Short answer: The AIV score (AI visibility score) is a 0 to 100 score that the open-source tool geo-score gives a website from the outside on a public rubric: it measures whether AI answer engines can find, read, trust and cite the site, and anyone can rerun it. It measures readiness, meaning how well your site is set up for AI, not how often AI actually cites you. That second number is the citation rate, tracked separately and never added to the score.
Where the term comes from
AIV stands for AI visibility. It comes from geo-score's public scoring standard, the AIV rubric, whose version 1.1 opens with: "AIV stands for AI visibility: whether AI answer engines can find, parse, trust and cite a site." geo-score is our open-source checker (MIT license). The rubric, the scoring code and a public leaderboard of scored sites are all on GitHub, so anyone can read them and rerun them.
So the AIV score is not an internal grade of ours. Anyone can compute it with the rubric as published, and every geo-score report names the rubric version it used.
How the score is calculated
Each run samples 8 pages from your site (the homepage, main product or service pages, knowledge pages and recent content) and scores 21 checks in five areas:
| Area | Points | What it looks at |
|---|---|---|
| Reachable | 15 | Does robots.txt block AI retrieval crawlers; do pages open for their user agents; is the main text in the HTML the server sends |
| Understandable | 22 | Sitemap, llms.txt, Organization and other structured data, breadcrumbs, page-type markup |
| Content citability | 35 | Self-contained answer passages, headings phrased the way shoppers ask, dates, sourced figures, named authors |
| Brand credibility | 18 | Third-party listings, independent mentions, a knowledge-graph entry, sameAs links, video |
| Answer fit | 10 | Content shaped as one question, one answer; coverage of the questions shoppers ask; readiness for Chinese engines (only when Chinese pages are sampled) |
Four bonus checks (conventions few sites use yet, such as llms-full.txt and ai.txt) add up to 6 points and cost nothing if missing. Checks that cannot be observed leave the denominator, so a report reads "score / observable maximum" and is then normalised to 0 to 100.
The three Reachable checks are gates. If any of them scores zero, AI crawlers really cannot get your content, and the score is capped at 40. Nothing else counts until that door is open.
Normalised scores fall into five bands: 83 and up is Leading, 66 and up Solid, 51 and up Growing, 31 and up Early, and below that Not started. A band describes a stage, not a verdict on your site.
AIV score versus citation rate
| AIV score | Citation rate | |
|---|---|---|
| What it measures | How ready your site is: can AI get in, read it, and find something to cite | The share of AI answers to a fixed set of shopper questions that cite you |
| Same result for anyone? | Close to it: same-day reruns land within a few points | No: ask the same question twice and the cited sources can change |
| Can you change it directly? | Yes, every check maps to something on your site | No, it follows from readiness plus time |
| How we use it | Acceptance: the baseline and the lift go in the contract | Monitoring: logged weekly, read as a trend over weeks, reported as it is, outside acceptance |
Merge the two into one number and you can no longer tell whether your site changed or the AI simply answered differently this time. The rubric requires them to be reported separately.
How to check your own site
All you need is Python 3 on your computer: no account, no API key. Replace yourstore.com with your domain:
$ curl -sL https://raw.githubusercontent.com/jianruntech/geo-score/v1.8.3/cli/geo_score.py \
| python3 - yourstore.com --brief
If you have uv installed, uvx geo-score@1.8.3 yourstore.com --brief works too. Rather not use a terminal? Leave your site on the free check page and we will look at it by hand.
Two runs on the same site can differ by a few points. On October 12, 2026, geo-score 1.8.0 scored the leaderboard three times in one day; of the 317 sites scored every time, 99.1% stayed within 5 points between the first two runs (reproducibility report). A gain under 5 points after a change is not a result. Versions sometimes change how a check is judged, so compare a site with itself on the same version; the geo-score changelog lists what changed in each release.
How we use it, and what it does not tell you
- The contract covers the AIV score's baseline and lift, nothing more. A change under 5 points is not a result.
- Whether AI mentions you, and the visits and leads it sends, we track with you and report as recorded, outside acceptance: seasons, ads, prices and the AI products themselves move them too much.
- A high score means AI can reach you, read you and find something to cite. That is the precondition for being cited, not a promise that AI will recommend you. We have not shown that higher-scoring sites get cited more often, so we do not claim it.
- Gate problems (crawlers blocked, text that only appears after JavaScript) usually take days to fix and can be rechecked the same day; when each AI recrawls and starts citing you is up to each AI.
Common questions
What is a good AIV score?
Look at the band: 66 and up is Solid, 83 and up is Leading. More important is whether any of the three gates scored zero, because then the score is capped at 40 and that comes first. Across the 324 well-known sites we scored in public, the median is 55 (data).
Is the AIV score the same as an SEO tool's site score?
No. SEO scores mostly track rankings in search results and backlinks. The AIV score measures whether AI can find, read and cite your pages when it answers a question. They overlap (pages must be crawlable and indexed), and each looks at things the other ignores.
If my AIV score goes up, will ChatGPT recommend my store?
Not necessarily. The score is a precondition, not an outcome. Whether you get recommended also depends on answering the questions shoppers actually ask and on real discussion of you elsewhere, and every answer can differ. That is why citation rate is tracked on its own.
Why accept work against the AIV score and not traffic or leads?
Because it comes out the same whoever runs it and you can rerun it yourself, so both sides can check what changed and by how much. Traffic and leads matter just as much and we track them with you, but too much outside anyone's control moves them for them to be a fair test of the work.
What does your site score today? Run the command above, or leave your site and get a free review within 48 hours. The term is also in our GEO glossary.
Sources
- geo-score: AIV rubric v1.1: the AIV definition, two scores reported separately, scoring rules, points per area, gates and the 40% cap, bands; checked 2026-10-15
- geo-score: How reproducible is a single run? (the 2026-10-12 retest with 1.8.0); checked 2026-10-15
- geo-score README: sample output that names the rubric version; checked 2026-10-15