All posts
AI visibilityMarketing analyticsAI search

AI visibility tracking: brand mentions and competitor share of voice

Measure AI brand mentions, citations and competitor share of voice with comparable question sets, clear definitions and evidence you can act on.

Magister Marketing5 min read

AI visibility tracking means checking how AI answers describe your brand, which competitors they mention, and what sources they cite. A useful report separates those signals, repeats the same questions under comparable conditions, and shows where coverage is missing.

The goal is not a screenshot of your brand in an answer. It is evidence you can use to decide what to explain, correct or publish next.

Magister’s November 12, 2026 changelog announced competitor share of voice over time, branded and unbranded reporting, and multiple prompt sets per brand. Those features raise a practical question: what should a founder or marketer measure, and what do the numbers actually mean?

Separate mentions, citations and recommendations

Start with three distinct questions.

Was your brand mentioned? Record whether the answer names your company or product. For a simple manual report, count a brand once per answer, even when its name appears repeatedly.

Was a source cited? Record the linked page and domain. Distinguish your own website from a third-party page discussing you. A link to your article does not, by itself, mean the answer recommends your product.

Was your brand recommended? Read the surrounding text. For this guide, treat a recommendation as an answer presenting your brand as an option to consider, rather than merely naming it. This is a suggested review rule, not a claim about a built-in Magister recommendation-rate metric.

OpenAI’s ChatGPT search guide says search responses may include citations. It also warns that sources can be incomplete, outdated or incorrect. Open the linked page before treating it as evidence.

Choose a question set you can keep comparing

Build your baseline around buying decisions your audience needs to make. Include the problems they are trying to solve, the capabilities they need and the alternatives they might evaluate.

Keep branded and unbranded questions separate. In this guide, a branded question names your company; an unbranded question does not. They test different things: how an assistant describes a named product versus whether that product appears during category discovery.

Record the question set, supported AI source, test date, language, location settings where relevant, and competitor list. Avoid blending a personal conversation full of company context with fresh-session tracking results.

Magister’s AI-visibility documentation describes rounds that ask supported AI sources questions in fresh sessions, score the answers and preserve their history. It advises comparing full rounds with full rounds, not treating incomplete coverage as a finished comparison.

Define the denominator before reporting a rate

For a straightforward manual mention rate, use:

Completed answers naming your brand ÷ all completed answers in the same question set, AI source and reporting period × 100.

Report the completed-answer count alongside the percentage. Failed or unavailable answers are missing coverage, not proof that an assistant omitted you.

Share of voice needs a different definition. One manual convention is to divide your brand’s appearances by the total appearances of your brand and a fixed competitor list. Count each included brand once per answer; an answer naming several brands contributes to several brands’ totals. If no included brand appears, report the result as unavailable.

That is an optional reporting convention, not Magister’s scoring formula. Its public docs do not specify the exact share-of-voice denominator or weighting.

Methodology varies. Ahrefs’ AI Visibility Metrics documentation defines its AI share of voice using estimated impressions relative to tracked brands. Do not assume similarly named metrics from different tools are interchangeable—or that either measures market share.

Compare like with like, source by source

Before calling a movement an improvement, check what changed in the measurement itself.

Did you change the questions? Add an AI source? Remove a competitor? Compare a complete round against one with missing answers? Each change alters the comparison you are trying to make.

Keep the baseline stable. If the scope needs to change, label the break and create a separate comparison rather than presenting a continuous trend without explanation.

Look at each AI source before reading an overall score. Google’s guidance on AI features says AI Mode and AI Overviews may use different models and techniques, so their answers and supporting links vary. Google also says AI Overviews do not appear for every query. No Overview is a different observation from an Overview that appears but leaves your brand out.

Read the answers behind the competitor gap

A competitor mention tells you where to investigate, not what to copy.

For each relevant gap, review the answer and its linked sources. Check what problem the answer addresses, why the competitor is included, and whether the description of your category is accurate.

Then inspect your own pages. Is the relevant capability explained clearly? Are important limits visible? Is the answer available publicly, or only in a sales conversation?

Magister’s competitor reports show brand and tracked-competitor mentions model by model, with low-sample warnings. Its cited-source reporting includes domains and sample URLs. Use those details to choose a page or explanation worth investigating—not to assume every competitor citation is an endorsement.

Turn a finding into a specific content change

Write down the observed gap before proposing a fix. Keep the observation separate from your explanation of it.

An omitted capability is something you can see in an answer. The idea that unclear documentation caused that omission is a hypothesis until you have more evidence.

Choose a change that addresses the reader’s unanswered question. Explain the capability plainly, state relevant limits, and cite supporting material. Record what changed and when, then review subsequent comparable rounds.

Treat before-and-after movement as evidence to investigate, not proof that your edit caused it. Keep a record of other changes that could affect the comparison.

Keep visibility separate from business results

A mention count is not a visit count. A citation is not a signup. A recommendation in a sampled answer is not a sale.

Report visibility alongside—but separately from—website visits and verified conversion events. State each metric’s source and reporting period. Avoid attributing revenue to an answer just because visibility and sales moved in the same direction.

Google’s guidance says traffic from its AI features is included in Search Console’s overall Web performance reporting. That does not give you an isolated AI-Overview conversion report. Keep the limits of your acquisition data visible rather than filling the gap with a confident attribution claim.

Start with evidence, then choose the next task

A useful review ends with a specific next step: correct an inaccurate description, strengthen an unclear page, or collect more evidence before acting.

Keep readiness separate from visibility, too. Magister describes its AEO audit as a checklist of whether AI crawlers can discover, parse and understand a site. That is not the same as measuring whether an assistant mentions or recommends the brand.

If you need a starting point for the broader work, get your starter audit and first-week plan. They are included in Magister’s free trial; no credit card is required.

Use tracking to identify the question. Use the plan to decide what to do about it.

Drafted by Magister, reviewed by our team.