Research note · Checked October 31, 2026
Ahrefs published its AI Share of Voice (SOV) method on October 22, 2026. The metric compares a brand’s appearances with tracked competitors; its weighted version uses Google keyword volume multiplied by an Ahrefs platform ratio to estimate exposure. Those “impressions” are modeled estimates, not counts of people who saw an AI answer.
The formula can support comparison when its prompt set, platforms and competitor set are clearly defined. It should not be reported as direct AI query volume or used as the only measure of visibility.
How Ahrefs describes AI Share of Voice
Ahrefs gives a basic response-based formula:
AI SOV = responses mentioning your brand ÷ responses mentioning any tracked brand × 100
That calculation counts tracked brand appearances in the selected answers. Ahrefs’ Brand Radar also weights prompts by “AI-adjusted volume”: the Google search volume of the prompt’s parent keyword multiplied by a platform ratio derived from Ahrefs’ aggregated referral traffic data. In its October 22 article and current help documentation, Ahrefs lists ratios of 0.115 for ChatGPT, 0.011 for Gemini, and 0.010 for Perplexity.
In the weighted implementation, estimated impressions are derived from those adjusted volumes for prompts where a brand appears or its site is cited. Brand SOV then compares those estimates across tracked brands. Because the denominator and prompt weights matter, the result can differ from a simple unweighted share of responses.
What the metric does—and does not—measure
| Metric | What it records | What it does not prove |
|---|---|---|
| Mention | A brand appears in a response. | That its website was cited or recommended. |
| Citation | A page or domain is linked as a source. | That the brand was endorsed, or that a reader clicked. |
| Estimated impressions | A modeled exposure estimate based on adjusted prompt volume. | An observed count of answer views or people reached. |
| AI SOV | A comparison with other tracked brands under the chosen setup. | Market share, customer preference, leads, or revenue. |
Ahrefs itself distinguishes mentions, citations, estimated impressions and SOV, and notes that a brand can be mentioned without a citation—or cited without being recommended. The distinction matters because teams often combine these outcomes under a single “AI visibility” heading.
For a measurement framework built around directly observed answers, see Ranknizer’s AI Search Visibility metrics and KPIs. That guide also recommends showing the number of responses behind rates and retaining platform-level results.
Where the weighting assumptions enter
The weighted score is an estimate built from several choices: the prompt or parent keyword, its Google search volume, the platform ratio, tracked competitors, and the platforms included. The source of the ratios is Ahrefs’ own aggregated web traffic data. Those ratios are a vendor methodology, not an independently observed count of how many prompts users submit to each AI product.
A change in a prompt set, competitor list, brand aliases, market, platform mix, search-volume source, or ratio version can move the score even if the brand’s actual answer presence has not changed. The reverse can also happen: a combined score may look stable while one platform improves and another weakens.
A more defensible reporting approach
Use two layers in client reporting and label them clearly.
1. Observed answer visibility
- Prompt set, tested date, market, language, platform and model version
- Valid responses and repeat runs
- Brand mention rate, recommendation rate and citation rate
- Brand-owned versus third-party citations
- Answer accuracy, prominence, and competitor appearances
- Referral sessions and conversions as separate website outcomes
2. Modeled exposure
- The vendor’s formula and definition of an impression
- Ratio source and version or review date
- Keyword volume source and parent-keyword mapping
- Platforms, tracked entities and competitor denominator
- Any change in prompts, location, aliases, weights or tool settings
Report each platform on its own before presenting a combined weighted number. Keep the numerator, denominator, prompt count, and methodology date beside the score. If a number comes from a vendor model, call it an estimate and avoid implying it is a directly observed audience count.
Common questions
Is AI Share of Voice a standardized industry metric?
No single formula applies across all tools. Ahrefs’ article explains its own definition and Brand Radar implementation. A comparison between vendors is meaningful only after checking their prompts, entity matching, platform coverage, weights and counting rules.
Does a higher weighted SOV mean more people saw my brand?
Not directly. The weighted exposure is modeled from Google search volumes and platform ratios. It is useful as a comparative estimate within a documented method, but it is not an observed audience measurement.
Can a combined AI SOV score hide a platform weakness?
Yes. Ahrefs notes that its combined score is impression-weighted, so a platform with larger modeled volume can dominate the result. Review each platform separately and retain the combined score only as a supplementary summary.
Prepared by Muhammad Sukhera, AI Search Visibility Consultant at Ranknizer. The ratios and implementation details can change; verify the current vendor documentation before comparing reports. Next review: November 30, 2026.