GEO Score: How to Measure Your Website's AI Visibility

You cannot improve what you do not measure. And in GEO, measurement is the main obstacle. There is no Google Search Console for ChatGPT. No rank report in Perplexity. No click data from Gemini.

Yet AI visibility can be measured. Not with the same tools or metrics as SEO, but with a structured approach that gives you a clear diagnostic and actionable priorities.

This article explains what a GEO score is, how it is calculated (as a complement to the complete 2026 GEO guide), how to interpret your results, and most importantly where to start to improve your visibility in AI answers.

Why measuring AI visibility is hard

In SEO, the measurement ecosystem is mature. Google Search Console tells you which queries lead to your site, which pages are indexed, what your CTR is. Tools like Ahrefs, Semrush, or Moz track your rankings daily.

In GEO, none of that exists — for three fundamental reasons.

AI answers are dynamic. Unlike Google SERPs, which are relatively stable, ChatGPT or Perplexity answers change with every query. Asking the same question twice can produce different answers, citing different sources. There is no fixed "position 1."

AI engines do not share their data. OpenAI, Google, and Anthropic do not provide an analytics console for site publishers. You cannot know how many times ChatGPT cited your site, or for which queries.

Citations are contextual. An AI may cite your site for one query and not for another very similar one. The citation depends on the conversation context, the question wording, and sometimes even the user's profile.

Despite these challenges, two complementary approaches let you measure your AI visibility.

Approach 1: manual testing

This is the most direct method. It consists of querying AI engines the way a potential customer would, and observing whether your site appears in the answers.

How to proceed

Step 1 — Define your target queries

List 10 to 15 questions your customers actually ask. Think in three categories:

• Recommendation queries: "What is the best [service] in [city]?", "What solution for [problem]?"
• Expertise queries: "How do I [do X]?", "What's the difference between [A] and [B]?"
• Brand queries: "What does [your company] do?", "Reviews of [your product]"

Step 2 — Query 3 AI engines

Ask each question on ChatGPT, Gemini and Perplexity. For each answer, note:

• Is your site cited as a source? (yes/no)
• Is your brand mentioned? (yes/no)
• Which competitors are cited?
• Does the answer reference content from your site?

Step 3 — Analyze the results

Calculate your presence rate: across your 10 to 15 queries × 3 engines, in what percentage of answers do you appear? A practical benchmark, with no statistical value: if you are missing from more than 4 answers out of 5 (below 20%) on questions directly about your offer, AI visibility should be a priority.

Limitations of manual testing

This approach gives you a snapshot at a single point in time, not ongoing tracking. It is also subjective: results vary based on wording and context. It is a good starting point, but it is not enough to drive a GEO strategy.

Approach 2: technical analysis

Technical analysis evaluates a page on objective criteria linked to AI citation. Unlike manual testing, it measures citation potential: the technical and editorial conditions that increase your chances of being selected as a source.

This is the approach Detekia uses: a free, automated GEO score that evaluates a page on 6 weighted criteria, after a readability prerequisite, and assigns it a score out of 100. Since September 30, 2026, the weights have come from our study of 317 pages, cited or not by AI engines.

The 6 criteria of the Detekia GEO score

Each criterion is measured by analyzing the actual HTML of your page, not by estimation: no AI is involved in the calculation. For a detailed explanation of each criterion and of the study that calibrated them, see the Detekia score methodology.

The prerequisite: readable by AI

Before the criteria, Detekia checks that AI engines can read the page:

• AI search crawlers (OAI-SearchBot, ChatGPT-User, PerplexityBot, Perplexity-User, Claude-SearchBot, Claude-User, Googlebot, Bingbot) are allowed in robots.txt; training crawlers (GPTBot, ClaudeBot, Google-Extended) do not count
• the page has no noindex tag
• the content is present without JavaScript

This prerequisite earns no points, but if any condition is missing, the score is capped at 40.

→To make your site readable by AI, read llms.txt, robots.txt and AI accessibility.

1. Answer structure (40 points)

This criterion measures whether the page reads as a series of answers ready to be reused. The analysis checks:

• a short answer (15 to 70 words) right after H2 and H3 headings, in proportion to the headings (up to 14 points)
• question-format headings (up to 9 points)
• a data table (8 points, or 3 points for at least two lists)
• paragraphs of 20 to 80 words (up to 4 points)
• a clear hierarchy: one H1 and at least two H2s (5 points)

What loses points: vague headings followed by long blocks of text, no table, no question in the headings.

2. Content depth (15 points)

This criterion counts the words of the page's main content (main or article tag, otherwise the whole page). Thresholds depend on the page type: from 150 words (2 points) to 1,500 words (15 points) for a homepage, from 300 words (2 points) to 3,000 words (15 points) for other pages.

What loses points: a page of only a few paragraphs, or main content drowned in menus and footer.

3. Evidence & data (15 points)

This criterion counts figures (up to 6 points), external sources linked in the content, excluding social networks (up to 5 points), and explicit attributions such as "according to", "a study" or "source:" (up to 4 points).

What loses points: messaging with no figures, no links to primary sources, no attribution.

4. Freshness (15 points)

This criterion looks at the age of the most recent date on the page (under 90 days: 9 points; under a year: 6; under two years: 2), whether a date is visible to readers (3 points) and whether the current or previous year appears in the content (3 points).

What loses points: no visible date and an old last update.

5. Entity clarity (5 points)

This criterion checks that AI engines know who is speaking: brand name in the title tag (1 point), in the H1 or the first 120 words (2 points), a meta description of 70 to 320 characters (1 point), an organization named in structured data (1 point). Our study did not confirm the effect of these signals, hence the low weight.

What loses points: a generic title ("Home") and a brand missing from the text.

6. External signals (10 points)

This criterion evaluates what exists about your brand outside your site: a Wikidata entity whose official website is your domain (5 points), a French or English Wikipedia article (2 points), and customer reviews or testimonials on the page (3 points).

What loses points: no Wikidata entry linked to your domain and no visible customer reviews.

How to interpret your score

Thresholds

Where most sites stand

Most sites are not at all optimized for generative engines. That is bad news for them, but good news for you: by optimizing now, you gain a considerable head start over your competitors.

The heaviest criterion

Answer structure is worth 40 points out of 100. It is what most clearly set cited pages apart from uncited ones in our calibration study: short answers under headings, question headings, tables. Many pages have content, but not in a form AI engines can reuse as is. That is where the biggest gains are.

How to read the thematic groups

Detekia organizes the 6 criteria into 3 thematic groups that make reading and prioritization easier:

The "Readable by AI" prerequisite is not a group: if it is missing, it caps the score at 40, whatever the other results.

Prioritize your actions by impact

Not all criteria carry the same weight, and not all fixes have the same effort-to-impact ratio. Here is how to prioritize:

Immediate-impact actions (this week)

• Lift the cap: if your robots.txt blocks OAI-SearchBot, PerplexityBot or Claude-SearchBot, if the page is set to noindex or if its content only exists in JavaScript, the score is capped at 40. Removing a few lines from robots.txt takes 15 minutes. Impact: a prerequisite for any citation.
• Add a short answer under each heading of your key pages, in two or three sentences. Effort: 1-2 hours. Impact: up to 14 points in answer structure.
• Show an update date and fill in dateModified. Effort: 30 minutes. Impact: up to 12 points in freshness if the page was just reviewed.

Medium-term impact actions (this month)

• Rephrase headings as questions and add a table: up to 9 points for headings, 8 points for a data table.
• Add figures and sources: replace every vague assertion with data linked to its primary source. Effort: half a day. Impact: up to 15 points in evidence and data.
• Complete your most important pages: definitions, steps, examples, frequently asked questions. Impact: up to 15 points in content depth.

Long-term impact actions (this quarter)

• Create or complete your company's Wikidata entry, with your domain as the official website. Impact: 5 points in external signals, and a clearer identity for AI engines.
• Show customer reviews or testimonials on key pages. Impact: 3 points in external signals.
• Earn mentions on third-party sites (media, comparison sites, industry directories): this does not count in a page's score, but in our study, 79 to 89% of citations went to third-party sites rather than brand websites.

Track progress over time

A single score is useful for diagnostics. But tracking over time is what lets you steer a GEO strategy.

What to track

• Overall score — is it increasing after each optimization wave?
• Per-criterion scores — which criterion is progressing, which is stalling?
• Score vs. competitors — analyze your competitors' sites on Detekia to benchmark yourself
• Manual test results — redo the 3-question test (recommendation, expertise, brand) every month

The score measures a page's citation potential; it doesn't tell you what AI engines actually answer in your market. To complete this tracking, our guide explains how to run a full GEO audit, and Beeleven, the agency that built Detekia, can measure your mention rate in ChatGPT, Gemini, Claude and Perplexity answers, against your competitors.

Recommended frequency

• Technical score: after each optimization wave, then monthly
• Manual test: monthly on your top 5 queries
• Editorial review: quarterly for content updates

Correlations to observe

After a technical optimization (robots.txt, noindex, rendering without JavaScript), run the manual test again at regular intervals. How long it takes for AI answers to change varies by engine and by how often they re-read your pages.

After a content optimization (answer structure, evidence), same principle: measure before, then at regular intervals (for example monthly), rather than waiting for a specific date.

External presence work (Reddit, press) takes the longest to build, but it is also the most durable lever.

What the score does not measure

The GEO score is an indicator of potential, not a guarantee of citation. A few important nuances:

The score does not predict exact queries. A good score means your site meets the technical and editorial conditions for being cited. But AI engines decide case by case for each query. A score of 85/100 does not mean you will be cited in 85% of answers.

The score does not measure topical relevance. If your site is perfectly optimized technically but your content does not address the topic a user is asking about, the AI will not cite you. GEO optimizes the conditions for citation, not content relevance.

The score evolves with the market. A score of 60/100 can be excellent in an industry where the average is 25, and insufficient in an industry where your competitors are at 75. Always interpret your score in context.

Frequently asked questions

Does a good GEO score guarantee being cited by AI?

No. The score measures citation potential — the conditions your site meets for being selected as a source. AI engines remain unpredictable in their selection. But a site that meets these conditions starts with an edge over a site that blocks AI crawlers or sources nothing.

What score should I aim for?

From 70/100, the score is good: your page has most of the expected signals. The goal is not 100/100, but to leave no important criterion aside and to do better than your competitors on the pages that matter.

Does the score change if I do nothing?

It can decline, yes. If your competitors optimize and you do not, your relative position degrades. And if your content ages without updates, the freshness criterion will naturally decrease.

Can I compare my score to my competitors'?

Yes. Test the sites of your 3-5 main competitors with Detekia's free GEO score. Compare overall scores and per-criterion scores. This tells you exactly where you lag behind — and where you have an edge.

How do I improve quickly?

The three most impactful actions: 1) make the page readable by AI (search crawlers allowed in robots.txt, no noindex, content without JavaScript), otherwise the score is capped at 40; 2) put a short answer under each heading, phrase headings as questions and add a table (Answer structure criterion, 40 points); 3) date your pages and source your figures (Freshness and Evidence & data criteria, 30 points combined). Then run the score again to measure the difference.

Get your score

The first step is knowing your current score. No guessing, no gut feeling: an objective measurement across 6 GEO criteria.

Analyze your site for free on Detekia: a score out of 100, 6 detailed criteria, recommendations prioritized by impact and a first citation test (5 questions put to one of OpenAI's models). To go further, Beeleven can analyze your brand's visibility on ChatGPT, Gemini, Claude and Perplexity.

→SEO vs GEO: what are the differences in 2026?