How to track AI search share of voice
AI search share of voice describes how observed brand appearances are divided among the tracked brands in a fixed prompt and engine panel. Record mentions, source links, recommendation classifications, competitor appearances, unavailable runs, and the exact denominator, then compare like-for-like panels over time. The metric describes those sampled answers; it does not measure authority, audience awareness, market demand, or why an engine produced them.
What is AI search share of voice?
AI search share of voice is the proportion of tracked brand appearances assigned to your brand across eligible answers in a defined panel. It measures observed presence under those conditions, not classic search rankings or the wider market.
A strong share-of-voice report shows more than whether your brand appeared. It shows how the brand appeared: cited as a source, mentioned in text, included in a shortlist, recommended as an option, or replaced by a competitor.
- Mention share: how often your brand is named.
- Citation share: how often your brand or domain is cited.
- Recommendation share: how often your brand is included as a suggested option.
- Competitor replacement: where competitors appear and you do not.
- Source share: whether cited evidence comes from owned or third-party sources.
Why AI share of voice matters
Generated answers can contain shortlists, comparisons, recommendations, mentions, and source links. A panel can record whether a tracked brand appeared in those outputs and how that result varied by prompt and engine.
Share of voice turns those observations into a trend for the registered panel. It does not show whether a person noticed the brand, considered it, clicked, bought, or whether the prompts represent total demand.
- It records prompts where sampled answers list competitors but omit your brand.
- It lets the team compare those observations with its own stated priorities.
- It separates citation presence from mere brand mentions.
- It shows differences among the named services in the registered panel.
- It creates a baseline for later like-for-like observations.
Step 1 — Define the competitor set
A share-of-voice metric is only meaningful if the competitor set is clear. Include direct competitors, substitutes, and other brands that recur in the registered answers or are documented in the team's research scope.
The list should be stable enough for tracking but flexible enough to add competitors that repeatedly appear in AI answers.
- Direct competitors: products solving the same problem.
- Frequently observed brands: names that recur in broad prompts in the registered panel.
- Substitutes: tools buyers may use instead of your category.
- Emerging competitors: brands appearing in new AI answers.
- Your own brand aliases: spelling, product name, company name, and domain.
Step 2 — Build the prompt set
Prompt selection determines what the share-of-voice number describes. Choose multiple documented decision contexts instead of treating one broad query as the market.
Each prompt class creates a different observation scope: category answers, alternatives, comparisons, use cases, problems, or evidence questions. The panel does not establish demand or source trust.
- Category prompts: best AI visibility tools.
- Alternative prompts: Profound alternative, Semrush alternative for AI visibility.
- Comparison prompts: RankEcho vs Profound, AI citation tracker vs rank tracker.
- Use-case prompts: AI visibility software for agencies, SaaS, ecommerce, or enterprise.
- Problem prompts: why ChatGPT does not mention my brand.
- Measurement prompts: how to document observed AI visibility change.
Step 3 — Record mentions, citations, and recommendations separately
Do not collapse every appearance into one score. A brand mention, a returned source link, and a recommendation classification are different observed events. None by itself establishes source authority, audience consideration, brand awareness, or the engine's reason for including the item.
Track the events separately first, then combine them into a weighted summary only if the weighting is clear.
- Mention: the answer names the brand.
- Citation: the answer links to or cites a source connected to the brand.
- Recommendation: the answer includes the brand as an option or shortlist candidate.
- Competitor-only answer: competitors appear but your brand does not.
- Misdescription: the brand appears but is described incorrectly.
Step 4 — Calculate prompt-level share of voice
At the prompt level, share of voice compares your brand's counted appearances with competitor appearances for the same prompt under the registered outcome rule.
An answer that omits your brand while listing three competitors is an observed gap. An unlinked mention is a different state from an owned source link. Use the raw answer and source list to form hypotheses; neither state identifies the fix or cause.
- Prompt SOV = your brand appearances divided by all tracked brand appearances for that prompt.
- Citation SOV = your citations divided by all tracked brand citations for that prompt.
- Recommendation SOV = your recommendations divided by all tracked brand recommendations.
- Competitor replacement = prompt where competitor appears and your brand is absent.
- Zero-result prompt = no tracked brands appear; review the prompt scope before interpreting it.
Step 5 — Roll up by category, engine, and prompt class
A single overall score can hide the real problem. Rollups should show where visibility is strong and where it breaks: by engine, prompt class, competitor, source type, and time period.
For example, your brand may have strong mention share in ChatGPT but weak citation share in Perplexity, or strong category visibility but weak alternative-prompt visibility.
- Engine rollup: ChatGPT, Perplexity, Claude, Gemini, Copilot, AI Overviews.
- Prompt-class rollup: category, comparison, alternative, use case, problem, proof.
- Competitor rollup: who replaces you most often.
- Source rollup: owned, review, community, editorial, directory, documentation.
- Trend rollup: observed results across comparable windows, with shipment dates noted.
Step 6 — Classify the visible source mix
Source mix describes the links returned with the sampled answers. An owned link, third-party link, or competitor-linked page is an observable category; it does not explain why share of voice changed or prove that the source caused the answer.
Use the categories to choose follow-up research. Inspect the intended page when an owned link is absent, and review relevant third-party coverage when those pages recur, but keep access, content, source selection, and other explanations open until direct evidence distinguishes them.
- Owned citation share: citations to your own domain.
- Third-party citation share: citations from independent sources that mention you.
- Competitor-linked sources: returned pages that mention or link to competitors.
- Community source presence: Reddit, forums, and niche communities.
- Review/source presence: G2, Capterra, directories, marketplaces, and roundups.
Step 7 — Turn losses into testable hypotheses
A low share-of-voice number is not a diagnosis. Review access, page clarity, public identity facts, visible sources, prompt fit, competitor appearances, and answer accuracy as separate hypotheses.
A useful report links each hypothesis to the evidence that supports or weakens it and keeps an unresolved state when the observations cannot identify a cause.
- Access hypothesis: a verified intended crawler or delivery path fails for the page.
- Page-clarity hypothesis: the requested fact is missing, unclear, or unsupported in public content.
- Identity-consistency hypothesis: public brand or category facts conflict across sources.
- Source-coverage hypothesis: returned third-party pages omit material, supportable facts about the brand.
- Prompt-fit hypothesis: the tracked question does not match the decision scope the team intended to measure.
- Accuracy finding: the observed answer describes the brand incorrectly.
Step 8 — Use share of voice to prioritize fixes
Prioritization depends on the team's stated business goal, the prompt's role in its research plan, the size and completeness of the observed gap, and whether a bounded issue can be verified and changed.
Do not treat a broad score, a returned source, or a recommendation label as proof of buyer intent, source trust, awareness, or likely commercial impact.
- Review competitor-only results for prompts the team has already marked important.
- Keep an unlinked mention separate from an owned source-link result.
- Correct inaccurate public brand facts when the evidence verifies them.
- Inspect the intended page before proposing a content or delivery test.
- Research recurring third-party pages without assuming they caused the answer.
What a useful AI share-of-voice report includes
A good report should show the exact scope behind every number: prompts, engines, eligible answers, unavailable runs, outcome definitions, source URLs, competitor names, and reviewable next tests.
It should show what changed since the last comparable panel, what was shipped between windows, and the observed re-test movement without saying the shipment caused it.
- Overall AI share of voice and trend.
- Citation share, mention share, and recommendation share.
- Prompt-level table with engines and cited URLs.
- Competitor replacement table.
- Source mix and off-site gaps.
- Follow-up list ranked by stated business priority and evidence completeness.
- Before-and-after observations following shipped changes.
How RankEcho helps
RankEcho tracks share of voice at the prompt level. It records brand mentions, source links, competitor appearances, source types, repeat counts, and unavailable results, then connects each gap to a reviewable hypothesis.
The workflow is: measure the prompt, inspect the evidence, choose and ship one bounded change, and record later answers. It does not prove why an engine changed or that the shipment worked.
Frequently asked questions
It is your brand's share of counted appearances among tracked brands in a defined prompt-and-engine panel. State the panel, eligible answers, outcome rules, and denominator with the result.
No. Citation rate measures how often your brand or domain is cited. Share of voice compares your brand presence against competitor presence across the same prompt set.
There is no universal weighting. They are different observed events, so report them separately first. If you publish a combined score, disclose the weights, denominator, and decision purpose without treating the score as authority or awareness.
There is no universal prompt count. Define the decisions and prompt classes the panel is meant to cover, document how prompts were selected, and keep the set stable enough for like-for-like comparison. State what the panel does not represent.
There is no universal cadence. Choose it from the decision deadline, expected variability, provider limits, and collection budget, then keep comparable windows and disclose any schedule change.
That depends on the stated business goal and evidence. A competitor-only result on an important registered prompt can be a review priority, but it still requires raw-answer inspection and a testable hypothesis rather than a presumed diagnosis.