio framework for AI platform citation mix: allocate GEO effort across ChatGPT, Perplexity, Gemini, Claude, and Copilot by source bias.
AI readability score: how to measure machine comprehension of your pages
AI readability scoring: which classic readability metrics still matter for LLMs, plus the structural and semantic signals AI parsers reward.
AI Search Citation Types: How AI Attributes Sources
Reference for AI search citation types — inline, footnote, source card, attributed quote, implicit — with platform differences and how to optimize.