Skip to main content
LLM Metrix
Analytics · Scoring

One number for your
AI visibility.

The Visibility Score blends mention rate, answer position and sentiment into a single 0–100 metric. Watch it move as you ship GEO improvements.

Free forever plan. No card required

app.llmmetrix.com/dashboard
Illustrative, sample figures in the product's real layout
Visibility Score
84+36
AI Visibility Score over the last 30D. The underlying data follows as a table.
AI Visibility Score over the last 30D
PeriodScore
Sample 148/100
Sample 252/100
Sample 355/100
Sample 458/100
Sample 562/100
Sample 665/100
Sample 771/100
Sample 874/100
Sample 978/100
Sample 1084/100
Mention Rate
68%
+9%
Sentiment
71
+4 pts
Position
#2
up 3 places

Sentiment is a 0–100 mean of how each answer described you, not a signed score: 60 is every answer neutral, 100 every answer positive, 0 every answer negative.

Works with the AI engines your customers use

ChatGPTPerplexityGeminiClaudeGrokMeta AIDeepSeekGoogle AI OverviewsMicrosoft Copilot

Why it matters

One score your whole team can act on.

A single source of truth

Stop juggling 12 metrics. One score that the whole team, exec to engineer, can rally around.

Trend over time, 7 / 30 / 90 days

See exactly how your score moves over 7, 30 and 90 day windows with sparklines and breakdowns, as far back as your plan's retention window keeps.

Composited from real signals

Mention rate, answer position and sentiment, weighted 50%, 35% and 15% into a single 0–100 score.

Per-engine breakdown

Drill into each engine to see exactly where the score is being earned or lost.

Benchmarked vs competitors

See your score next to the competitors you track, so one number tells you where you stand in the category.

Track the impact of your work

Watch the score respond over the scans after you ship changes, so you can gauge what's moving the needle.

Mechanics

From answers to one number.

  1. 01

    Engines answer the same prompts

    Every enabled engine receives each active prompt in one scan. One credit covers the prompt across all of them, and errored calls drop out of every downstream figure rather than scoring as zeros.

  2. 02

    Each answer is graded

    An analyzer reads the full text and returns whether the brand was named, its one-based place among all the options that answer listed, and the tone around the mention.

  3. 03

    Signals combine into one number

    Mention rate at half weight, placement at about a third, sentiment the remainder form a per-engine score. The headline figure combines those by delivered-answer counts and rounds to a whole number.

  4. 04

    The reading is frozen and plotted

    The score writes onto the scan row, immutable afterwards, and plots per completed brand scan in the home market, oldest first, competitor-subject and other-market rows stay off your line.

  5. 05

    Movement is compared, not assumed

    Each scan is compared with the previous one: a score move past your threshold, a competitor gaining share of voice, sentiment sliding. Thresholds are editable and every rule has an off switch.

Who it's for

Who runs on the score

CMO reporting one AI KPI to the boardThe deck needs one AI number next to SEO traffic.

You present marketing results monthly and need AI visibility expressed the way the rest of the pack is, one figure, one direction, one line.

  • The score trend plots one point per completed scan, oldest first, ready to lift into the board pack.
  • Competitor scans read back score-for-score, so the number arrives with context rather than alone.
  • Token-gated report links put the trend in front of people who will never log in.
  • The print view renders the scan as a branded document with the headline figures on it.
  1. 01Run a scan so the trend line has its first point.
  2. 02Open Reports and lift the score chart into the board pack.
  3. 03Send non-login readers the token-gated link instead of screenshots.

Metrics this role tracks: Visibility score · Share of voice · Mention rate

Open Reports
GEO lead proving the sprint moved the numberYou ship schema and content fixes and must show the delta.

You run GEO improvements in sprints and need a defensible before-and-after reading rather than an impression.

  • Read the score across the scans either side of each change, and expect movement on the engines’ refresh timescale, not on your deploy.
  • Rule out measurement confounders first, a scan that delivered fewer engines or prompts moves the line on its own.
  • Recommendations from the same scan arrive tagged critical through low, sorted by that tag, each carrying the finding behind it.
  • Generated artefacts, an llms.txt file, schema JSON-LD, a content brief, come off the recommendation board with copy and download.
  1. 01Work the critical-tier rows on the recommendation board.
  2. 02Generate the content brief from a content-gap row and ship it.
  3. 03Compare the score across the scans either side of the deploy before claiming the delta.

Metrics this role tracks: Visibility score · Mention rate · Average position

Open Recommendations
Founder benchmarking against the category leaderYou want the gap to the leader, not a vanity score.

You are up against named rivals and judge every metric by relative position rather than absolutes.

  • Scan a competitor as a first-class subject and read their score against yours on identical prompts.
  • Share of voice per engine shows which engines favour whom in the scan you open.
  • Each tracked rival’s share of voice is a line over time, so a squeeze builds visibly before the score absorbs it.
  • Once five paid workspaces in your industry and region opt in, the dashboard card ranks you among them as a percentile with a peer count.
  1. 01Add the category leader on the Competitors page and scan them as a subject.
  2. 02Read their score against yours on the identical prompt set.
  3. 03Watch each rival’s share-of-voice line build over the first few scans.

Metrics this role tracks: Visibility score · Share of voice · Mention rate

Open the Competitors page
Analyst who audits the measurement itselfYou will not report a number whose arithmetic you cannot reconstruct.

You have been burned by opaque vendor scores before and need to rebuild the figure yourself before anyone else sees it.

  • The formula is short and published: mention rate at half weight, placement at about a third, sentiment the remainder, combined per delivered-answer counts.
  • Errored answers are excluded from every term, so an outage cannot masquerade as a score drop.
  • Alert thresholds are editable in Settings → Notifications, so what counts as a move is your decision, not ours.
  • The scan on screen exports to CSV or JSON when the spreadsheet is the deliverable.
  1. 01Rebuild the figure by hand from the published weights on one delivered scan.
  2. 02Set your own move threshold in Settings → Notifications.
  3. 03Export the scan to CSV or JSON and check the arithmetic offline.

Metrics this role tracks: Visibility score · Mention rate · Average position · Sentiment score

Agency lead putting one comparable number in every client deckForty clients, forty decks, one metric that means the same thing in each.

You report across a portfolio of brands and need a single like-for-like AI figure in every review, whatever the client sells.

  • The same formula computes every client’s score from their own delivered answers, so the numbers are comparable across decks rather than re-derived per account.
  • A token-gated report link per client puts the reading in front of them without buying a seat or mixing accounts.
  • Scan a client competitor as a subject and the deck carries context, the score beside the rivals they actually name.
  • Print any client’s scan as a branded document when the deliverable leaves the screen.
  1. 01Give each client its own project so scores never blend.
  2. 02Share each client’s token-gated link from the Reports page after the monthly scan.
  3. 03Benchmark each brand against its named rival before the review call.

Metrics this role tracks: Visibility score · Share of voice · Mention rate · Sentiment score

Open Reports

The formula

How is the Visibility Score calculated?

Three signals, weighted 50 / 35 / 15. Mention rate is the share of delivered answers that name you at all. Placement converts where you sat among the brands that answer named into a factor: first place scores 1.0, second 0.8, third 0.6, fourth through seventh 0.4, and eighth or later, or named with no readable order at all, 0.25. Sentiment maps positive answers to 1, neutral to 0.6 and negative to 0.

The denominators differ, deliberately. Mention rate is measured over every delivered answer, including the ones that never named you. Placement and sentiment average over the mentioning answers only, so an absence cannot be double-counted against you. With zero mentions the sentiment term is forced to 0 rather than inheriting the neutral default: a brand no engine names scores 0, not a floor of nine.

One worked example: named in every delivered answer, mid-list, neutrally, the arithmetic lands at 73, ubiquity alone reaches the top bands without a single first mention. Per-engine scores then combine into your headline figure weighted by how many answers each engine actually delivered, so one flaky engine answering twice cannot move the number like an engine answering twenty times. The result rounds to a whole number and freezes onto the scan when it is written.

  • Citations, competitor mentions and accuracy findings are recorded on the same answers but are not score inputs.
  • Errored answers drop out of every term, so a provider outage never depresses the score.
  • The same scoring function grades deep-research reports, so the two surfaces stay comparable.

Context

What is a good AI visibility score?

There is no universal pass mark, and a tool that publishes one is selling confidence rather than measurement. The same figure means different things depending on category density, five serious players splitting the naming produces lower leaders than a niche with one obvious answer, and on your query set, because broad informational prompts behave differently from commercial ones.

Judge the number three ways instead. Against your own trend across several scans, since consecutive readings of an unchanged site still vary. Against the competitors you track, scanned as subjects on identical prompts. And against your priority queries specifically, where a modest overall score can hide dominance on the handful that drive revenue.

For cohort context, an opt-in industry index ranks your latest score among contributing paid workspaces in the same industry and region once at least five have opted in. You see a percentile and a peer count, never another workspace’s score, and buckets below five participants are never published.

Movement

Why did my score change between scans?

Start with measurement. The score moves whenever a scan delivers different inputs: an engine added or dropped, a prompt set edited, fewer answers coming back. The scan row records the mode each reading ran under, grounded or not, which market, which language, because mixing instruments on one trend line is change that looks like movement and is not.

Beyond that, attribution is probabilistic and the product says so: nothing in it records what you shipped, so pair the chart with your own log of content and schema changes rather than expecting the graph to name a cause. Generative answers also vary run to run, which makes a few points of movement between scans the measurement rather than the market. Judge direction over several scans.

  • A move of five or more points against the previous scan raises the score-change alert by default; the threshold is editable, with immediate, digest and off modes.
  • An absolute floor alert fires on every scan while the score sits below a crisis level, a standing condition, not a one-off change.
  • Free workspaces keep seven days of scan history, so a ninety-day trend window fills only on plans with longer retention.

FAQ

Visibility Score questions, answered.

How is the Visibility Score calculated?
It is a weighted composite of mention rate, answer position (first vs fine-print) and sentiment, expressed on a 0–100 scale. Mention rate carries half the weight, position about a third and sentiment the remainder, 50%, 35% and 15%.
Is the score comparable across companies?
The score uses the same formula for every brand, so it is calculated consistently across companies. It reflects raw visibility signals rather than being normalised against an industry cohort. Compare it against the competitors you track for category context.
How quickly does the score react to changes?
The score updates every time a scan runs, automatically every day on paid plans and once a week on free, plus any scan you run yourself. How fast the engines themselves reflect your content changes depends on the engine, but you will see the score move on the scans after you ship updates.
How is this different from answer-engine ranking?
Answer-engine ranking tags where you sit inside each individual AI answer, first, prominent, mid-list, fine-print or absent. The Visibility Score compresses those signals into one number per scan: how often you were mentioned, how high you placed when you were, and how positively. Use the ranking views to debug a score; report the score to summarise the rankings.
Why did my score drop?
First rule out measurement: the score moves whenever a scan delivers different inputs, so a run on fewer engines or fewer prompts moves the line for that reason alone. Beyond that, treat attribution as probabilistic, nothing in the product records what you shipped, so pair the trend with your own log of content and schema changes rather than expecting the chart to name a cause.
Can I benchmark my score against my industry?
On paid plans you can opt your workspace into an anonymised industry index. Once at least five contributing workspaces in the same industry and region have opted in, the dashboard card ranks your latest score among them, you see a percentile and a peer count, never another workspace's score. The public industry-benchmarks page is a separate league of well-known brands, not customer percentiles.
Which engines feed the score?
Every engine your project tracks contributes answers to each scan, and the score is computed from those answers. Free monitors a fixed four engines; paid plans choose any four of the seven. ChatGPT, Perplexity, Gemini, Claude, Grok, Meta AI and DeepSeek. Google AI Overviews, Google AI Mode and Microsoft Copilot run on the weekly SEO-data pass for paid plans and are billed separately rather than through credits.
Start on the free plan. No card, no time limit.

The AI era of search
is already here

See what AI engines say about your brand before your competitors do. Start free today. No card required.