Citerank
What Is an AI Visibility Score? How It's Calculated (and What Counts as Good)
Key takeaways
- An AI visibility score measures how discoverable, trustworthy, and citable a website is to AI engines like ChatGPT, Perplexity, Gemini, and Google AI Overviews, expressed as a single number and letter grade.
- A real score is not one number. It has to break into layers, because a site can be strong on content and invisible on discoverability or agentic access at the same time.
- Grade bands that hold up in practice: 85+ is an A (Elite), 70 to 84 is a B (Strong), 55 to 69 is a C (Developing), 40 to 54 is a D (Emerging), and under 40 is an F (Foundational).
- A single snapshot score is close to meaningless on its own. AirOps found only 30 percent of brands stay visible across consecutive AI answers on the same query, so the score has to be tracked over time, not checked once.
- Most tools calling themselves an 'AI visibility checker' only measure whether your brand gets mentioned. That is one input, not the whole picture. Discoverability, entity clarity, and agentic access matter just as much and are rarely measured at all.
An AI visibility score measures how likely AI engines are to find, trust, and cite your website when answering a question a real person asked. A well-built score is not just "did ChatGPT say my brand's name." It grades the specific, fixable reasons an engine would or would not cite you: whether your schema makes your entity unambiguous, whether crawlers and agents can actually reach your content, whether that content is written in a format worth quoting, and whether you show up consistently across engines rather than once by luck.
Most of the tools ranking for this term right now are checkers, not scores. They run a handful of prompts and report a mention count. That tells you an output happened. It does not tell you why, or what to fix. Being cited once is not the same as being reliably citable, which is the distinction this score is built to measure. Here is the methodology we use, the real grade bands, and the one thing almost every guide on this topic leaves out: a single check is close to worthless without a volatility number attached to it.
What is an AI visibility score measuring, exactly?
It is measuring citability, not popularity. A site can have huge brand recognition and still score poorly if an AI engine cannot parse its schema, cannot crawl its content, or has nothing callable by an agent. Conversely, a small, focused site with clean entity signals and direct-answer content can outscore a household name.
A score built for this has to separate at least five things, because they fail independently:
- Schema and entity clarity: does structured data make it unambiguous who you are and what you offer
- AI discoverability: can crawlers and agents actually reach your content (robots.txt, llms.txt, sitemap, rendering)
- Agentic protocol support: is anything on the site callable by an AI agent, not just readable by a human
- Content citability: is the content written in a direct-answer, quotable format an engine can lift cleanly
- Brand presence: does the entity have enough corroborating signal (sameAs links, consistent NAP, third-party mentions) to be trusted
A site strong on one or two of these and weak on the rest still ends up with a mediocre blended score, which is the point. The score should punish imbalance, because engines do not cite sites that are strong in theory and unreachable in practice.
How is an AI visibility score calculated?
Here is the actual weighting we use in Citerank's scoring engine, published so you can see exactly what moves the number.
Quick score (on-site only):
| Layer | Weight |
|---|---|
| Schema & Entity | 25% |
| AI Discoverability | 25% |
| Content Citability | 22% |
| Brand Presence | 18% |
| Agentic Protocol | 10% |
Deep score (adds live market data, on-site layers scaled to 55% of the total):
| Layer | Weight |
|---|---|
| Live AI Citations (real prompts run against ChatGPT, Perplexity, Gemini) | 22% |
| AI Brand Recognition (do engines already know the entity unprompted) | 8% |
| Knowledge Graph presence | 7% |
| Backlink Authority | 8% |
| On-site layers (scaled) | 55% |
The deep version is the one that actually answers "does an AI engine cite me today," because it runs real prompts against real engines instead of only inferring citability from on-site signals. The quick version is what you can check in seconds with no account.
If a domain has no reachable content at all, one dimension can be a hard override. Everything else becomes irrelevant if an engine cannot parse the page in the first place, which is why discoverability carries as much weight as schema.
What is a good AI visibility score?
These bands are on a 0 to 100 scale with a five-tier system:
| Score | Grade | Tier | What it means |
|---|---|---|---|
| 85–100 | A | Elite | Rare. Strong across every layer, agentic access included. |
| 70–84 | B | Strong | Solid fundamentals, one or two layers still leaking points. |
| 55–69 | C | Developing | Most common band. Content and schema exist, discoverability or agentic access is weak. |
| 40–54 | D | Emerging | Basic signals present, multiple layers close to zero. |
| 0–39 | F | Foundational | An AI engine has almost nothing reliable to work with. |
To calibrate this against something real rather than a made-up benchmark: we ran 19 well-known SaaS brands through this exact scoring engine. The average was 44, a D. The single highest score across all 19 was Zapier at 69, a C. Not one brand reached a B or an A. The weakest shared layer was agentic protocol, where 11 of 19 sites scored a flat zero.
That is the honest baseline. If your own score comes back in the 60s, you are already outperforming most recognizable software brands on this axis. If it is under 40, you are in the same position most of the internet is in right now, which is also the opportunity: the bar is low enough that focused fixes move the number fast.
What is the 30 percent rule, and why does it matter more than the score itself?
This is the part most guides on this topic skip entirely, and it is the most important one.
AirOps' research into AI answer volatility found that only about 30 percent of brands that appear in one AI-generated answer are still present in the very next answer to the same prompt. Only about 1 in 5 stay visible across five consecutive runs of the identical query. AI models are not deterministic the way a Google ranking used to feel deterministic. A model update, a slightly different phrasing of the same question, or nothing you can observe at all can flip which sources get cited.
That means a single AI visibility check, run once and treated as a verdict, is close to meaningless. A score of 70 today could be a 40 next week for reasons that have nothing to do with anything you changed. The number that actually matters is not one snapshot. It is the trend across repeated checks over weeks, and specifically whether your citation rate holds steady, climbs, or degrades when the same query is run again.
This is why an AI visibility score is a measurement practice, not a one-time report. Check it, fix the weakest layer, check it again a week later, and watch the trend rather than the single number.
How do you check your own AI visibility score?
Run your domain through Citerank's free AI Visibility Score. It returns the overall score and grade along with all five on-site layer scores broken out individually, so instead of a single number you can see exactly which layer is dragging the average down. For the deep version with live citation testing across ChatGPT, Perplexity, and Gemini plus repeat-run volatility tracking, that is available in the full Citerank Score audit.
Either way, the fix is always the same shape: find the weakest layer, fix that one thing, re-check in a week, and watch whether the number holds instead of assuming one good result means the work is done. If discoverability turns out to be your weak layer, our complete GEO guide covers the fix in depth; if the gap is on the content side, start with why sites don't get cited by AI.
Sources: AirOps, "How to Measure AI Search Visibility" (2026).
See how AI search sees your site
Get your free AI Visibility Score in seconds, no signup required.
Get my free score