The AI Visibility score.
Share-of-voice 45%, AI access 25%, citability 15%, off-site footprint 15%. What each one measures, how to move it, and why a component we can't measure gets reweighted out rather than zeroed.
One number, out of 100, for a question search rankings can't answer: when someone asks an AI assistant for a recommendation in your category, do you come up? It's the headline on SEO & AI Visibility → AI Visibility, and it's built from four components you can each move independently.
This page explains what the number is made of. For the actual fixes, go to Getting cited by AI assistants — that's the page that changes your score.
The four components
| Component | Weight | The question it answers |
|---|---|---|
| AI Share-of-Voice | 45% | Are you actually being recommended? |
| AI access & readiness | 25% | Can the AI crawlers even reach you? |
| Content citability | 15% | Is your writing quotable once they get there? |
| Off-site footprint | 15% | Do you exist anywhere other than your own website? |
The order is the priority order. Share-of-voice is nearly half the score because it's the outcome; the other three are the levers that produce it. Your band follows the total: 90+ excellent · 70+ good · 50+ needs work · below 50 critical.
Share-of-voice is a comparison — it asks whether AI names you or them. With no competitors on your brand there is nobody to compare against, so the largest component can't run at all. Add at least one under Brand → Competitors before you run a check.
AI Share-of-Voice · 45%
We ask ChatGPT, Perplexity, Gemini and Claude real questions from your category and count how often you're named against how often your rivals are. It's the only component that costs anything to run, and it's the one that moves slowly — it reflects months of published work, not a config change.
It has a page of its own, because the sampling is worth understanding before you read a number off it. How share-of-voice is measured →
AI access & readiness · 25%
The plumbing. This is the component you can fix in an afternoon, and the one most sites are quietly failing. Its 25 points break down like this:
- Retrieval crawlers — 12 points. We check six user-agents in your
robots.txt: GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot, Google-Extended and Bingbot. You lose points pro-rata for each one you block — block all six and you score zero here. (We deliberately ignore CCBot: that's training data, not live retrieval, so blocking it doesn't stop you being cited.) - Server-rendered content — 5 points. We fetch your homepage without running JavaScript. If there's almost no text in the HTML, AI crawlers see almost nothing.
- Organization schema — 5 points. Organization, LocalBusiness, ProfessionalService or Brand JSON-LD on your homepage. This is how a machine confirms you're a real entity.
- An
llms.txtfile — 3 points. Optional, low payoff today, cheap to add. We'll flag it, but we won't pretend it's urgent.
If your homepage is unreachable when we look, we don't punish you for it — the server-render and schema checks go neutral rather than scoring zero.
Content citability · 15%
Being reachable gets you considered. Being quotable gets you cited. We crawl your high-intent pages — homepage first, then FAQs, comparisons, pricing, about, guides and product pages, up to your plan's page budget — and score five levers. This is a text analysis, not an AI judgement: it's fast, it's repeatable, and it costs you nothing.
| Lever | Share of citability | Full marks means |
|---|---|---|
| Answer-first intros | 25% | Your first paragraph is a substantive, direct answer — roughly 30 words or more — not a warm-up. |
| Statistics & data | 25% | Real figures: percentages, currency, thousands, multiples. Around five per thousand words. |
| Quotes & cited sources | 20% | A blockquote, or at least three outbound links to sources. |
| Question-shaped headings | 15% | At least two H2s or H3s that are genuine questions. |
| Content freshness | 15% | A published or modified date within the last six months. |
Three details that trip people up:
- A bare year isn't a statistic. A "© 2026" in your footer earns nothing — we only count real figures. Dates are handled by the freshness lever instead.
- "How It Works" isn't a question. A heading counts only if it ends in a question mark, or opens with an interrogative and is a real, five-word-or-longer question.
- Undated pages aren't punished. No date at all lands neutral. A date older than two years scores far worse than no date — so an old stamp is worse than none, and a current one is better than both.
The AI Visibility tab shows all five levers as their own bars, so you can see which of the five is dragging you down. The editing habits that fix them →
Off-site footprint · 15%
AI assistants cross-check. A business that only exists on its own domain looks thinner than one that turns up in reviews, discussion and video. We look for you — and for your rivals — on YouTube, G2, Trustpilot and Reddit, and we check Wikidata for an entity matching your brand name.
The score is the share of those places where you actually appear. Two honest caveats built into the maths:
- A platform where we got no data at all doesn't count either way — it isn't scored as a win.
- A platform where nobody in your market appears isn't flagged as a gap. If none of your rivals is on G2 either, being absent there isn't costing you anything.
The missing Wikidata entity is the one worth acting on: the notability bar is lower than people assume, and it's the register AI engines query directly to confirm you're real.
What we can't measure, we reweight out
A component we couldn't measure is not scored as zero — it's removed from the denominator. If share-of-voice can't run (no competitors), your score is calculated out of the remaining 55 points of weight and scaled back to 100.
This is the honest way round, but it has a consequence worth knowing about: the first check that does include share-of-voice can move your score a long way in either direction, because a 45% component just switched on. That isn't a bug and it isn't your site changing. It's the score finally measuring the thing that matters most.
The per-engine cards
Below the score sit one card per AI assistant we sampled. Each shows:
- A mention rate — the percentage of that engine's answers you appeared in.
- In X of Y answers — the raw count behind it, so you can see how big the sample was.
- Cited instead — when you appeared in none of an engine's answers, the rival it named most. That name is the most useful thing on this page.
Engines disagree, and the disagreement is the signal. Being strong in Perplexity and invisible in ChatGPT usually means you're citable but not crawlable — check whether GPTBot is blocked.
Running a check
Add a competitor
At least one, or share-of-voice can't run. More is better — the comparison prompts get sharper.
Open AI Visibility and run the check
The setup popup lets you choose the depth, how many prompts, which engines, and how many runs — all capped to your plan. The per-plan limits →
Work the recommendations
They're sorted by points gained, and the content ones open straight into BlogCraft with the topic loaded.
The first check on each brand is free. After that a check is billable, and you can run at most one per brand per day. See what a check draws on your balance on the credits page — and note that the access, citability and off-site components refresh on every check at no extra cost.