
AI visibility: share of voice, sentiment and position explained
Know how often you appear, where you appear and what AI says about you.
An AI visibility report should separate being named, linked and recommended. It should also show your share of the conversation, your position in an answer and how you are described. Start with those questions instead of relying on a single score.
- Mention: does the answer name you?
- Citation: does it link to a page as evidence?
- Recommendation: does it suggest choosing you?
- Share of voice: how much of the tracked conversation is yours?
- Position: where do you appear in the answer?
- Sentiment: what does the answer say about you?
Mentions, citations and recommendations: did you appear?
A mention means the answer names your company. Ahrefs counts it once per answer, even if your name appears again.
A citation is a link used as a source. It may lead to your page or to another site talking about you. Being referenced is different from being recommended.
A recommendation suggests you as an option for the buyer. Read the sentence, rather than counting every appearance of your name as an endorsement.
For our suggested report, calculate mention rate by dividing answers that name you by all valid answers in the test. Calculate recommendation rate the same way, using answers that recommend you. Keep failed requests out and report their count separately.
Share of voice: how much of the conversation is yours?
Share of voice compares your visibility with a defined group of competitors. Ahrefs uses your share of estimated impressions across the brands you track. Those impressions are based on Google search volumes, not a count of actual AI readers.
For a simple internal report, we suggest a mention-based share of voice: your mentions divided by all mentions across the companies you track. Count each company once per answer. Label it “share of tracked mentions” so the formula is clear.
Keep the competitors, questions, engines and period fixed when comparing results. Ahrefs notes that changing the tracked brands can change share of voice.
A score can improve because a rival appears less often. Read it alongside your own mention rate. Neither measure tells you your share of sales.
Swipe across to read all columns.
| Basis | Company | Mentions | Share |
|---|---|---|---|
| Illustrative example | Your company | 8 | 40% |
| Illustrative example | Company B | 7 | 35% |
| Illustrative example | Company C | 5 | 25% |
In this made-up example, there are 20 mentions across the tracked companies. Your share is 8 ÷ 20 × 100 = 40%. The total here is mentions, not answers: one answer can name several companies.
Position: where does your company appear?
For your reporting sheet, define position as the order in which companies first appear in the answer. If yours is the third distinct company named, record third. This is our suggested counting rule, not a universal AI ranking standard.
Keep that separate from recommendation order. A company might be mentioned early as background but recommended later, or not recommended at all. Save the sentence that explains its role.
For an ordered shortlist, record the place on that list. For a comparison table or prose answer, record the display order without assuming it means best to worst. If the layout has no meaningful order, use “not applicable”.
Report position only where a company appears. Keep “not mentioned” as a separate outcome, rather than turning it into a zero. Otherwise an average can hide how often the company is absent.
If a tool gives earlier positions more weight in its share-of-voice score, ask for the weighting rule. Do not compare that score directly with an unweighted count of mentions.
Swipe across to read all columns.
| Basis | What the answer shows | What to record |
|---|---|---|
| Suggested method | A numbered shortlist | The company’s place on the list |
| Suggested method | Companies named in prose | Order of first appearance; do not call it a best-to-worst ranking |
| Suggested method | An unranked comparison table | Display order only, or not applicable if order is not meaningful |
| Suggested method | Your company is absent | Not mentioned; no position |
Sentiment: is the description helping or hurting you?
Sentiment describes the tone of what AI says about your company. Ahrefs recommends checking whether the description is positive, neutral or negative.
We suggest adding “mixed” when the same answer gives meaningful praise and criticism. Judge the words about your company, not the mood of the whole answer.
Keep the exact wording beside the label. “Good for small teams, but limited reporting” needs a different response from an entirely negative description. This sentence is an invented example.
Record the reason too: price, support, ease of use or a missing feature. Review unclear cases by reading the answer. A sentiment percentage without the reasons gives you little to act on.
Swipe across to read all columns.
| Basis | Example wording | Label | What to investigate |
|---|---|---|---|
| Illustrative example | A strong choice for small teams | Positive | What strengths are being recognised? |
| Illustrative example | Offers project management software | Neutral | Is there enough detail to help a buyer choose? |
| Illustrative example | Support is slow | Negative | Which source makes that claim, and is it current? |
| Illustrative example | Easy to use, but reporting is limited | Mixed | Check the praise and criticism separately |
For example, if 10 answers mention you and 6 are positive, report “6 of 10 mentioning answers were positive”. Keep negative, neutral and mixed counts beside it. This is an illustrative calculation, using only answers that mention you.
Accuracy and citation share: what else needs checking?
A positive description can still be wrong. Ahrefs recommends saving the question, response and cited sources when a company is described inaccurately or with outdated information.
Bing’s Citation Share measures your site’s citations against all citations for the same search query used to support an answer. It is not traffic share or a content-quality score.
Visible citations are not a complete reading history. OpenAI’s sources list can include consulted pages that were not cited in the answer.
Track visits and business enquiries separately. Traffic alone does not show whether people saw your company mentioned, cited or recommended without clicking.
Build a report you can act on
Keep a row for each answer. Record the engine, date, question, memory and conversation conditions. Save the answer itself so someone else can check the labels.
Add columns for Mention, Recommendation, Citation, Position, Sentiment and Accuracy. Summarise share of voice across a fixed group of competitors, using the formula written beside your report.
Keep answers with web search separate from answers without it where you can identify the difference. Report each engine separately before combining results.
Repeat the same questions across sessions and periods before calling a change a trend. Aleyda Solís recommends repeated testing for reliable presence measures.
Use the result to choose work. Low mention rate calls for a visibility investigation. Negative descriptions call for reading the criticism and its sources. Wrong facts call for correcting the information behind them.