The Agent Visibility Score: How We Measure AI Visibility Across 5 Dimensions

In this article

1. Why AI visibility needs its own score
2. The five dimensions
3. How we build and run the prompt set
4. A worked example
5. Reading the score
6. From score to dollars: pipeline at risk
7. What the score doesn't do
8. Frequently asked questions

Why AI visibility needs its own score

SEO has decades of shared metrics: rankings, impressions, clicks. AI search has almost none. An assistant doesn't show a ranked list. It names two or three companies and moves on, the answer changes between runs, and most citations don't match Google's top results. Ahrefs found that only 8% of ChatGPT's in-text citations ranked in Google's top 10 for the same prompt.

So "we rank on page one" doesn't tell you whether ChatGPT recommends you, and a single screenshot of one answer tells you even less. We built the Agent Visibility Score to answer two questions every month: how often are we named compared with competitors, and what's stopping us from being named more?

The five dimensions

1. Access (20 points)

Can the crawlers behind each assistant reach your pages and read them? We check:

• robots.txt rules for 15 AI crawlers across 8 engines. Search and user-fetch crawlers (such as OAI-SearchBot and ChatGPT-User) count three times as much as training crawlers, because they decide whether you can be cited at all.
• Firewall and CDN behavior: whether verified crawlers get your content or a 403, a 429 or a challenge page (the six usual culprits).
• Readability without JavaScript on your key templates: how much of the text is in the raw HTML (why it matters).
• Indexability: noindex tags, a working sitemap, and whether key pages are in Bing's index as well as Google's.

The robots.txt and raw-HTML parts of this dimension are what our free LLM Readiness Check scores in 10 seconds.

2. Quotability (20 points)

We review your 10–20 most important pages (core services or products, industries, pricing, about, top resources) for the things that let an assistant quote you accurately:

• An answer-first summary that states what, for whom and where.
• Concrete facts: specs, certifications, capacity, lead times, locations and pricing information.
• Coverage of the narrower questions assistants search for ("fan-out" queries), which Grow and Convert observed at 2–4 per prompt.
• Real FAQs, self-contained sections, current dates and matching structured data.

The benchmark is our answer-ready page template.

3. Corroboration (20 points)

Assistants trust what several sources agree on. We check whether your company appears, with consistent facts, on the third-party sources the assistants actually cite for your category: industry directories and marketplaces, review platforms, association member lists, comparison and "best of" articles, regional business press, and forums. We also check entity signals such as LinkedIn, Crunchbase, Google Business Profile, Bing Places and Wikidata, and whether your Organization schema links to them.

4. Accuracy (20 points)

We ask each assistant a fixed set of fact questions about your company (locations, service area, certifications, products, pricing, size, ownership, contact details) and grade every claim as correct, outdated or wrong. Errors are weighted by severity: a wrong location or certification, or confusion with another company, costs more than a slightly old detail. Each error comes with its likely source, so the fix list writes itself (how we fix them).

5. Share of voice (20 points)

This is the outcome. We run 30–40 prompts your buyers would really type on all six assistants and record every company named and every source cited. Two numbers come out of it:

• Mention rate: the share of prompts in which at least one assistant names you. This drives the 20 points.
• Share of voice: your share of all vendor mentions, by assistant and overall, compared with each competitor.

How we build and run the prompt set

The prompts are the foundation, so we build them from how buyers actually ask, not from SEO keywords. A typical set of 30–40 mixes:

• Category prompts: "best MRO supplier for mid-size manufacturers in the Midwest"
• Requirement prompts: "ISO 9001 certified industrial distributor Ohio"
• Problem prompts: "how to cut PPE stockouts on a factory floor"
• Comparison prompts: "alternatives to big-box MRO suppliers for small industrial buyers"
• Local and logistics prompts: "who stocks stainless hex bolts in bulk near Columbus"
• Commercial-terms prompts: "industrial supply company with net-30 terms for new accounts"

Each prompt runs on ChatGPT, Perplexity, Gemini, Claude, Microsoft Copilot and Google AI Overviews, in fresh sessions with web search available, from the target market. We store the full answers and cited sources, so every number in the report can be traced back to an answer. For monthly re-scoring the prompt set stays fixed, so changes reflect your visibility, not a different question list.

A worked example

Here's how the score looks for the fictional distributor in our sample report, Northfield Industrial Supply (240 employees), across 40 buyer prompts and 6 assistants:

Reading the score

The bands are a guide. The trend against your direct competitors, on the same prompts, matters more than the absolute number.

From score to dollars: pipeline at risk

A visibility score alone won't get budget approved, so every score comes with an estimate your CFO and VP of Sales can check. The method, with the sample report's numbers:

1. AI-assisted vendor searches per month in your category and region, estimated from prompt-volume data for the tracked prompts: 2,400.
2. Share of those answers that don't name you, weighted by volume: 93%.
3. Buyers who would have contacted a named vendor, a conservative 3%, gives about 67 lost inquiries a month.
4. Your inquiry-to-qualified-opportunity rate, from your CRM (40%), gives about 27 lost opportunities a month.
5. Your average first-year account value ($18,000) gives about $480,000 of pipeline at risk per month.

These are estimates, not forecasts. We show every assumption, report a range in live audits, and narrow it once real AI referral traffic can be measured in GA4 and Bing's AI Performance report.

What the score doesn't do

• It doesn't guarantee recommendations. No one controls how AI systems choose answers, and we don't promise rankings or mentions.
• It isn't a traffic metric. Many AI answers never produce a click. Pair the score with AI referral traffic and "how did you hear about us?" answers on your forms.
• It depends on the prompt set. That's why we build prompts with your sales team and keep them fixed for re-scoring.

Want your own score? The free Agent Blindness Report gives you all five dimensions, competitor share of voice and the pipeline estimate within 48 hours. See the sample report first.

Frequently asked questions

Free tool

Start with your Access score in 10 seconds

The free LLM Readiness Check scores the crawler access part of the method: 15 AI crawlers, raw HTML, structured data, sitemap and llms.txt.

Run the free check