How Wombid measures local AI visibility
The method behind the 0 to 100 score. Published so anyone can audit the measurement, not just the marketing.
In plain terms
We ask the same kinds of questions a real customer would ask, on the five named AI engines plus Google Maps, for the trade and suburb on the audit. Then we count whether those answers name the business.
- The score is a snapshot of one run, not a forecast of jobs.
- A missing Google AI Overview is left out of the score. It is not treated as a miss.
- Two scores can be compared only when the method has not changed between them.
- No one can guarantee that an AI engine will recommend a business.
The score does not predict calls, bookings or revenue. It measures whether named engines recommended the business for the questions we asked, on that run.
What we check
| Surface | What we record |
|---|---|
| ChatGPT | Named businesses in a live, web-grounded answer |
| Gemini | Named businesses in a live, search-grounded answer |
| Google AI Overviews | Named businesses in the real AI Overview box when Google shows one for that query. If Google shows no Overview, the cell is n/a |
| Claude | Named businesses in a live, web-grounded answer |
| Perplexity | Named businesses in a live answer |
| Google Maps | Local-pack positions for the same questions, from Australian mobile search. Maps is evidence beside the score, not a sixth scoring engine |
Google search, AI Overviews and Maps are read as Australia on a mobile device, in English. That is the coverage an Australian customer actually sees.
Organic Google rankings for the same questions are included only when a website address is on the audit. Without a website, the AI surfaces and Maps still run. A missing website is recorded as not measurable, never guessed from titles.
The six buyer questions
Every trade audit asks six questions in this shape, filled with the trade label and suburb:
- Who is the best [trade] in [suburb]?
- Can you recommend a reliable [trade] near [suburb]?
- I need a [trade] for [typical job] in [suburb]. Who should I call?
- Which [trade] in [suburb] has the best reviews?
- For emergency trades: Emergency [trade] in [suburb], who can help fast? For planned trades such as builders: Which [trade] in [suburb] can start soon?
- Affordable [trade] for [second typical job] around [suburb]?
The question text is part of the run's identity. Changing a question makes the new run not comparable with the old one, even if the score epoch is unchanged.
How the 0 to 100 score is calculated
Each engine-and-question cell is a mention, a miss, or not applicable. Mentions use token-boundary name matching, so a short name is not credited inside a longer rival name.
- Mention rate (weight 0.55). Cells that name the business, divided by cells that were actually checked. n/a cells are out of the denominator.
- Position quality (weight 0.20). When the business is named, earlier in the list scores higher than later.
- Share of voice (weight 0.25). Mentions of the business divided by mentions of the business plus mentions of other named businesses in those answers.
The number is then rounded to a whole score and labelled:
- 0 to 20: Invisible
- 21 to 45: Barely visible
- 46 to 70: Emerging
- 71 to 100: Strong
If an engine cannot be checked, that engine is dropped from the run rather than scored as six misses. An incomplete check is labelled Check incomplete, not Invisible.
The Absent-Overview Exclusion
Google often shows the local map pack for a trade query and no AI Overview at all. When that happens, the AI Overview cell is n/a. It does not lower the score. Treating a missing Overview as "AI did not name you" would punish suburbs and trades where Google simply does not draw the box.
What the arithmetic still cannot see
A listing that might be the same business under a slightly different name can still count as a rival in share of voice. Where we flag that ambiguity we say so next to the score, because excluding those rows would inflate the number. The score can understate a business that is competing with its own second listing.
When two scores can be compared
A later pack only reports movement when all three identities match the earlier run:
- Score epoch (currently wombid-score-v3)
- Coverage contract (the Google AU-mobile search shape used for Overviews and Maps)
- Question mask (the exact six questions asked)
If any of those differ, the new score is a new baseline. We do not describe a method change as the business rising or falling.
Version archive
| Epoch | What changed | Comparable with |
|---|---|---|
| wombid-score-v1 | Original mention matching and engine set. | Other v1 runs with the same questions and coverage |
| wombid-score-v2 | Coverage identity moved with the Google AU-mobile pull. Unstamped records are treated as v1. | Other v2 runs with the same questions and coverage |
| wombid-score-v3 (current) | 4 August 2026. Name matching changed, so which answer cells count as a mention changed. Trends across this line are not comparable. | Other v3 runs with the same questions and coverage |
Engine-set identity is separately labelled wombid-engine-set-v1. Adding or removing an engine would rebase the score and need a new engine-set label. That has not happened in v3.
Data dictionary
- Run
- One audit of one business, one trade, one area, one timestamp.
- Cell
- One engine answering one of the six questions. Values: mention, miss, or n/a.
- Mention
- The engine named the audited business in that answer, under token-boundary matching.
- n/a
- The cell was not checked. Typical case: Google showed no AI Overview. Excluded from mention rate.
- Score epoch
- The named version of how a mention is decided. Current value: wombid-score-v3.
- Coverage contract
- The named version of the Google AU-mobile search pull used for Overviews and Maps.
- Question mask
- The exact question texts used on that run. Embedded in the comparability identity.
- Share of voice
- The business's mentions divided by all named-business mentions in checked cells.
Honest limits
- We do not guarantee a recommendation, a rank, or a job.
- We do not set up, submit or verify directory listings.
- We do not invent opening hours, phone numbers or service areas the customer did not give us.
- Repeat runs ask live engines. The same questions can return different names the next day because the engines changed, not because the method did.
See it on your own suburb
The free audit runs this method on your trade and area. No signup, no card.
Run the free AI Visibility Audit