How we calculate your SEO, GEO and GEI scores
This page describes the calculation exactly as it is programmed. Every number below is read from the scan's code: when the calculation changes, this page changes with it.
Method last updated: November 3, 2026
Three scores, three different questions
- SEO, out of 100: can Google read your site properly? We start at 100 and deduct points for each problem found.
- GEO, out of 100: is your site ready to be read by AI assistants? Same principle, with AI-specific criteria.
- GEI: do AI assistants actually name you? We ask buyer questions to an AI engine and count the answers that cite you.
SEO criteria and points deducted
When two numbers are shown, the deduction grows with the share of pages affected, up to the maximum.
| Criterion | Severity | Points deducted |
|---|---|---|
| Site could not be reached | critical | −100 |
| Store not served over HTTPS | critical | −20 |
| … page(s) without a title | major | −5 to 15 |
| Not mobile-ready (no viewport) | critical | −15 |
| … internal page(s) return an error | major | −4 to 12 |
| … page(s) without a meta description | major | −4 to 12 |
| … pages share the same title | major | −8 |
| … page(s) without an H1 heading | minor | −2 to 8 |
| … page(s) under 150 words | major | −8 |
| … page(s) without a canonical URL | major | −6 |
| No sitemap.xml | minor | −5 |
| … of … images without alt text | minor | −5 |
| … title(s) over 60 characters | minor | −4 |
| … page(s) over 1.5 MB of HTML | minor | −4 |
| No robots.txt | minor | −3 |
| … title(s) under 20 characters | minor | −3 |
| … page(s) with several H1 headings | minor | −3 |
| Page language not declared | minor | −3 |
| … page(s) without share preview (Open Graph) | minor | −3 |
GEO criteria and points deducted
Store criteria: only counted for a site that sells products.
| Criterion | Severity | Points deducted |
|---|---|---|
| robots.txt blocks … AI crawlers | critical | −25 |
| No Product structured data (JSON-LD) (store) | critical | −20 |
| No llms.txt file | major | −10 |
| …/… products with a weak or missing description (store) | major | −10 |
| No FAQ markup (FAQPage) | major | −8 |
| No Organization/WebSite schema | major | −8 |
| Customer reviews missing from the schema (store) | minor | −5 |
| Home meta description weak or missing | minor | −4 |
| Sitemap missing or empty | minor | −4 |
| … pages without an H1 | minor | −3 |
A score can never exceed what we actually checked
Both raw scores measure preparation, not results. The score shown is capped:
- Too thin a sample (fewer than 3 pages and no product): 69 at most, for SEO and GEO.
- GEO, never cited in the answers tested: 39 at most.
- GEO, cited in fewer than half of the answers: 59 at most.
- GEO, cited but not in every answer: 74 at most.
- GEO, citation not measured: 59 at most.
- SEO, fewer than 3 pages read: 39 at most; fewer than 10: 59; fewer than 20: 74.
- SEO, without a connected Google Search Console: 85 at most.
AI engines and number of questions
Free scan: 3 buyer questions asked to one AI engine. Fallback order: Agnes AI → Mistral → Gemini → DeepSeek → Qwen → GLM. The first engine that answers is used, and the report names the engine actually asked.
The questions, asked in English with the category inferred from your site:
- What are the best online stores to buy [your category]?
- Where can I buy high-quality [your category] online? Name specific stores.
- I want to buy [your category]. Which brands or stores do you recommend?
These are the questions for a store; other wordings are used for software or a service.
Paid tracking: up to 20 questions (4 question families in 2 languages, or your own questions), each asked to at most 3 engines among: Qwen, GLM, Gemini, DeepSeek, Mistral, ChatGPT, Claude, and Perplexity. Only engines configured on our servers at measurement time are asked.
One question asked to one engine counts as 1 against your plan's monthly volume: Measure 30 per month, Fix 150 per month, Steer 400 per month, Agencies 800 per month, Agency+ 1,500 per month, and Agency Max 3,000 per month.
Being named first counts more than being named sixth. Points per position: 1 → 10, 2 → 8, 3 → 6, 4 → 5, 5 → 4, 6+ → 2; absent = 0. The visibility score is the average over the questions asked (0 to 10).
Date and frequency of measurements
The free scan measures at the moment you run it: its result is valid for that day and time only.
Paid tracking can be scheduled: every day, every week, and every month. Each measurement is recorded with its date.
Uncertainty interval
With few questions, chance weighs heavily. We therefore show a 95% confidence interval (Wilson method) around your citation share.
Example computed by the code: cited 0 times out of 3, the true share lies between 0% and 56%.
Below 10 questions, the report flags a small sample.
Visibility index (Analyst)
With Steer, Agencies, Agency+, and Agency Max, the visibility index combines four independent measures, out of 100:
- preparation (GEO score): 20%
- citations by AI assistants: 40%
- visits coming from AI assistants: 25%
- revenue coming from AI assistants: 15%
A missing measure is never counted as zero: the index uses only the measures available, and how many were taken is always shown next to it.
Limits
- A score is not a guarantee: a good score guarantees neither a Google ranking nor a recommendation by an AI assistant.
- AI answers vary from one day to the next, from one engine to another and from one language to another: two close measurements can give different results.
- Without Search Console, we see neither indexing, nor inbound links, nor your actual Google rankings.
- We read a sample of pages, not necessarily the whole site.
What makes the result vary
- the AI engine that answered (fallback when an engine is unavailable);
- the category inferred from your site, used to write the questions;
- the number of pages read and what they contain;
- the language of the questions;
- changes to your site between two measurements;
- an engine that fails: the question is not counted, neither as cited nor as not cited.