Full transparency

How Detekia
calculates your score

Our GEO Score is calculated from 8 measurable and verifiable criteria, based on academic research in Generative Engine Optimization and the public guidelines of major AI models.

The process

3 automated analysis steps

01
Content scraping
Detekia retrieves the full HTML of your page via a specialized AI scraping service. The raw HTML, text and metadata are extracted.
02
Heuristic analysis
Our analysis engine computes 7 heuristic scores by inspecting the DOM: tags, attributes, ratios, counts. These scores are reproducible and deterministic.
03
AI analysis
Our AI analysis engine evaluates the text content to assess editorial neutrality and generate contextualized recommendations. The final score incorporates this result.
3 AUDIT LEVELS

Same methodology, 3 levels of depth

The 8 criteria and scoring are identical across all 3 audits. What changes is the depth of analysis.

Free
1 page — €29
Complete — €99
8 GEO criteria analyzed
✓
✓
✓
Pages analyzed
1
1
10
AI visibility test
Partial
10 queries
30 queries
Recommendations
Preview
Detailed + code
Detailed + code
Cross-page patterns
—
—
✓
Prioritized action plan
—
—
✓
Per-page breakdown
—
—
✓
Downloadable PDF
—
✓
✓
Report accessible
24h
10 years
10 years
Delivery time
30 sec
1 min
5-10 min
See our detailed plans →
Score calculation

Score out of 100 = 7 heuristics + AI neutrality

The first 7 criteria are calculated by deterministic rules on the HTML. The 8th criterion (editorial neutrality) is scored by Claude out of 10, then converted into a -3 to +3 point bonus/penalty added to the total.

25
pts
Extractability
20
pts
Verifiability
15
pts
E-E-A-T Authority
15
pts
Crawlability
10
pts
Structured data
10
pts
Editorial neutrality
5
pts
External presence
5
pts
Freshness
Criteria details

The 8 criteria explained

🎯
Extractability & direct answer
25 pts
What we measure
AI models look for "ready-to-cite" answers in your content. A good website answers the question in the first few lines, with lists, tables and clear headings.
✓ Good signal
Intro that directly answers in 1-2 sentences, bullet lists, data tables, logical H2/H3 structure, short paragraphs (50-300 characters).
✗ Bad signal
Generic introduction with no direct answer, dense text blocks, no lists or tables, flat structure with no hierarchy.
🔬
Verifiability & evidence
20 pts
What we measure
AI models prioritize content that backs up its claims: sourced figures, precise dates, links to primary sources, citations.
✓ Good signal
Figures with sources, explicit dates, external links to studies or reference organizations, comparison tables, expert quotes.
✗ Bad signal
Unsourced claims ("it has been proven that..."), missing or unsourced figures, no external links, no publication date.
🏆
E-E-A-T Authority
15 pts
What we measure
E-E-A-T (Experience, Expertise, Authoritativeness, Trustworthiness) is the framework AI models use to evaluate the credibility of a website and its authors.
✓ Good signal
Author page with bio and photo, "author" schema in JSON-LD, detailed About page, Contact page, complete legal notices, Organization schema.
✗ Bad signal
Anonymous content with no identified author, no About or Contact page, no legal notices, no schema.org for entities.
🤖
AI Crawlability
15 pts
What we measure
AI bots (GPTBot, ClaudeBot, OAI-SearchBot) need to be able to read your website. A blocked or poorly accessible site cannot be indexed.
✓ Good signal
Lang attribute defined, correct canonical, content not blocked by JavaScript, AI bots mentioned in robots.txt or headers, sitemap.xml.
✗ Bad signal
Meta robots noindex, content loaded only via client-side JS without SSR, missing lang attribute, redirect loops, no canonical.
🧩
Structured data
10 pts
What we measure
Schema.org is the language AI models read first to understand what your website is about, who wrote it, and what it covers.
✓ Good signal
FAQPage, Article, HowTo, Organization, Person schemas in JSON-LD. Priority schemas (FAQPage, HowTo) are particularly valued.
✗ Bad signal
No schema.org, incorrect or incomplete schemas, missing schemas for key entities (Organization, Author).
⚖️
Editorial neutrality
10 pts
What we measure
AI models avoid citing overly promotional or biased content. Factual, nuanced text is far more likely to be used as a source.
✓ Good signal
Informative and factual tone, nuances present, sourced claims, precise vocabulary without excessive superlatives.
✗ Bad signal
Unsourced superlatives ("the best", "revolutionary"), unproven claims, purely promotional tone, lack of nuance.
🌐
External presence
5 pts
What we measure
AI models cross-reference multiple sources to evaluate your credibility. Being mentioned in the press or cited on other websites strengthens your perceived authority.
✓ Good signal
Press mentions with links, active presence on LinkedIn/Twitter, referenced testimonials and reviews, citations in third-party articles.
✗ Bad signal
No identifiable inbound links, no social media presence, no press or third-party mentions.
📅
Freshness & maintenance
5 pts
What we measure
For evolving topics, AI models prefer recent content. A visible update date and current copyright signal that the website is actively maintained.
✓ Good signal
dateModified in JSON-LD, visible update date, current year copyright, recent content in meta tags.
✗ Bad signal
No visible date, copyright from 3+ years ago, missing dateModified schema, articles without publication date.
Score normalization
The raw score from 7 criteria (excluding neutrality) totals 95 points. The Editorial Neutrality criterion, evaluated by AI, adds an adjustment of -3 to +3 points. The final score is capped at 100.
Documented case studies
Each report includes real-world case studies on the 3 weakest criteria (SEO Vendor, Ahrefs, Stackmatix, etc.) to illustrate concrete impact of optimizations.
Analysis limitations
Page-level analysis
Detekia analyzes precisely the page you submit (homepage by default, or any other URL of your site), not the entire site. For most sites, the homepage concentrates the global signals (robots.txt, Organization schema, authority, social presence) and provides a representative diagnostic. To analyze multiple pages with a consolidated report, see our Pro offer (coming soon).
Non-measurable factors
Intrinsic content quality, offline reputation, inbound backlinks, and AI citation history cannot be measured by our technical analysis.
Algorithm evolution
AI citation criteria evolve regularly. Our methodology is updated but may lag behind the latest practices of AI models.
Dynamic websites
Websites that load their content entirely via client-side JavaScript (SPA without SSR) may receive underestimated scores because the content isn't accessible to the scraper.
Sources—Aggarwal et al., Princeton / Georgia Tech, KDD 2024·Google E-E-A-T (2024)·OpenAI GPTBot·Anthropic ClaudeBot·Schema.org·SE Ranking (2025)·Otterly.AI (2026)·AirOps (2026)·Seer Interactive (2025)·Edelman (2026)

Test your website
for free

Full analysis in under 60 seconds. No signup required.

Analyze my website →