Free tool · No account · No pitch

Free GEO audit: is your site ready to be cited by AI answers?

A GEO audit measures whether answer engines can reach, understand, trust, and cite your page.ChatGPT, Perplexity, Gemini, Claude, and Google AI Overviews don't rank pages — they quote passages. Enter any URL plus your email: our research-backed engine scores it 0–100 across 5 dimensions in about 30 seconds, shows the top fixes, and emails you a human-verified full report within 3 business days.

Last updated: September 24, 2026 · By Growthegy Team

Download the 42-point checklist

Same checks the engine runs, as a Markdown file you can keep, print, or hand to a developer. Enter your email and it downloads instantly — plus one follow-up email, nothing else.

Preview all 42 checks

Crawl access (AI bots must reach you)

robots.txt exists and does not block OAI-SearchBot, Claude-SearchBot, PerplexityBot, Googlebot, Applebot
Training opt-out is a separate decision: GPTBot / ClaudeBot / CCBot and the Google-Extended token (blocking training never affects citations)
Never use a blanket "User-agent: * / Disallow: /" on a public site
XML sitemap exists, is reachable at /sitemap.xml, and is referenced from robots.txt (≤50k URLs and ≤50MB per file)
Curated llms.txt at domain root (10–30 key links, H1 title, descriptions) — Google ignores it (2026-06-15); ship for non-Google optionality, 0 citation weight
Do NOT create llms.txt, chunk content, or rewrite phrasing "for AI" expecting citations — Google explicitly rejects all three (AI optimization guide, myths section)

Structure (passages, not pages, get cited)

Exactly one H1 per page; sequential H2 > H3 hierarchy, no skipped levels
Question-form headings that mirror real buyer queries
2–4 sentence paragraphs; each section self-contained (paste-alone test)
Key claim in the first sentence of every section and in the first 30% of the page
Real HTML tables with labeled thead (never image-tables); lists for steps and features
5–8 real user-phrased FAQs, above the fold or TOC-linked

Citability (give engines a reason to quote you)

40–60 word self-contained answer block at the top of each key section
"X is…" definition sentence within the first 60 words
At least one dated, sourced statistic per section (never invent numbers)
At least two named-expert quotes per flagship page (name + title)
3–6 outbound citations to primary sources (.gov, .edu, papers, official docs)
One uniquely attributable number per piece (survey, benchmark, product data)

Authority and trust (E-E-A-T + Who/How/Why)

Named author with title and linked bio on every article (never anonymous)
Visible publication date and last-updated date; refresh money pages quarterly
Who/How/Why passes: who made it (byline), how (process/AI-use disclosure where readers would ask), why (to help, not to chase clicks)
Corrections path visible (updates logged, errors fixed openly); HTTPS everywhere
Contact page + about page + privacy policy linked site-wide (recourse path)
Reply to reviews within 24–72 hours; keep NAP identical everywhere
Wikipedia/Wikidata presence where eligible (within the rules); Reddit/YouTube/review footprint

Schema and technical (currency matters)

Valid JSON-LD with @type on every block (Article+author, QAPage for genuine Q&A, Product/Offer, LocalBusiness, VideoObject)
https://schema.org context, absolute URLs, ISO-8601 dates, zero placeholder text
Never ship HowTo, SpecialAnnouncement, CourseInfo, ClaimReview, EstimatedSalary, LearningVideo, or VehicleListing (rich results retired 2023-2026)
FAQPage kept for semantics only — Google retired FAQ rich results for ALL sites 2026-05-07; no removal order, but expect no SERP benefit
JSON-LD in server-rendered HTML (JS-injected markup faces delayed processing, Google Dec 2025)
Title 30–60 characters and distinct from the meta description; meta description 120–160 characters with a CTA
Alt text on all informative images (10–125 chars); width/height set; lazy-load below the fold; WebP/AVIF with fallback
Server-rendered key content (no JS-only text); LCP ≤2.5s; no X-Robots noindex surprises

Agent readiness (report-only, no score weight)

Primary content present without JavaScript (GPTBot/PerplexityBot-class fetchers read raw HTML)
Page HTML under 2MB so key content + JSON-LD sit inside Googlebot’s fetch window
Absolute self-referencing canonical; intentional meta robots (never accidental noindex)
Open Graph (title/image) + twitter:card for unfurls; hreflang with x-default where multi-language
AI discovery endpoints present (.well-known/ai.txt, /ai/summary.json) — optionality, not ranking
Interactive elements expose accessible names/roles (agent + screen-reader baseline)

Falsifiability (verify every fix)

Each fix states how you would know it failed (re-crawl, re-audit, or cited-answer check)
Re-run the audit after every change batch — scores move only when checks flip
Track cited-answer presence per engine (distributions over runs, never single-run promises)

Run these checks inside your own AI tools

Copy the universal prompt into any assistant. Honest note: each tool wires prompts differently — the guides below match how each one actually works. No tool can promise citations; these prompts audit readiness, same as this page.

Claude Code — Skills directory

.claude/skills/geo-audit/SKILL.md (project) or ~/.claude/skills/geo-audit/SKILL.md (personal)

Create the folder and save the prompt below as SKILL.md inside it.
Restart Claude Code (or run /init) so it picks up the skill.
Invoke it by asking: "audit this page for AI citations" with a URL.

Cursor — Rules file

.cursor/rules/geo-audit.mdc

Create the file with the frontmatter shown, then paste the prompt below the frontmatter.
Cursor attaches it automatically to matching work; or @-mention the rule in chat.

Codex (OpenAI) — AGENTS.md

AGENTS.md in your repo root

Append the prompt below under a "GEO audit" heading in AGENTS.md.
Codex follows repo instructions automatically on every task.

ChatGPT — Custom instructions / Custom GPT

Settings → Personalization → Custom instructions (≤1500 chars per field)

Paste the short prompt below into custom instructions.
For repeatable audits, create a Custom GPT and paste the full prompt as its instructions.

Grok (X) — Pinned prompt (no skills system)

No install location — Grok has no skills directory

Paste the prompt below at the start of each session, then give it a URL.
For repeat use, keep it in a note and re-paste; results vary run to run, so audit twice.

Windsurf / Kiro — Rules / steering files

.windsurf/rules/geo-audit.md or .kiro/steering/geo-audit.md

Save the prompt below at the path for your tool.
Reference it in chat when you want an audit run.

How does the audit work?

The engine fetches your page the way an answer engine does, then runs deterministic checks grouped into five weighted dimensions. Every check is shown in your result — nothing is a black box:

Dimension	Weight	What it checks
Crawl access	20	robots.txt allows AI-search bots; curated llms.txt; AI discovery endpoints
Structure	20	One H1, sequential headings, tables, lists, FAQ blocks, self-contained passages
Citability	25	Real statistics, named quotes, primary-source citations, definitions, unique data
Authority	20	Named author, publication dates, page depth, trust signals
Schema	15	Valid JSON-LD, titles, meta descriptions, alt text, language declaration

Scores of 70+ with zero critical issues pass our ship-gate — a threshold calibrated to published audit research where pages above it earn citations at roughly triple the rate of pages below it. Critical issues (blocked crawlers, noindex headers, concealed manipulative text) fail the gate on their own, whatever the score.

Why should you trust this audit?

Four reasons, each verifiable:

Open methodology. The rubric mirrors published, peer-reviewed research — the Princeton GEO study (KDD 2024), which found quotation addition lifting visibility ~41% and statistics ~33–40% — plus open-source audit engines whose scoring weights are public.
No invented inputs. The engine only measures what it can fetch. Rewrite suggestions use your real facts or explicitly flagged slots — we never fabricate statistics or quotes.
No guaranteed outcomes.AI answers vary run to run (measured similarity between runs sits around 0.34–0.42), so honest measurement reports distributions, not promises. Anyone promising citations is selling certainty that doesn't exist.
Stated limits.JavaScript-only content, logged-in pages, and video without transcripts can't be fully assessed by automated fetching — your result says so when that applies.

What Google says you do NOT need

Google's AI optimization guideis explicit: from Search's perspective, optimizing for generative AI search is optimizing for the search experience — still SEO, not a separate discipline. AI Overviews and AI Mode are grounded in the same ranking and quality systems as classic Search, layered with retrieval-grounded generation and query fan-out. Google explicitly rejects five popular claims:

llms.txt as a citation lever — Google Search ignores llms.txt entirely. We report it for non-Google optionality with zero citation weight.
Chunking content into small pieces for AI — not required; self-contained passages win on readability, not compliance.
Rewriting phrasing or keywords "for AI" — synonym understanding is sufficient.
Chasing inauthentic mentions across blogs, forums, and videos.
Over-investing in structured data specifically for AI features — schema still matters for eligibility, not as an AI shortcut.

Eligibility floor: a page must be indexed, eligible for snippet display, and included via the Search Console generative-AI control (include is the default) to appear in any Google AI feature. There is no separate "AI index". Everything this audit checks is SEO fundamentals applied through that lens.

How to verify every fix (falsifiability)

Every finding states how you would know it failed: re-crawl the URL, re-run the audit, and watch the specific check flip. Track cited-answer presence per engine across repeated runs — AI answers vary run to run, so honest measurement reports distributions, never single-run promises. Third-party figures in our copy (for example, citation lifts or freshness windows) are labeled with source and date, or left out.

Primary sources this audit follows

Google AI optimization guide (docs, updated 2026-07-10)
Google creating helpful content / Who-How-Why
Google third-party SEO tools guidance (2026-06-05)
Princeton GEO study (KDD 2024)

Frequently asked questions

What is a GEO audit?

A GEO (Generative Engine Optimization) audit measures whether answer engines — ChatGPT, Perplexity, Gemini, Claude, and Google AI Overviews — can reach your page, understand it, trust it, and cite it. Ours scores five dimensions (crawl access, structure, citability, authority, schema) out of 100 and lists exact fixes.

Will this audit get my site cited by ChatGPT or Perplexity?

No audit can promise that, and you should distrust any tool that does. AI answers vary run to run and engines change constantly. What the audit does is make citation possible and measurable: it finds the concrete gaps — blocked crawlers, missing schema, thin passages, anonymous authorship — that keep engines from quoting you.

Why do you ask for my email before showing results?

Two reasons, both stated upfront: the full human-verified report (rewrite demo, platform notes, measurement starter) is delivered by email within 3 business days, and one follow-up email carries the report summary. No newsletter trap, no list-bombing, deletion on request any time.

How does the scoring work — can I verify it?

Yes. Every check is deterministic and listed in your result: robots.txt bot access, llms.txt presence, heading hierarchy, tables and lists, FAQ blocks, statistics and quotes, outbound citations, author and date metadata, JSON-LD validity, title and meta lengths, brand-name consistency, and freshness. The rubric mirrors published research (Princeton GEO, KDD 2024) and open-source audit engines. Nothing is a black box.

Which AI engines does this cover?

ChatGPT Search, Perplexity, Google AI Overviews and AI Mode, Gemini, Claude, and Grok — plus agentic browsers that read pages directly. Each has different retrieval habits (for example, Perplexity re-crawls in hours while ChatGPT leans on training data), so the report includes per-platform notes where they matter.

How is this different from an SEO audit?

SEO audits optimize for ranked lists and clicks. A GEO audit optimizes for being quoted inside synthesized answers: self-contained passages, machine-readable structure, named authorship, and verifiable facts. The two complement each other — technical SEO is the foundation GEO stands on.