AI share of voice is your brand’s share of a clearly defined set of brand mentions in AI answers. To measure it, choose buyer questions and competitors, run the same questions across selected engines, and divide your brand’s counted mentions by the total counted mentions in that competitive set. Report answer presence and source citations separately: they answer different questions.
What does AI share of voice actually measure?
A share-of-voice figure is only useful when its denominator is clear. One tool may count every brand occurrence; another may count answers containing each brand. Those methods can produce different percentages from the same responses. Write down the method before comparing reports.
In WildSEO’s competitor reporting, share of voice uses counted brand citation spans against the configured competitive set. An answer repeating a brand several times can contribute several spans. The number of answers containing the brand is a separate measure. A linked source URL is also different from a brand occurrence: an answer can recommend your product while citing someone else’s page.
This is a measurement of the prompts, engines and dates you sampled. It is not your share of all AI conversations, market demand or revenue. WildSEO’s checks do not reveal private user chats. See the WildSEO MCP overview for how scheduled checks and workspace data are exposed.
How should you calculate and report it?
Use three measures together. Share of voice describes the competitive mix; mention rate describes how often you appear; linked citations describe which pages support the answer.
- Competitive share of voice: your counted brand spans divided by all counted brand spans for you and the selected competitors, multiplied by 100.
- Answer presence: completed eligible answers naming your brand divided by all completed eligible answers, multiplied by 100.
- Own-site citation rate: completed eligible answers linking to your domain divided by all completed eligible answers, multiplied by 100.
Define exclusions consistently. Record failed runs and missing providers rather than treating them as answers with no mention. Do not silently filter negative mentions out of an overall visibility metric. If you report positive mentions separately, label that filter.
A fictional worked example
The following numbers are invented solely to demonstrate the arithmetic, not measured results. Suppose 100 completed answers contain 30 counted mentions of your brand, 50 of competitor A and 20 of competitor B. Your competitive share is 30 / 100 = 30%. If your brand appears in only 20 of those answers, its answer-presence rate is 20%. If eight answers link to your website, the own-site citation rate is 8%. These numbers are illustrative, not WildSEO customer results.
Now suppose one long response repeats your brand ten times. A span-based share can rise even if you appear in no additional answers. Inspect the underlying responses before calling that change a discovery win. If the denominator is zero, report the share as unavailable rather than a meaningful zero.
How do you build a useful tracking baseline?
- Choose actual buyer problems. Start with sales questions, support requests and relevant search queries. For a marketing team, an example might be “How can a small team compare brand recommendations across AI search engines?” Avoid questions that instruct the model to praise or remember your brand.
- Separate intent groups. Keep category discovery, comparisons, implementation questions and branded support questions in separate groups. Branded questions often produce high mention rates by construction.
- Fix the competitor set. Select relevant alternatives and document the selection. Adding a competitor changes the denominator, so start a new comparable baseline or clearly annotate the change.
- Record engine and context. Keep model/provider, date, language, location where supported, and retrieval settings with each result. Use fresh conversations for manual checks. A fresh conversation avoids including the previous thread in the prompt; variation can still remain.
- Repeat on a cadence. Compare equal windows with comparable prompt and provider coverage. Show sample sizes and run failures beside the percentage. A single answer is an example, not a durable trend.
The WildSEO platform combines tracked questions, responses, competitors and cited sources. Use it to inspect the evidence behind a change rather than relying on a headline score.
What should you investigate when competitors appear instead?
Read the answer first. Is the competitor recommended for a use case you serve? Is its description accurate? Which source pages support the answer? A competitor mention is a starting point for investigation, not proof that copying its article will work.
Build a small action log with the buyer question, observed gap, cited URL, proposed page change, owner and date shipped. For a fictional software vendor, the action could be to add a verified integration walkthrough and link it from the related feature page. That is a testable improvement with value for readers.
If the model repeats outdated pricing, compare the claim against current approved facts, then inspect your pricing page and relevant third-party listings. Correct what you control and document what remains inaccurate. Avoid claiming that a page edit will immediately change a model’s answer.
Which improvements are worth testing?
Prioritize useful, accurate pages that people can access and understand. Add first-hand examples, clear product details, original analysis where you have permission to share it, and contextual links between related pages. Use descriptive headings and tables when they make the material easier to read.
For Google’s AI search features, Google’s official guidance says foundational SEO remains relevant and a supporting page must be indexed and eligible for a search snippet. Its generative AI optimization guide emphasizes unique value rather than special AEO hacks. Clear formatting helps readers; it does not establish why an engine selected a source. Structured data should match visible content.
Earn relevant third-party coverage through useful expertise and genuine product information. Do not manufacture reviews, seed deceptive recommendations or buy spam links. Source diversity can be worth studying, but an observed association is not proof of an engine’s ranking mechanism.
How do you connect visibility to business results?
Review qualified organic visits, AI referral visits, trials, demos and conversions alongside visibility. Keep the dates and attribution limits visible. An increase in monitored mentions does not establish that it caused additional sales, and referral reporting may not capture every visit originating from an AI experience.
Use a weekly summary with four lines: what changed, which responses demonstrate it, which useful page improvements shipped, and what happened to qualified visits and conversions. Keep unsuccessful experiments in the record too. For terminology, use the AEO glossary; for related practical guides, browse Trailhead.
Ready to establish your own baseline? Start a WildSEO trial, choose a focused set of buyer questions, and inspect the first responses before expanding your monitoring.
