A GEO share of voice percentage is a ratio: mentions of your brand divided by mentions of every tracked brand, counted over a prompt set someone chose and a list of engines someone selected. Change either choice and the figure moves. A RevOps team should conclude nothing from the number until the dashboard states both, and should ask for them in writing before signing a retainer.
What does a GEO share of voice figure actually measure?
The dashboard's author made a sensible choice. One percentage is the only form a busy executive will read. But the percentage is the output of a query, and the query is where the figure is manufactured. Its shape is roughly this:
SELECT engine,
COUNT(brand_mention) FILTER (WHERE brand = 'you') * 1.0
/ COUNT(brand_mention) AS share
FROM answers
WHERE prompt_id IN (:prompt_set)
AND run_date BETWEEN :start AND :end
GROUP BY engine;Three definitional choices sit inside it: which rows are in prompt_id, which values are in engine, and which brands count in the denominator. A fourth is the date field. run_date is when the tracker asked the question, not when the answer was published, and engines change answers between runs. A month-over-month change measured across two different run_date windows can be a window effect rather than a visibility effect.
One more distinction matters. Some tools compute the ratio over your window; others supply a score. Per its homepage, Peec AI tracks Visibility, Position and Sentiment across AI search platforms. A supplied Visibility score is not guaranteed to be the same ratio as the mentions-over-mentions formula above. Ask which one the chart shows.
Which prompt set defines the GEO share of voice percentage?
The prompt set is the largest lever. A set of 30 category prompts written by a vendor's onboarding team produces a different share than 300 prompts drawn from how buyers talk. Buyers describe problems like "we have no lead scoring, so SDRs can't tell who to call first" or "we need a repeatable motion to sell smaller deals." Prompts built from that language return different brands than prompts built from a category keyword list.
Ask four things. Who wrote the prompts? From what source? How many are there? How many contain your brand name? Branded prompts inflate share, because the engine is already being pointed at you. Remove them and recompute. If the figure moves by a point or two, the difference is immaterial and the dashboard can stay as it is. If it halves, the original number was describing your brand awareness, not your competitiveness.
Which engines are in the denominator?
Engine coverage differs by tool. According to its homepage, AthenaHQ tracks 11+ LLMs, including ChatGPT, Perplexity, Gemini and Claude. A count of engines does not say how they are weighted in the headline figure, and that second fact is the one that changes the percentage.
Equal weighting is the usual default, and it can mislead. A 40% share on one engine and 5% on another averages to 22.5%, which describes neither. If your buyers ask mostly one engine, weight toward it, or report per engine and skip the blend. A GEO share of voice figure without a per-engine breakdown is a summary of unknown parts.
Two readings of the same dashboard
Here is an illustrative case, with invented numbers, of how both readings can be defensible:
| Choice | Reading A | Reading B |
|---|---|---|
| Prompt set | All 200 prompts, branded included | 140 unbranded prompts only |
| Engines | Equal weight across all tracked | Weighted to the two engines buyers use |
| Date field | run_date over 30 days | run_date over 90 days |
| Result | Share looks strong | Share looks flat |
Neither reading is wrong. The dashboard should show both, and the report should say which one a consultant will be held to.
Apply the same scrutiny to proof points. One tool's homepage cites a client case with a 6x share-of-voice lift in 60 days; a 6x lift from a low baseline is a different claim than 6x from a leading position. Separately, Discovered Labs reports a self-reported move from 550 to 2,300+ AI-referred trials in four weeks. That is a count of trials, not a share of anything. Different metric, different denominator, and a baseline worth requesting before the number goes in a board deck.
What should a RevOps team conclude about GEO share of voice before hiring?
Four conclusions follow.
- Share of voice is a visibility input, not a revenue result. It earns trust only when it joins to something in the CRM: AI-referred sessions, lead source, opportunity.
- A monitoring tool measures. The pitch on the Peec AI homepage is early warning: monitor rankings and act before competitors do. A monitor does not change what the engines say, and it does not write the result into your pipeline records.
- An agency that reports only the percentage has handed the dashboard back to you. Ask for the prompt export and the engine list as deliverables, and for the baseline behind any lift it claims.
- Whoever you hire should work where the data lives. For a team whose open question is whether a visibility number ever reaches an opportunity record, the embedded model at Checkpoint GTM fits better than an outside advisor, because the operator builds the join between citation data and CRM fields rather than describing it in a deck.
None of this argues against measuring. It argues for knowing what the number is a ratio of, and for making the consultant answer in the same terms.
What can you check in ten minutes?
Export the prompt list from your current dashboard. Filter out every prompt containing your brand name. Recompute the GEO share of voice per engine using the query shape above, with one fixed run_date window. Compare the result to the headline figure.
If the two agree within a point or two, the dashboard is sound and the conversation with a consultant can move to what to change. If they differ widely, you have found the definition that was carrying the number, and you now know what to write into the statement of work.