Not sure where to start? Take a guided tour.
AI Site Grader
AI Query Scanner
llms.txt GeneratorDefinition and formula
Quotable and reproducible
Definition
“AI share of voice is the percentage of a defined class of brand events captured by one brand within a fixed set of prompts, AI engines, markets, languages and runs. Unless a report says otherwise, use mention share as the default event class and publish recommendation share, citation share, visibility rate and position separately.”
Canonical mention-share formula
mention share = your brand-answer mention units ÷ all tracked brand-answer mention units × 100
- Observation unit
- One completed answer for one prompt × engine × market/language × run. Count a tracked brand at most once in that answer, even if the name appears several times.
- Denominator
- Every brand-answer mention unit earned by your brand and all declared competitors in the exact sample. Brand-free answers add no mention units, but they remain in the visibility-rate denominator.
The five measures
Do not blend them by default
| Measure | Numerator | Denominator | When it is useful |
|---|---|---|---|
| Mention share | Answers in which your brand is present, counted once per answer | Brand-answer mention units for every brand in the declared comparison set | The default competitive presence measure. It answers: how much of the category conversation did we occupy? |
| Recommendation share | Answers that present your brand as a suitable option or shortlist choice | Recommendation units for every tracked brand under the same classification rule | Use for buyer-intent prompts. A passing mention must not count as a recommendation. |
| Citation share | Citation appearances attributed to your brand under a declared domain rule | Citation appearances attributed to every tracked brand under that same rule | Use to compare evidence ownership. State whether citations mean owned domains, named-brand sources, or another rule. |
| Visibility rate | Eligible answer observations in which your brand appears | All eligible completed answer observations, including answers with no tracked brand | Use to answer: are we appearing at all? Unlike share metrics, the denominator is answers, not brand events. |
| Position or prominence | Observed order, first-place count, or an explicitly weighted position score | Present mentions, or all brands’ weighted scores when reporting a share | Use when order is measurable. Keep the raw position and weights beside any weighted score. |
Scroll to compare all columns
One blended number can hide a bad trade. A brand might win 45% of citation share because its documentation is frequently sourced, yet earn only 25% of recommendations. Averaging those values into 35% erases the useful diagnosis: the evidence is present, but the brand is not making the shortlist.
How to calculate it
Shared with the free tool
Mention share
14.6%
35 ÷ 240 × 100
The denominator contains your brand plus every other tracked brand. A zero denominator returns 0%.
Worked example
Illustrative data
The same brand can lead one measure and trail another
This example is illustrative, not a Trakkr benchmark. Forty eligible answers contain 60 tracked-brand mentions, 24 classified recommendations and 20 citations attributed under an owned-domain rule.
AI visibility audit
A repeatable operating workflow
- 01
Choose the prompt set
Start with 20 to 50 prompts that reflect real category, use-case, comparison and problem questions. Keep a smaller priority set for weekly checks. Version the library when prompts change.
- 02
Declare the comparison set
Include your brand and the alternatives a buyer would genuinely consider. Publish the list. Adding or removing a competitor changes the denominator, so start a new comparison series when the set changes.
- 03
Choose engines and surfaces
Measure each surface separately before any roll-up. A Google AI Overview observation is eligible only when an Overview appears. A chat observation is eligible when the engine returns a completed answer.
- 04
Fix market and language
Record country, language, device or account context when it can affect the answer. Do not compare a US English run with a UK English run as though they were the same observation.
- 05
Capture the event fields
For every prompt, engine, market and run, store completion status, brands mentioned, recommendations, source links, mention order and the raw answer. Keep missing data distinct from a checked absence.
- 06
Set frequency and thresholds
Run priority prompts weekly and the full audit monthly. A practical starting rule is at least 95% sample completion, then investigate a 5 percentage-point move only when it persists for two weekly runs. These are operating defaults, not universal statistical thresholds.
- 07
Decide from prompt-level evidence
Inspect the prompts and engines behind every material move. Act when the same loss pattern repeats, then record the content, PR or product change so the next audit has a clear hypothesis to test.
Your first benchmark is structural, not universal: equal share is 100 ÷ the number of tracked brands. With five brands, that reference is 20%. Use it as orientation only. The more useful benchmark is your stable trend by engine, prompt group and market.
Engine benchmark
2025-08-12 to 2026-03-11
Tracked-brand appearance rate by engine
The share of valid analyzed prompt responses in which an engine returned at least one tracked brand recommendation. This is an engine-coverage benchmark, not market share and not a target for your brand.
- Prompts analyzed
- 825,343
- Valid comparisons
- 797,644
- Model responses
- 6,439,133
- Average model agreement
- 43.3%
- Perfect agreement
- 4.0%
Sample: matched prompts with sufficient coverage across eight tracked model families. Geography, raw prompts, raw responses, brand-level rows and exact model versions are not published.
Citation benchmark
2025-10-03 to 2026-10-14
The most-cited domains in the verified public slice
The current snapshot contains 80,769,585 citation appearances across 725,777 domains and 1,928 tracked brands. The prior page total was stale and has been replaced with this verified snapshot.
The three leading domains account for 5.8173% of citation appearances in this snapshot.
Methodology and limits
What the figures can support
Citable methodology fragment
“The public Citation Index sums citation appearances from the latest valid snapshot for each of 1,928 tracked brands. It covers 3 October 2025 to 14 October 2026. The tracked panel is not a random sample or a census of all AI answers. The aggregate does not retain engine, geography, language, answer-level denominators or customer identity. The downloadable domain slice includes only aggregate rows seen across at least three contributing brands; raw URLs and customer data are excluded.”
- Data period
- 2025-10-03 to 2026-10-14
- Observation counted
- Sum of citation appearance counts in the latest valid snapshot for each tracked brand.
- Sample rule
- Latest valid citation snapshot per tracked brand. Counts are appearance counts within those snapshots, not a census of all AI answers.
- Engines
- The public aggregate does not retain provider-level attribution, so model comparisons and response-level citation rates are unavailable.
- Markets
- Prompt and response geography is not retained in the public aggregate.
- Excluded
- Raw prompts, Raw responses, Customer identities, Account identifiers, Model-level denominators, Geography
- Publication rule
- Downloads include aggregate domain rows seen across at least 3 contributing brands. Raw URLs and customer data are excluded.
- Main limitation
- Not unique citations and not a census of all AI answers. A URL can appear more than once.
Put the audit into operation
Keep the denominator stable
Questions, answered.
Visible FAQ
AI share of voice is the percentage of a defined class of brand events captured by one brand within a fixed set of prompts, AI engines, markets, languages and runs. Unless a report says otherwise, use mention share as the default event class and publish recommendation share, citation share, visibility rate and position separately.
Measure the shares separately, then act on the evidence
Keep the prompt set, comparison set and classification rules fixed. When a metric moves, inspect the answers that moved it before you change the strategy.
Return to the definition