When AI answers,are you in it?
Search is no longer ten blue links. The GEO audit scores how ready your site is for AI engines and lists what's strong and what's missing, item by item.
Rather hand the job over? In our SEO & AI Visibility service we run this tool for you; scope and monthly price are published.
SEO servicesWhat a GEO audit checks, and how GEO differs from SEO
GEO (generative engine optimization) is the work of making a site easy for AI search and chat assistants to read and cite in their answers. It does not replace SEO; it adds a few checks on top of the same foundation.
| Question | Classic SEO | GEO |
|---|---|---|
| Written for | The results page | The AI engine writing the answer |
| Access | Googlebot and other search crawlers | AI crawlers such as GPTBot, OAI-SearchBot, PerplexityBot and ClaudeBot |
| Structure | Titles, internal links, speed | Schema, question-style headings, short definitions |
| Trust | Links and authority | Wikidata entry, organisation identity, author, cited sources |
| Measured by | Rankings and clicks | Being cited; this audit measures readiness |
53 checks across 6 categories
When you enter a domain, the scan runs in two phases. First, robots.txt, llms.txt, sitemap.xml and the homepage are requested at once. If the site responds, the second phase requests the homepage again as AI crawlers and looks up a Wikidata entry linked to your domain.
| Category | Points | Checks | What it checks | Read from |
|---|---|---|---|---|
| Citability | 22 | 11 | Question headings, a definition sentence, a 40-80 word opening, lists, tables, figures, sources, heading order, paragraph length | Homepage body |
| Content and expertise | 22 | 11 | Content depth, title, meta description, author, date, expert review, customer proof, alt text, freshness | Homepage |
| Brand and entity | 20 | 9 | Wikidata and Wikipedia, sameAs, social profiles, contact details, company identity, certificates, address | Homepage, Wikidata |
| Access and rendering | 15 | 7 | Crawler permissions, responses to crawler requests, noindex, nosnippet, server-side rendering, HTTP status, speed | robots.txt, homepage |
| Structured data (schema) | 13 | 8 | Valid JSON-LD, organisation schema and its fields, WebSite, FAQ, content, breadcrumb, review and offer schema | Homepage JSON-LD |
| Platform and technical signals | 8 | 7 | llms.txt and its format, sitemap, the sitemap line in robots.txt, Open Graph, Twitter card, canonical | llms.txt, sitemap.xml, robots.txt, homepage |
AI crawlers and robots.txt access
Crawler permission is worth 5 points, split across 18 crawlers by weight. Whether a crawler may open your homepage is worked out by merging robots.txt rules as RFC 9309 describes.
Weights are read from the fixed table in the audit engine and add up to 98. Each crawler's share is its weight over that total, times 5 points.
| Purpose | Crawlers | If blocked |
|---|---|---|
| AI search | OAI-SearchBot, Claude-SearchBot, PerplexityBot, Googlebot, Bingbot, Amazonbot, DuckAssistBot | Critical |
| User request | ChatGPT-User, Claude-User, Perplexity-User, MistralAI-User | Critical |
| Model training | GPTBot, ClaudeBot, Google-Extended, Applebot-Extended, CCBot, Meta-ExternalAgent, Bytespider | Medium |
If only training crawlers are closed, visibility in answers does not drop directly, so the finding is medium.
- All rule groups naming the crawler are merged; if it has no group of its own, the catch-all (*) group applies.
- The longest rule matching the homepage path wins, and Allow wins a tie. The * and $ characters are read as patterns.
- No robots.txt means access is treated as open; a robots.txt that cannot be read is reported as “undetermined”.
- A server or firewall can still refuse a crawler that robots.txt allows. So the homepage is requested as GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, PerplexityBot, CCBot; this check is worth 3 points.
Can ChatGPT, Gemini and Claude reach my site
Your client knows the product, not the crawler. So the results for 18 crawlers are grouped into 7 answer platforms. A crawler counts as closed for its platform if robots.txt blocks it or if a request sent as that crawler fails.
| Platform | Crawlers read |
|---|---|
| ChatGPT | GPTBot, OAI-SearchBot, ChatGPT-User |
| Perplexity | PerplexityBot, Perplexity-User |
| Gemini · AI Overviews | Google-Extended, Googlebot |
| Claude | ClaudeBot, Claude-SearchBot, Claude-User |
| Copilot | Bingbot |
| Meta AI | Meta-ExternalAgent |
| Apple Intelligence | Applebot-Extended |
- Ready: every crawler for the platform is open.
- Partly open: some crawlers are open and some are closed.
- Blocked: none of the platform's crawlers can reach the site.
- Not measured: robots.txt could not be read; the platform is not counted as blocked.
Why a Wikidata entry carries so many points
AI engines confirm a brand against knowledge bases. The audit looks for a Wikidata item whose official website (P856) is your domain. That check is worth 5 points, and a linked Wikipedia article is worth 3 more.
- With no Wikidata item, the Wikipedia check drops out of the score; the same gap is not penalised twice.
- If the Wikidata query does not answer, the check is reported as not measured, never as zero.
- Adding an item requires the brand to be covered by independent sources; the rules are on the Wikidata notability page.
Score, findings and the 30-day plan
- 1Enter the domain
The last audited domain is pre-filled, so you do not retype it.
- 2Four sources read
robots.txt, llms.txt, sitemap.xml and the homepage are requested together.
- 3Crawler requests and Wikidata
If the site responds, crawler requests and the Wikidata query go out together.
- 4Score and level
Measured checks are scaled to 100 and the level follows from the score.
- 5Findings and plan
Findings are ranked by severity, then by points lost, and spread over the weeks.
- 6Fix and re-audit
If an earlier audit exists, the change is shown in points.
| Score | Level | What it means |
|---|---|---|
| 80 and above | Strong readiness | Most measured checks pass |
| 60-79 | Good readiness | Solid base with a few gaps |
| 40-59 | Moderately ready | Points lost across several categories |
| 0-39 | Low readiness | Crawler access or basic structure is missing |
| No measurement | Not measured | The site did not respond; no zero is written |
| Severity | Checks | Examples |
|---|---|---|
| Critical | 8 | Search crawler blocked, noindex, empty HTML shell, organisation schema |
| Medium | 25 | Question headings, author, meta description, Wikidata entry |
| Opportunity | 20 | llms.txt, Open Graph, comparison table, summary box |
| Week | Which findings go in |
|---|---|
| Week 1 | Low-effort critical and medium fixes |
| Week 2 | Low-effort opportunities and medium-effort critical fixes |
| Week 3 | Remaining medium-effort fixes |
| Week 4 | High-effort content, brand and infrastructure work |
Empty weeks are skipped. Each week shows the score you reach once its fixes are done.
Client reports and share links for agencies
When the audit finishes, the side panel turns the same result into documents for your client. Every output is built from the same scorecard, so one audit always produces the same text.
- PDF report: saved through the browser's “Save as PDF” print option.
- Report text: score, platform readiness, 30-day plan and strengths, ready to paste.
- Talking points: a short summary to read out in a meeting.
- Proposal draft: scope and expected outcome, with no price written in.
- Share link: valid for 30 days and needs no login.
Who it suits, and where it stops
- Agencies that want to show a client their AI search readiness as a concrete score.
- Business owners who do not know whether their site is closed to AI crawlers.
- The audit measures readiness; it does not measure whether AI answers actually mention your brand.
- The sitemap is looked for at /sitemap.xml; a sitemap at another address shows as “not found”.
If question headings and short definitions are missing, you can rebuild the copy with the AI content generator. On multilingual sites, the wrong page getting indexed is a separate problem: when Google indexes the wrong language.
Frequently asked questions
What is a GEO audit?
A check of how ready your site is to be read and cited by AI engines such as ChatGPT, Perplexity and Gemini. It runs 53 checks across 6 categories: citability, content and expertise, brand and entity, access and rendering, structured data, and platform signals.
How long does a GEO audit take?
In the first phase four sources are read at once, and the longest wait is 20 seconds for the homepage. The crawler and Wikidata requests in the second phase take at most 10 seconds. A slow site returns its result within 30 seconds.
Do I need an llms.txt file?
No. llms.txt is still a proposed format and, with its format check, is worth 3 points of the score. If it is missing, the finding is listed at opportunity level.
What happens if I block GPTBot?
GPTBot is a training crawler and costs about 0.5 points of the robots.txt score; if only training crawlers are closed, the finding is medium. If OAI-SearchBot or ChatGPT-User is closed, the finding becomes critical, because visibility in ChatGPT answers drops directly.
What if my brand has no Wikidata entry?
The Wikidata check misses its 5 points and the Wikipedia check drops out of the score. The finding carries a high-effort label and goes in the last week of the plan; an entry needs coverage of the brand in independent sources.
Does an unreachable site score zero?
No. A check that cannot be reached drops out of the score and is never written as zero. If nothing could be measured, the score shows “Not measured”.
Can I re-run the audit after fixing things?
Yes. Audit the same domain again and the difference from the previous audit is shown in points. The number of audits per month depends on your plan.
How do I send the report to my client?
Create a share link; it is valid for 30 days and needs no login. You can also take a PDF, the report text, talking points or a proposal draft.
Sources
The official documentation behind the technical points on this page:
- [1]Google Search Central: AI features and your website
No extra requirements for AI Overviews; Search fundamentals apply.
- [2]Google Search Central: Common crawlers
Google-Extended does not affect Search inclusion or ranking.
- [3]OpenAI: Overview of OpenAI crawlers
GPTBot, OAI-SearchBot and ChatGPT-User are separate crawlers.
- [4]RFC 9309: Robots Exclusion Protocol
Merging rule groups, longest match and wildcard characters.
- [5]Wikidata: Official website property (P856)
The property that links an item to a domain.
- [6]Wikidata: Notability
The conditions an item must meet to be added.
- [7]The llms.txt proposal
Purpose and format of the llms.txt file.
Related pages
- Free SEO analysisA quick look at your site's overall technical health.
- SEO servicesHand the fixes from the audit over to us.
- ReportingAnalytics, Search Console and ad data in one report.
- AI content generatorWrite content in a question-and-answer structure.
- ChatGPT AdsChatGPT ads are sponsored placements shown below the answer in OpenAI's chat assistant, to users on the free and Go plans; ads went live in Turkey on 17.09.2026.
If AI can't find you, you're not there.
Run the GEO audit on a paid plan. The full panel is free with no time limit: brand, accounts, reports. Allowance: 10 captions a month, 2 images a month, 1 videos a month, 1 brand, 1 user.