The short answer
AI share of voice can be a useful visibility indicator for a defined question set. It is not revenue attribution or market share. Report it alongside the underlying counts, relevant referral traffic and separately sourced pipeline evidence. Use it to explain a decision, and clearly label what the data cannot establish.
Sooner or later, someone will point at the visibility chart and ask: “How much revenue did that generate?” It is a fair question. It also asks something the chart alone cannot answer.
The way through is to give each measure a clear job. A sampled AI answer can tell you who was recommended in that test. Your analytics can record some visits. Your sales process can collect evidence about how a deal developed. Connecting those records takes more than putting them next to each other.
Keep three layers of evidence separate
| Layer | What to report | What it does not prove |
|---|---|---|
| Visibility | Brand appearances and recommendations in a defined sample. | How many real people saw those answers. |
| Visits | Recognizable AI referral sessions and what those visitors did. | Every visit influenced by an assistant. |
| Commercial outcomes | Qualified leads, opportunities and customer-reported discovery. | That a change in sampled visibility caused the outcome. |
An assistant may name you without producing a click. A buyer may later search for your brand or arrive through a colleague. Conversely, a referral session does not prove the user received a favorable recommendation. Keep referrer information, customer-reported discovery and your attribution model as distinct fields where possible.
If you ask buyers how they found you, allow a free-text answer and keep the original response. “ChatGPT helped us make a shortlist” contains more useful context than forcing the deal into one perfectly tidy acquisition bucket.
Write the one-page brief before making the chart
Under that paragraph, include the provider breakdown, collection dates, repeat spread, missing answers and a link to the raw evidence. Put the proposed action, owner and review date on the same page. The reader should not have to decode five charts to find out what you want to do.
Then show commercial observations separately. For example: recognizable AI referral sessions, qualified inquiries from those sessions and the count of opportunities where a buyer explicitly mentioned an assistant. Explain the attribution window and avoid double-counting an opportunity that appears in more than one view.
Use first-party evidence where it is available
Bing’s AI Performance documentation describes citation reporting across supported Microsoft AI experiences and selected partners. Its scope is useful to know: it is not an all-assistant audience measurement, and citation activity is not a traffic or revenue figure.
Google also documents measurement for its generative AI search features. Keep platform reports labelled with their own coverage and definitions. Do not add a citation count from one platform to a sampled brand-mention count from another and call the result reach.
A third-party panel can help answer competitive questions that your own-site analytics cannot. First-party reports can add evidence about your site’s actual activity within their scope. They are complementary views, with different denominators.
Give the metric a decision to earn
Before adding share of voice to a recurring dashboard, name the decision it will inform. Correcting a recurring product misconception is one. Prioritizing a comparison page for a valuable use case is another. Watching an overall percentage because competitors have a dashboard is a weaker reason.
- Define the buying questions and market in scope.
- Choose a review cadence that matches how often you can act.
- Keep the baseline and measurement changes visible.
- Review evidence behind consequential movements.
- Report the action taken and what happened next, including no clear change.
Visibility can be a useful leading indicator without pretending to be booked revenue. A candid explanation of its limits makes it easier for a skeptical colleague to use the information. That is a better outcome than a metric nobody dares question.
Sources & context
This guide responds to a question raised in Is “share of voice in LLMs” a real KPI or a vanity metric? on Reddit. The discussion informed the topic; it is not evidence of the effectiveness of any product mentioned there. Platform documentation is linked alongside the claims it supports.
Examples and figures are illustrative unless explicitly stated otherwise. Written and published by Share of Voice, a Superstellar LLC product. We sell AI visibility reports; our commercial perspective is worth keeping in mind.
Spotted something we should correct? Let us know.