What does this GEO test check?
GEO (Generative Engine Optimization, or AI search optimisation) means making a site readable, understandable and citable by answer engines: ChatGPT, Claude, Perplexity, Google AI Mode. This test measures the technical prerequisites, the ones observable in the code without interpreting the assistants' answers.
- AI crawler access: your robots.txt is parsed per RFC 9309 for sixteen named user agents (GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, Claude-User, Claude-SearchBot, PerplexityBot, Perplexity-User, Google-Extended, GoogleOther, CCBot, Amazonbot, Applebot-Extended, meta-externalagent, Bytespider, MistralAI-User).
- Extractability: the raw HTML of the homepage is read without executing JavaScript, as most AI fetchers do: title, description, single h1, h2, canonical, language, text share, empty-page signals.
- Structured data: JSON-LD blocks are validated, their types listed, and the identity entity (Organization, LocalBusiness…) is checked for name, url, sameAs and a legal identifier.
- llms.txt: presence and structure of the file, and whether an llms-full.txt exists.
- Freshness and identity: machine-readable dates, Open Graph tags, hreflang when the site is multilingual.
- Response time (TTFB) measured on the homepage, with deliberately generous thresholds.
Every check is deterministic: no language model takes part in the analysis, the same site gives the same result, and every point is something we implement for our clients.
How to run a GEO audit: the six dimensions of the SEOForge score (v1, September 2026)
The score is our own - it follows no industry standard, as none exists yet. It is published here in full so it can be verified and cited. Each dimension is scored from 0 to 100 then weighted; a dimension that cannot be assessed is excluded and the remaining weights are re-normalised.
| Dimension | Weight | Why this weight |
|---|---|---|
| AI crawler access | 35 | Absolute prerequisite: a blocked crawler reads nothing. Answer crawlers count for 85 points of the dimension, training crawlers for 15, because blocking the latter is a legitimate choice. |
| Content extractability | 20 | What a text extractor sees without JavaScript is what the model will be able to cite. |
| Structured data | 15 | JSON-LD is the most reliable way to tell a machine who you are, what you sell and at what price. |
| llms.txt file | 10 | Emerging convention (llmstxt.org): useful and cheap, but not yet universally read. Moderate weight, absence lightly penalised. |
| Freshness and identity | 10 | A dated, identifiable source is retained more easily than an anonymous, timeless one. |
| Response time (TTFB) | 10 | A very slow server makes crawlers give up; beyond that, speed is a Core Web Vitals matter, measured by the speed test. |
Caps (hard blockers)
Some findings cap the final score whatever the other results:
- homepage in error or unreachable: maximum score 0
- noindex on the homepage: maximum score 30
- every AI answer crawler blocked: maximum score 40
Grades
A from 90, B from 80, C from 65, D from 50, F below.
Known limits
- The test reads the homepage and four well-known files (robots.txt, llms.txt, llms-full.txt, sitemap.xml), at most 6 requests and 15 seconds. It does not crawl the site.
- It does not measure your actual citations in assistants' answers: nobody can do that reliably today, and we would rather say so.
- A block at CDN or firewall level (Cloudflare, for instance) is not visible in robots.txt: only server-log analysis, included in the GEO Audit, reveals it.
- TTFB is measured from our servers in France, on a single request: an order of magnitude, not a field measurement.
Results are cached for one hour per URL, with no public result page. Our crawler identifies itself as SEOForge-GEO-Test and honours your robots.txt. For Core Web Vitals, use the speed test.
Every dimension is detailed, with the engines' documentation and the fixes in the code, in the complete guide to AI search optimisation (GEO)
This methodology was applied to 1,000 European SME sites in September 2026: read the study and download the data
Each dimension has its detailed guide (most are in French):
- Crawlers IA et robots.txt : les seize robots, un par un (FR)
- Cloudflare and AI crawlers: proxy, Bot Fight Mode, WAF
- Vérifier l'accès de ChatGPT et des assistants à votre site (FR)
- Le contenu sans JavaScript pour les moteurs IA (FR)
- Schema Organization et identité d'entité (FR)
- llms.txt : spécification, faits documentés, écriture (FR)
- Measuring your visibility in AI assistant answers
- Missing from ChatGPT: the step-by-step diagnosis
- How to get cited by ChatGPT and other AI assistants
- Choisir une agence GEO : les critères à vérifier (FR)
This free test vs the GEO Audit
The test stops where the work begins: observing is free, understanding and fixing happens in your context.
| Free GEO test | GEO Audit - €490 excl. VAT | |
|---|---|---|
| AI crawler access | robots.txt, 16 user agents | robots.txt + CDN + firewall, verified in your server logs (which crawlers really get through, how often) |
| Scope | Homepage and well-known files | Your site's priority pages: services, offers, expert content |
| Structured data | Types present and identity signals | Complete entity schema, disambiguation against namesakes, machine-readable offers and prices |
| Content | HTML extractability | Answer-first structure, citable FAQs, dating policy, llms.txt generated from your pages |
| Off-site | Not covered | Official profiles, directories and third-party sources that models cross-check |
| Deliverable | Score and generic actions | Action plan ranked by impact / effort, with optional implementation in your code |
Frequently asked questions
What exactly does this test check?
Six technical dimensions observable in your site's code: access for sixteen AI crawlers in robots.txt, content extractability without JavaScript, structured data and entity identity, the llms.txt file, freshness and identity signals (dates, Open Graph, hreflang) and server response time. The sitemap is checked for information. Every check is deterministic and documented in the methodology above.
Does this test measure whether ChatGPT cites me?
No. This test does not measure your actual citations in AI assistants' answers, and no tool does so reliably today. It measures the prerequisites: if a crawler cannot read your site or does not understand who you are, a citation is impossible. A good score does not guarantee a citation; a bad score makes it very unlikely.
Is a perfect score possible?
Yes, and our own site gets it, since the test checks exactly what we implemented on seoforge.fr: robots.txt explicit bot by bot, llms.txt generated from the routes, Organization schema with company identifiers and official profiles, server-rendered content. Every point is publicly verifiable.
Does blocking training crawlers lower the score?
Very little. Training crawlers (GPTBot, ClaudeBot, CCBot…) weigh 15 points out of the 100 of the "access" dimension, and blocking them is reported for information, not as an error: it is a legitimate editorial choice. Blocking answer crawlers (OAI-SearchBot, ChatGPT-User, PerplexityBot…) is, however, what removes you from the answers.
What do you keep from a test?
The tested URL and the final URL, the date, the score, the per-dimension scores and the codes of the points found (never excerpts of your site), the test duration and, for 30 days, a truncated, irreversible fingerprint of your IP address to detect abuse. This data serves to run and improve the tool and is deleted after 24 months. If you request the report by e-mail, the test is attached to your request so we know what we are talking about.
Is this a complete GEO audit?
No. This free GEO audit reads your home page and four well-known files, at most six requests and fifteen seconds, and scores the prerequisites of a citation. The paid GEO Audit covers the six dimensions on your priority pages, reads your server logs over several weeks, measures your citations before and after, and ends with the fixes implemented in your code.
Free or paid GEO audit: what is the difference?
The free one answers one question in fifteen seconds: can answer engines read, identify and cite your site? The paid one answers the next: what to fix, in which order, and who does it. The table above details what each covers; the GEO Audit price is published on the service page.
Want to know what assistants say about your market today? That is what the 20-minute call included with the GEO Audit is for.