AI Visibility Score

What Pendium's 0–100 AI Visibility score measures, how a scan works across ChatGPT, Claude, Gemini, and Google AI Overviews, and how to read the results.

Pendium measures how AI assistants — ChatGPT, Claude, Gemini, and Google AI Overviews — perceive and recommend your brand. A visibility scan asks each platform the kinds of questions your buyers actually ask, reads the answers, and turns them into a single AI Visibility score from 0 to 100, plus the detail behind it.

This page explains what the score means, how a scan is built, and how to interpret the results.

Running a scan is a write plus LLM spend, so an account in read-only mode can't trigger one — it can still view existing scores and reports.

The score at a glance

The AI Visibility score is a 0–100 summary of how present your brand is in AI answers across the platforms scanned. Higher means AI assistants mention and recommend you more often, and more prominently, when people ask about your category.

Pendium maps the score to a level so you can read it at a glance:

Score	Level	What it means
80–100	Excellent	AI assistants reliably surface and recommend you in your category.
60–79	Good	You show up often, with room to win more queries.
40–59	Moderate	You appear sometimes, but competitors are named more consistently.
20–39	Low	You're rarely mentioned when buyers ask about your category.
0–19	Invisible	AI assistants almost never bring you up.

What the score measures

For every question in a scan, Pendium checks the answer from each AI platform for two things:

Mention — did the assistant name your brand at all? This is the biggest driver of the score. If AI doesn't mention you, nothing else matters.
Position — when you are named, how near the top of the list are you? Being the first recommendation counts for more than being the fifth.

A query's score combines those two signals — mention rate across platforms, plus a prominence bonus for ranking high. Sentiment (whether the mention is positive, neutral, or negative) is captured per mention and surfaced in the report.

For the exact formula — how mentions, citations, and ranking position are weighted, and why a citation counts for less than a recommendation — see How the score is calculated.

Buyer-intent queries set the headline

Not every question matters equally. Pendium tags each query with a reach level, from direct buyer-intent down to broad curiosity:

Reach level	What it captures	Example
Core	Direct, buyer-intent queries in your exact category	"best project management tools"
Adjacent	One step out — neighboring needs where you're a plausible answer	"how do I keep my team's work organized?"
Aspirational	Broader industry and thought-leadership themes	"the future of remote work"
Visionary	Loosest relevance, top-of-funnel curiosity	"how is AI changing productivity?"

Your headline score is the score of your core, buyer-intent queries. Adjacent, aspirational, and visionary queries are still scanned and scored — you'll see them in the per-reach-level breakdown — but they don't move the headline number. That keeps the score anchored to the question that matters: "when someone is ready to buy in my category, does AI recommend me?" (If a brand has no core queries at all, the headline falls back to a blend across whatever levels ran.)

Three different scores

When you run a preview / brand-page scan, you'll see three scores. They measure different things — don't confuse them:

AI Visibility (0–100) — discovery. "When AI helps someone in my category, does it recommend me?" Driven by how often you're named in answers to category and comparison questions, relative to competitors.

Direct-Brand Knowledge (0–100) — understanding. "When someone asks AI about my brand by name, how much does it actually know?" This comes from asking the assistant directly about your company and grading the depth and accuracy of what it knows:

Knowledge score	Level	What it means
60–100	Strong	AI has a clear, confident, fact-rich picture of your brand.
30–59	Partial	AI can describe you accurately but lacks depth.
10–29	Thin	AI knows you exist but can't say much.
0–9	Unknown	AI effectively has nothing to go on.

AI Sentiment (0–100) — regard. "When AI talks about my brand, how favourably does it talk?" This is about tone, not reach or depth:

Sentiment score	What it means
80–100	AI would actively recommend you and can point to specific outside endorsement.
60–79	AI speaks well of you and would put you on a shortlist.
40–59	AI describes you factually, with no praise and no criticism.
20–39	AI raises real caveats alongside anything positive.
0–19	AI would steer a buyer away, and unresolved complaints dominate.

Most brands sit in the 40–59 band, and that is a neutral result rather than a poor one. A business nobody has written about scores there by design: an absence of praise is not criticism, and sentiment is deliberately not lowered for brands that simply have no coverage. Knowledge score is the field that carries "AI has nothing to go on."

A brand can be well-known by name (high knowledge) yet rarely recommended in its category (low visibility), or vice-versa. It can also be barely known and still spoken of warmly by the few sources that mention it. Each gap is worth closing, and they call for different work.

How a scan is built

A scan turns your brand's strategy into questions, asks them across platforms from multiple buyer perspectives, and aggregates the results.

Topics and queries. Pendium organizes the questions into topics (themes like "Core Product" or "Competitor Comparisons"), each holding a set of queries (the actual questions asked). Every query carries a reach level (above).

Personas. A scan runs queries from the perspective of different buyer personas (for example, "Technical Startup Founder" vs. "Enterprise CMO"). The same question gets different answers depending on who's asking, so personas reveal where you're strong with one audience and weak with another.

Platforms. Every standard scan covers four AI surfaces:

ChatGPT (OpenAI)
Claude (Anthropic)
Gemini (Google)
Google AI Overviews (the AI answer box in Google Search)

Each platform gets its own breakdown — score, mention rate, and sentiment — so you can see, for instance, that you're strong in ChatGPT but invisible in AI Overviews. If a scan didn't query a given platform (a scoped or single-engine run), that platform's breakdown reports as not queried rather than a 0 score — a platform AI hasn't been asked about yet is unscored, not failing.

Coverage. The bigger lever you control is how many queries a scan runs — more queries means broader coverage. Twenty to forty is a good range; thirty is a sensible default. The MCP and REST API also accept a mode parameter (default batch); both batch and full scan all four platforms with fast, cost-effective models.

Grounding. When the URL you scan is a local-business listing (a Yelp business page, or a Google Business Profile / Google Maps listing) or an e-commerce storefront (a Shopify store), the preview scan grounds its analysis in the real signals it can read there — for a local business, that's your rating, categories, services, hours, and the themes customers praise; for a store, your catalog and best-sellers. Grounding makes the buyer questions and the brand profile match the business AI assistants actually see, so the score reflects your real footprint rather than a guess from a thin homepage. Non-listing URLs are scanned exactly as before.

Who we decide you are. Scans read the structured data most sites already publish about themselves: the business name, address, phone, and category, stated in a machine-readable block on your own pages. Those details settle any disagreement with what a web search turns up. This matters when a similarly named business exists at a nearly identical web address, because search engines quietly correct the spelling and hand back the other company. When your site states its address, that address is the one we use.

Separately from grounding, every scan classifies whether the business is local, judging the business model (does a customer have to be nearby to buy?) rather than the seed URL or the city named in its market. So a plumber, clinic, or restaurant scanned by its own website gets the local reading even with no directory listing to ground against, and a national brand headquartered in one city does not.

What a scan returns

Beyond the headline score, a completed scan gives you:

Per-platform scores — how each AI surface sees you. A platform the scan didn't query comes back unscored (not a 0) — treat those as "not tested," not as a real result.
Per-persona scores — where you win or lose by audience.
Query-level detail — for each question: were you mentioned, where did you rank, what was the sentiment, and which platforms answered.
Top competitors — the brands AI named alongside or instead of you, ranked by how often they came up across the scan.
Cited sources — the URLs the assistants leaned on when answering, flagged by whether they mention you. These are the pages shaping AI's view of your category — and the ones worth earning a mention on.
Recommendations — prioritized actions to improve where you're weakest.

Summaries, headlines, and recommendation copy in the report follow Pendium's plain-English content standards: concrete, specific, and free of AI-sounding jargon.

Running and reading a scan

In the product — run a scan from your dashboard and read the full report there.
Via an AI agent (MCP) — use scan_visibility to trigger a scan, get_scan_status to poll, and get_report for the full breakdown. get_cited_sources lists the sources, and get_recommendations returns the action items.
No-auth lookup — check any brand's public score with lookup_brand_score (MCP) or the score REST endpoint.
Over HTTP — trigger a scan, poll it, and fetch the report via the REST API.

Re-scan after you publish new content or earn mentions to see the score move — get_scan_history tracks the trend over time.