How Blizo measures AI visibility

The data, the formulas, and the guardrails behind every number in your dashboard.

How tracking works

Every update cycle, Blizo asks a curated set of real buyer questions — your prompts — to each AI platform we track: ChatGPT, Gemini, Claude, Perplexity, Grok, and Google's AI Overviews and AI Mode. We record the full answer, the brands it mentions, the order they appear in, and every web source it cites.

For ChatGPT and Gemini, we measure the actual consumer web experience through a geo- and language-specific browser scraper. Google's AI surfaces are measured directly from their search results. Claude, Perplexity, and Grok use their native web-grounded model routes as repeatable proxies. Results can still differ from any single session because AI answers are non-deterministic by nature.

Disclosure: ChatGPT and Gemini numbers come from their consumer web interfaces; Claude, Perplexity, and Grok use web-grounded API proxies. Every stored payload records its measurement source and method version.

The metrics

Visibility rate

Of the answers where a platform actually responded, the share that mention your brand. The denominator is answered runs — questions that produced no answer are excluded so a quiet day never looks like a loss.

visibility = mentions ÷ answered runs

Average position

When your brand is mentioned, how early it appears among all brands named in the answer (1 = named first). Lower is better. Counted only on generic prompts — prompts that name your brand or a competitor guarantee that entity a top mention, which would skew the average. We only report it once there are enough mentions to be meaningful.

avg position = Σ position ÷ mentions

Share of voice

Your brand's mentions as a share of all mentions across your brand and its tracked competitors — how much of the AI conversation you own.

share of voice = your mentions ÷ all tracked mentions (organic prompts only)

Citation rate

Of answered runs, the share where the AI cited a page on your own domain as a source. Being named is good; being cited as the source is better.

citation rate = cited runs ÷ answered runs

A brand that appears only as a cited source URL counts toward citation rate, not as a mention — a mention means being named in the answer text itself. On the Sources page, citations of a tracked brand's own domain are excluded when the prompt names that brand, since an AI citing a brand's site while being asked about that brand proves nothing.

Blizo Score

A single 0–100 headline that blends the three signals that matter, weighted toward being mentioned at all, then being cited, then ranking early:

Blizo Score = 0.60 · visibility + 0.25 · citation + 0.15 · position factor

The position factor rewards appearing earlier and is zero when you're unmentioned — so the score is meaningful even at zero visibility, and stays a clean baseline to grow from.

Sentiment

How positively AI answers describe your brand, read across every answer that mentions it. Each mention is classified as positive, neutral or negative; the balance of those reads gives the overall tone.

net sentiment = (positive − negative) ÷ mentions

With only a handful of mentions we don't claim a read — the tag appears once there's enough signal, and a genuine split between positive and negative shows as "Mixed" rather than averaging out.

Head-to-head win rate

How often your brand is ranked at least as high as every competitor named in the same answer. Ties count as wins. Counted only on generic prompts: when a prompt names your brand or a competitor, that entity's top spot is pre-decided by the prompt, not won — those answers show a sentiment comparison instead.

win rate = wins ÷ comparable answers

An answer is comparable only when your brand and at least one competitor are both ranked in it — answers that name no ranked competitor are never counted against you.

Perception: sentiment & head-to-head

Perception reads every tracked prompt (active and paused; archived prompts are excluded) — including prompts that name your brand. Asking an AI directly about a brand is exactly where tone shows, so unlike the organic competitive metrics (Share of Voice and the visibility headline, which exclude branded prompts), nothing is filtered out here. Sentiment and win rate answer "how does AI talk about you when you come up", not "how often do you come up".

A win rate is only displayed once the window holds at least 3 comparable answers — one comparison is an anecdote, not a rate. Per-competitor records on the Perception page follow the same rule, and negative mentions are backed by the verbatim quote where one was captured.

Recommended actions

The Actions page is derived from the answers Blizo tracked in the last 30 days — deterministically. Fixed rules run over stored data; no generative AI writes, ranks or embellishes them, so an action can never claim something the underlying answers don't show. Every action carries its evidence verbatim: the quotes, counts and domains that triggered it.

Content gap — a tracked prompt has modeled fan-out sub-topics where your site is absent or mentioned negatively. Priority follows the highest sub-topic priority score.
Competitor-owned prompt — your brand appears in none of a prompt's answers (at least 5 answered runs) while a competitor appears in at least 40% of them. Organic prompts only.
Citation gap — the domains AI cites most in answers where your brand is absent, once there are at least 5 such answers. Your own domain and competitor domains are excluded.
Reputation risk — any negative brand read in the window (1+), or 20%+ negative across 3 or more reads. Always ranked high: reputational damage compounds, so every negative is actively worked on.
Losing head-to-head — a prompt with at least 3 comparable answers where you win fewer than half.
Competitor cited — a tracked competitor's own domain is cited as a source in at least 1 answer(s) to generic prompts (prompts naming neither your brand nor a competitor). The cited pages are listed verbatim as the content to match.

The list shows at most 10 actions, capped at 3 per rule type, sorted by a 0–1 priority score that maps to the High / Medium / Low labels.

Answer presence on Google surfaces

Google's AI Overviews and AI Mode don't appear for every query. When no AI answer is generated, we record answer presence separately and keep that run out of your visibility denominator — so "Google didn't show an AI answer today" is never mistaken for "your brand lost visibility."

Why we use rolling windows, never a single run

The same question asked twice can return different answers. Blizo never reports a metric from one run. Rates are computed over rolling multi-day windows, and when a window has too few answers to be trustworthy we show "Collecting data" instead of a misleading number. This is why weekend or brand-new projects surface as collecting rather than a hard zero.

Below 5 answered runs we don't report a rate; average position needs at least 3 mentions.

Questions about the methodology? Every formula above lives in one place in the codebase and is unit-tested — the dashboard never re-derives a rate in the UI.