The AI Visibility Index Methodology: How TopSlot Measures AI Visibility

Anyone can open ChatGPT, type a question, and screenshot whether their brand shows up. That is not measurement — that is an anecdote. The moment you ask it again tomorrow, or on a different model, or phrased slightly differently, the answer changes. A single lucky mention tells you nothing about whether buyers actually see you.

The AI visibility index is TopSlot's answer to that problem. It is not a per-user number you can screenshot — it is the measurement system underneath the number: the branded engine, the query design, the scoring model, and the credibility layer that together decide how AI visibility gets quantified. Think of it the way FICO is a scoring model, or the way an S&P index is a methodology rather than a single stock price. The AI Visibility Index is how we measure; the AI Visibility Score is the 0–100 headline number a brand gets out of it.

This piece is the methodology in full. If you want to trust the number, you should understand exactly how it is produced.

Why a methodology matters more than a number

Every AI-visibility tool on the market can produce a number. The question that separates a credible index from a vanity metric is: would you get the same answer if someone else ran it, or if you ran it again next week? That is reproducibility, and it is the entire game.

AI models are probabilistic. Ask the same question twice and the wording of the answer shifts. Change one word in the prompt and a brand can appear or vanish. A methodology exists to control every one of those variables so that when your Score moves, the movement means something real about your visibility — not noise from a re-roll of the dice. The AI Visibility Index is built around four design decisions that make the number defensible.

Decision 1: Four models, not one

Buyers do not all use the same assistant. Some live in ChatGPT, some ask Gemini inside Google, some trust Perplexity for cited answers, and some use Claude. A brand can be the default recommendation in one and completely absent from another.

The Index queries four models — ChatGPT, Gemini, Claude, and Perplexity — for every audit, then composes a single Score from the results while preserving the per-model breakdown underneath. That breakdown is diagnostic, not decorative. A brand scoring well everywhere except Perplexity has a freshness-and-citation gap; a brand scoring flat-low across all four has a base-authority problem. Same Score, different fixes. Measuring one model would hide exactly the signal that tells you what to do next.

Decision 2: Zero-brand-name, buyer-intent queries

This is the single most important design choice, and it is where most tools quietly cheat.

If you ask an AI "tell me about Acme CRM," it will happily describe Acme — because you named it. That measures nothing except the model's ability to autocomplete a name you handed it. It is a mirror, not a measurement.

The AI Visibility Index uses a zero-brand-name methodology: every audit query is a generic, buyer-intent question that never mentions your brand, a competitor, or your domain. Questions like "what's the best CRM for a 12-person sales team?" or "which project tool should a remote agency actually pay for?" This is what real buyers type. The result reflects what they genuinely see — whether the model volunteers your name unprompted, or names five competitors and skips you.

Enforcing that discipline is non-trivial. Brand names leak into queries in subtle ways — near-misspellings, domain stems, product-line nicknames. TopSlot runs a multi-layer brand-safety classifier that filters every query before it reaches a model, so a category question never accidentally becomes a branded one. Zero-brand-name is not a slogan; it is enforced at the pipeline level. It is also the reason the Index maps cleanly onto emerging disciplines like generative engine optimization and answer engine optimization — both of which are about earning the unprompted mention, not the prompted one.

Decision 3: Intent-weighted scoring

Once the answers come back, they have to be turned into a number — and not every mention is worth the same.

Being named when a buyer asks a high-intent "which should I buy / best X for me" question is the most valuable placement there is. That buyer is at the checkout stage of their thinking. A broad category-awareness mention — showing up in "what is a CRM?" — matters, but far less. An evaluation or comparison mention sits in between. The AI Visibility Index reflects this reality by weighting mentions according to the buyer intent behind the question that produced them.

Here is how the factors stack, qualitatively:

• Mentions, weighted highest and scaled by buyer intent. Whether you appear at all is the dominant factor — and appearing in a high-intent purchase question counts far more than a broad awareness mention. This is the bulk of the Score.
• Prominence. Where you land in the answer. First sentence and top of the recommended list beats seventh item in a footnote.
• Citations. Does the model link your URL, or just say your name? A cited answer is a stronger signal of authority.
• Sentiment. Is the framing positive, neutral, or negative? A smaller refinement on top of the rest.

We deliberately publish the factors, not the exact coefficients — the same way Moz, Ahrefs, and FICO describe what drives their models without printing the formula. The weights are proprietary and they get recalibrated; the priority order does not. Mentions scaled by intent lead, then prominence, then citations and sentiment as refinements.

Decision 4: Honesty gates and read-stability

A methodology earns trust by what it refuses to claim. The AI Visibility Index has two protections built in specifically to stop it from lying to you.

Honesty gating. When the sample behind a Score is thin, the Index does not fake precision. It reports a band — "Strong Visibility" — alongside an explicit insufficient data signal, rather than a false-precise "67.3." And hard caps apply: zero mentions, a very low mention rate, or zero citations cap the Score no matter how the other factors land. You cannot buy your way to a high number on prominence and sentiment while the model never actually names you. The gates make the number honest before they make it flattering.

Read-stability. Because models are probabilistic, a single query run is a coin flip. The Index is designed to read the same underlying visibility consistently — so day-to-day movement reflects a real change in how AI sees you, not a re-roll. That stability is what makes the AI Search Tracker meaningful: when it snapshots your Score on a schedule and the line moves, you can trust the movement is signal. A wobbly, un-stabilized metric would make trend lines meaningless and every alert a false alarm.

Where the Index fits

The AI Visibility Index is the measurement engine. From it flows the AI Visibility Score — your 0–100 headline number — with score bands running from 0–10 (AI doesn't know you) through 51–70 (strong) up to the 85+ range that only category-definers reach. Around it sit related gauges: the AI Authority Rating is an off-site authority driver — referring domains and citation authority — that influences visibility without being a mathematical component of the Score, and Brand Health tracks operational readiness. But the Index is the spine: the reproducible, four-model, zero-brand-name, intent-weighted, honesty-gated system that decides what "visible" actually means.

If you are new to the number itself, start with what an AI Visibility Score is, or see how the Index differs from traditional SEO rankings — then come back here for how it is produced.

See the methodology on your own brand

Reading about a methodology is one thing; watching it run against your category is another. The fastest way to understand the AI Visibility Index is to run it. A free AI Visibility Scorecard fires real buyer-intent, zero-brand-name queries at live models and shows you your current band in about a minute. If you want the four-model, scheduled, read-stable version with trend lines and alerts, that is what the AI Search Tracker is built for.

The number is only as good as the method behind it. Now you have the method.