Free diagnosis · inspectable evidence

AI agent readiness test

See what an AI agent can actually read on your website. Find the obstacles, inspect the evidence and leave with clear next fixes.

View a sample report
  • 18

    checks linked to their sources

  • 45

    technical points, open scoring method

  • 3

    missions to run after the scan

Tool V6.4 · technical scoring method V5

Free score

Start your diagnosis.

No account · no credit card · shareable report

Add competitors or an industry Optional

Optional, used only to make the benchmark more relevant. You can skip it.

Competitor comparison Compare the same public signals across up to three websites.

The scan reads public pages only. It never requests access to your administration area.

  • robots.txt
  • sitemap
  • structured data
  • contenu lisible
  • key pages

URLs and observations are included in the shareable report. Understand what data is retained.

A real diagnosis, before you begin

Here is what you get.

We run the same test on our own website. This dated example shows the observations, their scope and the next decision to make.

  1. IdentifyA score backed by the pages that were actually read.
  2. VerifyOne finding, source and verification rule for every point.
  3. ActPriorities, followed by a new test after the fixes.
Analyse my website

Edikka example

edikka.com

45/45
Strong public signals

6/6 readable pages · 18 checks assessed

Next verification

The foundations pass these checks. The next step is to test whether an agent can understand the offer, then measure real citations separately.

Open the full report and evidence

Observation of our own website using our scoring method. This 45/45 is neither an independent certification nor a guarantee of citation. Frozen source: method 2026-09-09.v5.

Fictional teaching example · 1 HTML page

When the test finds obstacles.

Three fixes on a page intended for search engines. The same engine calculates both scores from the sources below.

Before32/45
After44/45
  1. Make this page indexable

    +2 pts

    robots: noindex

    This page is intended for search engines: remove the noindex directive, then rerun the check. Preserve intentional exclusions.

  2. Declare the reference URL

    +2 pts

    No canonical URL was extracted.

    Add a consistent canonical URL and verify its destination.

  3. Repair the JSON-LD

    +8 pts

    At least one block contains a JSON syntax error.

    Remove the trailing comma, validate the JSON, then review the information it describes.

    Three checks become verifiable again: types, syntax and entity (+4, +2, +2).

The first fix matters more than its score. Removing noindex adds only 2 points, but it is the priority for a page that should be indexed. The remaining point is the optional AI context signal; it does not prevent publication.

Inspect the teaching example sourcesBefore-and-after HTML, conditions and 18 calculated checks.

Fictional company and URL: atelier-exemple.test. No remote website was scanned and no observation was added to the Observatory. This is not a previous state of Edikka. Scoring method: 2026-09-09.v5.

Identical conditions: HTTP 200 response, robots.txt allowing all paths, sitemap present and no optional AI context file. One HTML page is assessed; links are not followed.

HTML before the fixes
<!doctype html>
<html lang="en"><head>
<title>Example Studio — Consulting for SMEs</title>
<meta name="description" content="Operations consulting for SMEs: diagnosis, action plan and progress tracking.">
<meta name="robots" content="noindex">
<script type="application/ld+json">{"@context":"https://schema.org","@type":"Organization","name":"Example Studio",}</script>
</head><body><main>
<h1>Operations consulting for SMEs</h1>
<p>Example Studio is a fictional company used to explain a technical diagnosis. Its offer includes a scoping workshop, a process analysis and a prioritised action plan. This demonstration page is intended to be accessible to search engines.</p>
<h2>How does an engagement work?</h2>
<p>The project lead describes their objectives and shares the relevant documents. A consultant identifies friction points, presents the findings and proposes verifiable steps. Deliverables include an executive summary and a decision list.</p>
<h2>Frequently asked questions</h2>
<p>What results can be expected? A documented plan with sources for each finding. This example guarantees no commercial outcome. All company information is entirely fictional.</p>
<a href="/en/expertise">Offer details</a><a href="/en/contact">Contact the studio</a>
</main></body></html>
HTML after the fixes
<!doctype html>
<html lang="en"><head>
<title>Example Studio — Consulting for SMEs</title>
<meta name="description" content="Operations consulting for SMEs: diagnosis, action plan and progress tracking.">

<script type="application/ld+json">{"@context":"https://schema.org","@type":"Organization","name":"Example Studio"}</script>
<link rel="canonical" href="https://atelier-exemple.test/">
</head><body><main>
<h1>Operations consulting for SMEs</h1>
<p>Example Studio is a fictional company used to explain a technical diagnosis. Its offer includes a scoping workshop, a process analysis and a prioritised action plan. This demonstration page is intended to be accessible to search engines.</p>
<h2>How does an engagement work?</h2>
<p>The project lead describes their objectives and shares the relevant documents. A consultant identifies friction points, presents the findings and proposes verifiable steps. Deliverables include an executive summary and a decision list.</p>
<h2>Frequently asked questions</h2>
<p>What results can be expected? A documented plan with sources for each finding. This example guarantees no commercial outcome. All company information is entirely fictional.</p>
<a href="/en/expertise">Offer details</a><a href="/en/contact">Contact the studio</a>
</main></body></html>
Points by check — same page, same method
CheckBeforeAfter
Page accessible over HTTP4/44/4
Verifiable robots policy3/33/3
XML sitemap detected3/33/3
No noindex directive0/22/2
Canonical URL declared0/22/2
Title and H1 present2/22/2
Access rules available1/11/1
Generic access to the tested URL5/55/5
Explicit AI context · bonus0/10/1
JSON-LD block present4/44/4
Structured types identified0/44/4
Readable JSON-LD syntax0/22/2
Organisation or website entity0/22/2
Title and description extracted2/22/2
Content organised into sections2/22/2
Links to useful pages2/22/2
Question-or-answer signals2/22/2
Evidence mentions detected2/22/2

Method

One scoring method. Every point explained.

Eighteen checks grouped into five families. The weights are an Edikka convention: they describe observed public signals, not a probability of being cited.

16 pts

Technical accessibility

HTTP, robots.txt, sitemap, indexability, canonical URL and headings.

7 pts

Crawler access

Policy and access to the tested path: 6 points. Optional AI context: 1 point, already included in both the 7-point family and the 45-point total.

12 pts

Structured data

Presence, JSON-LD syntax, types and entities. Accuracy still requires review.

8 pts

Readable content

Metadata, sections, useful links and question-or-answer signals.

2 pts

Evidence mentions

Signals of cases, sources or results. Authenticity still requires verification.

Method published by Bertrand Morel, founder of Edikka · updated . Maximum: 45 points, bonus included.

Tool version: V6.4. Technical scoring method: 2026-09-09.v5, retained for scan comparability. The mission protocol evolves separately.

Review all 18 scoring criteriaWeights, scope and references for method 2026-09-09.v5.

Edikka editorial scoring method. Weights are not a probability of citation. Answer and evidence signals detect mentions; their accuracy still requires review. An unverified check suspends the overall comparison.

Scope, limitations and versionsHow to interpret a result and compare two scans.

What the scan reads. The received HTML and public rules across a small set of key pages. It identifies readable pages and unassessed checks; it does not execute JavaScript or crawl the entire website.

What still requires review. Content accuracy, reputation, backlinks and citations in assistants. JSON-LD or words such as “reviews” and “results” do not prove that a claim is true.

Business context. Eight optional questions shown after the scan produce up to 55 declarative points. They remain separate from the 45 technical points and mission results.

Why V5 and V6 coexist. The interface and missions belong to V6. The technical scoring method remains 2026-09-09.v5 to preserve comparability with earlier diagnoses. Missions have their own protocol and do not change this score.

Read the full methodology in Markdown

Understand and extend the diagnosis

From the first check to real visibility.

France GEO Observatory: when is a comparison published?30 distinct websites per industry or 100 in the global sample.

One recent, comparable observation is retained per website. No median is published before these thresholds. Incomplete or unreadable scans remain useful to their owner but are excluded from statistical comparisons.

The sample depends on websites submitted voluntarily and is not representative of every French website. The 2026 GEO study is separate research with its own sample and scoring method.

One diagnosis. A clear way forward.

Take action Start with the observed obstacles. Edikka can help prioritise the pages, content and journeys that need improvement.

FAQ

Answers before you start

Cost, data, reading limitations and progress tracking.

7 featured questions See all FAQs

Is the test free, and do I need an account?

The technical diagnosis, evidence and exports are free, with no account or credit card. Automated missions use a limited shared allowance. If it is exhausted, the scan remains available and you can run the missions with your own assistant.

What does the 45-point score measure?

It combines 18 checks of the received HTML and public rules: technical access, robots, structured data, readable content and evidence mentions. A check that cannot be verified remains unassessed. The score does not measure citation frequency or the commercial quality of the website.

What happens if my website blocks the scan?

An inaccessible page or unreadable JavaScript-rendered page does not receive a zero. The report records the reading limitation. Check public access and server rules before concluding that content is missing.

What is the difference between the scan and the three missions?

The scan inspects technical signals. The missions ask an agent to understand an offer, retrieve sourced information and identify the next action. Responses are retained, quoted passages are verified and results remain separate from the technical score.

Are my URL and report public?

The scan covers public pages. Its shareable report contains URLs and observations, but no contact details. Avoid URLs containing secrets. Technical reports carry a noindex directive. The selected Edikka editorial example is visible on this page.

How can I measure progress after a fix?

Run the same website again in the same browser. Local history retains twelve diagnoses and compares complete scans with the same method and scope. Missions have their own retest. A scope change or incomplete reading suspends the numerical comparison.

Why is the industry comparison sometimes unavailable?

A median is published from 30 distinct comparable websites per industry, or 100 websites in the global sample. The observatory remains under construction below that threshold. This voluntary sample is not representative of every French website.