Cette page existe en français.
What the scanner checks, why it matters to AI assistants and what to change when a score is low.
· Updated on
The overall score is the weighted average of these 12 dimensions. The weights are fixed and published; every report states the engine version used.
robots.txt evaluated per RFC 9309 for 17 AI crawlers (OpenAI, Anthropic, Perplexity, Google, Apple, Mistral…), meta robots tag and X-Robots-Tag header.
Allow AI search crawlers in robots.txt while keeping your existing rules; remove noindex from public pages. The report provides the full robots.txt, ready to integrate.
XML sitemap (including indexes), lastmod freshness, canonical URL, internal linking and broken outbound links on a sample.
Declare an up-to-date sitemap in robots.txt, add a canonical URL to every page, link your key pages together and fix the broken links listed in the report.
Single H1, heading hierarchy, semantic HTML5, lists and tables, text volume and readability (Flesch for English, Kandel-Moles for French).
One H1 per page, H2 and H3 without skipped levels, main, article and section elements, lists for enumerations and short sentences.
Self-contained passages, quantified facts, sources and authority links, definitions: what AI can quote verbatim.
Write paragraphs that stand on their own, give verifiable figures with their source and define your terms ("X is…").
JSON-LD (including @graph) and microdata: entity identity, content type, WebSite, breadcrumbs, Product and reviews for shops.
Add Organization (name, url, logo, official profiles) and WebSite JSON-LD, then the right type on each page. The report generates these blocks from your site data.
Presence and quality of llms.txt in the llmstxt.org format (a proposed standard): title, summary, sections and links to key pages. It serves AI assistants and agents that read your site; Google does not state that it uses it to rank search results.
Publish /llms.txt with a title, a one-sentence summary and sections linking to your key pages. The report generates a ready-to-integrate file.
Online shops only (detected automatically): price, availability, identifiers and consistency between the page and Schema.
Mark up every product page as Product with an offer (price, currency, availability), brand, image and SKU or GTIN, matching what the page shows.
Marked-up FAQ, question headings followed by a short answer, comparison tables and numbered steps.
Answer your customers’ real questions as headings with a 10 to 80 word answer right below, and mark them up as FAQPage.
HTTPS, identified entity, legal notice, privacy, contact and about page, authors, dates and reviews.
Link your legal notice, privacy policy, contact page and about page from every page; sign and date your content.
Server response time, compression, HTML weight, caching, mobile viewport, modern image formats, blocking scripts, HSTS.
Enable compression and caching, serve images as WebP or AVIF, defer scripts and aim for a server response under 800 ms.
Meta description, Open Graph, X (Twitter) card, title consistency and linked official profiles (sameAs).
Give every page a meta description, Open Graph tags and an X card consistent with its title, and link your official profiles in JSON-LD (sameAs).
20 WCAG 2.2 A/AA rules checked on the page code (alternatives, forms, accessible names, zoom, ARIA…) and the statement required by the European Accessibility Act.
Fix the elements listed per rule in the report (HTML snippets provided) and publish the pre-filled accessibility statement it generates.
For a site that is not an online shop, the Product data dimension is excluded and its weight is redistributed over the other dimensions.
The same loop for a company site, a shop or a media site: measure, fix in order of expected gain, publish, then check.
Paste a page URL. Paid plans add AI analysis; from Pro, the real-browser test (contrast, keyboard, Core Web Vitals) and whole-site crawls.
Actions are ranked by expected score gain. The free plan shows the first 5; paid plans give the ready-to-apply fix for every failed check.
robots.txt, llms.txt, JSON-LD and accessibility statement: copy them to your server, publish them in one click with the WordPress plugin, or fetch them through the API or the MCP connector.
On paid plans, your customers’ questions are asked to 5 answer engines with web search (ChatGPT, Perplexity, Gemini, Claude, Mistral): mentions, citations and competitors cited.
Re-run the audit after each fix or turn on monitoring: the scanner re-checks the site at the chosen frequency and alerts you if the score drops.
Every report produces files built from your site’s real data and compared with what it already has. Nothing is invented: a value that cannot be found on your site is not written into the file; it is listed separately, to complete before integrating.
Keeps your file as it is, including deliberate refusals, and adds only what is missing (Sitemap line). Reopening it to blocked AI answer engines is offered separately, as a fix.
Site summary and links to key pages, in the llmstxt.org format.
Entity identity: name, URL, logo, contact and official profiles found on the site.
From the questions and answers actually present on the site.
For shops: price, currency, availability and identifiers read from the page.
Only the missing or weak headers, in your host’s format (Netlify, Apache, Nginx or Vercel), with your existing settings kept.
Vulnerability reporting contact (RFC 9116), to place in /.well-known/, if your site has none.
Pre-filled with the non-conforming WCAG criteria detected (European Accessibility Act).
Failed checks, ranked, each with its fix.
The list of delivered files, their status (add, replaces, merge, already compliant) and where each one goes.
Files produced by the scanner’s generator for a fictional site (www.example.com, a domain reserved for examples).
Current robots.txt of the site
User-agent: * Disallow: /admin/ User-agent: GPTBot Disallow: / User-agent: ClaudeBot Disallow: / User-agent: PerplexityBot Disallow: /
Generated robots.txt
User-agent: * Disallow: /admin/ User-agent: GPTBot Disallow: / User-agent: ClaudeBot Disallow: / User-agent: PerplexityBot Disallow: / # --- Added by GEO/AEO Scanner on 2026-11-01 --- Sitemap: https://www.example.com/sitemap.xml
The current file is kept in full: GPTBot and ClaudeBot (training crawlers) stay blocked, the owner’s choice. Only the missing Sitemap line is added.
Fix provided in the report
# Replace the groups of these robots in your robots.txt with this one (your private paths stay closed): User-agent: PerplexityBot Allow: / Disallow: /admin/
PerplexityBot answers users’ questions and cites its sources: blocking it prevents being cited. The fix reopens public pages to it and keeps /admin/ closed.
Generated file
# Atelier Exemple > Atelier Exemple fabrique des meubles et agencements en bois massif sur mesure. Menuiserie sur mesure en bois massif ## Key pages - [Réalisations](https://www.example.com/realisations) - [Demander un devis](https://www.example.com/devis) - [À propos](https://www.example.com/a-propos) ## FAQ - **Quel est le délai de fabrication ?** Six à huit semaines après validation du devis. ## Contact - Email : contact@example.com
Every audit produces the same structured report: viewable online, exportable as PDF and CSV, and shareable through a time-limited link.
Overall GEO score, accessibility score, AI-agent readiness score and the score of each dimension.
Failed checks ranked by expected gain, with the fix (paid plans).
20 WCAG 2.2 rules with the offending HTML snippets, and the real-browser test from the Pro plan.
robots.txt, llms.txt, JSON-LD, action plan and accessibility statement ready to integrate.
10 AI-agent readiness checks and an MCP server generated for the site.
Pages analysed, engine version, bot protections encountered and the limits of the analysis.
GEO refers to making a site readable and quotable by AI assistants such as ChatGPT, Claude, Gemini or Perplexity (Generative Engine Optimization). AEO refers to optimising for answer engines (Answer Engine Optimization): clear questions followed by short answers these assistants can reuse.
It fetches the HTML served to bots (without running JavaScript, like most AI crawlers), reads robots.txt for 17 AI crawlers, llms.txt and the sitemap, explores up to 3 key pages and scores 12 dimensions, including 20 WCAG 2.2 accessibility rules.
From 5 to 30 seconds depending on the site. With AI analysis (paid plans), allow 1 to 2 minutes. A whole-site crawl takes a few minutes depending on the number of pages.
3 analyses per month without a credit card: 12-dimension score, accessibility, AI-agent score, the first 5 actions (without fixes) and one site monitored monthly. Paid plans add fixes, AI, visibility measurement, the real browser, site crawls and the API.
Yes, on paid plans: your customers’ questions are asked to ChatGPT, Perplexity, Gemini, Claude and Mistral with web search. The report counts mentions of your brand, citations of your site and competing brands, and keeps the full answers. AI answers vary from one run to the next: this is a measurement, not a guarantee.
No. The score measures whether your site is accessible, understandable and quotable. Whether to cite a source is up to each assistant and also depends on the question asked and the other sources available.
Yes, any public page can be analysed, as a crawler would. Only pages available without signing in are read.
For a plain audit, none: we only read the public pages requested. AI analysis and visibility measurement send the analysed content and your questions to the chosen models through OpenRouter; the real-browser test loads the page at Browserless.
Copy the fixes and files from the report, publish them in one click with the WordPress plugin, or fetch them through the REST API or the MCP connector for Claude Code, Codex and Cursor.
Free audit, 3 analyses per month, no credit card.