How to Track AI Visibility: Brand Mentions, Citations and Share of Voice

Tracking AI visibility means measuring how often your brand appears, how often your pages are cited, which competitors appear beside you and what types of prompts trigger those results. The useful unit is not one impressive screenshot. It is a repeatable prompt set measured over time.

This is the measurement problem I care about most when thinking about AI visibility in CrawlerQue: turning generated answers into comparable signals instead of treating every response as an anecdote.

Why AI visibility is harder to measure than rankings

Traditional rank tracking is relatively structured.

You choose a keyword, location and device, then record where a URL ranks. The search result can change, but the output format is still predictable.

Generative search is more variable:

• The same prompt can produce different wording on different runs.
• The model may mention a brand without linking to it.
• It may cite one page today and another tomorrow.
• Different users can receive different results.
• The answer can depend on live web retrieval, model version and geography.
• One prompt can contain several hidden subqueries.

Because of that, the right goal is not perfect deterministic tracking. It is consistent sampling.

The six metrics that matter most

1. Brand mention rate

Brand mention rate answers a basic question: in what percentage of tracked responses does your brand appear?

Brand mention rate = responses mentioning your brand ÷ total valid responses

If you test 100 prompts and your brand appears in 23 valid answers, your mention rate is 23 percent.

Segment it by intent. A 40 percent mention rate on educational questions may be less commercially useful than a 15 percent rate on high intent recommendation prompts.

2. Citation rate

A mention and a citation are not the same.

Citation rate measures how often the answer provides a visible link or source attribution to your domain.

Citation rate = responses citing your domain ÷ total valid responses

Also track citation without mention. Sometimes your page supports an answer even when your brand name is not stated in the prose.

3. Share of voice

AI share of voice compares your presence with a defined competitor set.

One simple version is:

Your mentions ÷ mentions of all tracked brands

A better version weights prompts by commercial value. For example, a recommendation prompt can carry more weight than a basic definition prompt.

4. Prompt coverage

Prompt coverage shows where visibility exists.

Tag each prompt into categories such as:

• Awareness.
• Problem research.
• Comparison.
• Recommendation.
• Purchase evaluation.
• Troubleshooting.

Then calculate mention and citation rates by category.

This tells you whether the brand is merely known or actually present during buyer decisions.

5. Source page distribution

Track which URLs from your domain are cited.

If one article earns 80 percent of all citations, you have both a strength and a risk. The page is valuable, but the rest of the site may not be contributing much to AI retrieval.

Useful dimensions include:

• URL.
• Page type.
• Topic cluster.
• Citation count.
• Prompt category.
• Assistant or search surface.

6. Mention context

A raw mention can be positive, neutral or irrelevant.

You may want to classify whether the answer:

• Recommends the brand.
• Lists it as one option.
• Uses it only as an example.
• Cites it as a source.
• Mentions a limitation.
• Confuses it with another entity.

This is especially important for brands with common or ambiguous names.

Build a prompt set that represents real demand

Your tracking quality depends on the prompt set.

Do not start with 500 random questions. Start with 30 to 100 prompts that map to actual customer journeys.

Use five prompt groups

1. Category prompts

“What are the best tools for technical SEO audits?”

2. Problem prompts

“How can I find pages that are indexed but getting no impressions?”

3. Comparison prompts

“What is the difference between an SEO audit platform and an AI visibility platform?”

4. Recommendation prompts

“Which tools combine SEO auditing with AI visibility tracking?”

5. Brand prompts

“What is CrawlerQue and what does it do?”

Brand prompts are useful for entity accuracy, but they should not dominate the score because they already contain the answer you want the system to recognize.

How to generate better prompts

A weak prompt list often sounds like a keyword export with question marks added.

Use real variables:

• Company size.
• Industry.
• Budget.
• Existing tools.
• Goal.
• Constraint.
• Geography if relevant.
• Stage of the buying process.

Instead of:

best CRM

Try:

I run a small digital agency with five sales and delivery users. We need one CRM for leads, follow ups and project handoff. What features should I prioritize?

Longer prompts expose whether your content covers the real decision context.

Normalize brand names before scoring

AI responses can write a brand in multiple ways.

For accurate mention tracking, normalize:

• Capitalization.
• Common spacing differences.
• Official abbreviations.
• Product name variants.
• Domain name mentions.

For example, CrawlerQue, crawlerque.com and a lowercase crawlerque may all represent the same entity.

Be careful with fuzzy matching. A short brand name can generate false positives if it also appears as an ordinary word.

Track citations separately from source domains

A response may cite:

• Your homepage.
• A blog post.
• A third party page that talks about you.
• A review platform.
• A social profile.

Do not collapse all of these into one “citation” metric.

I recommend at least three fields:

1. Owned citation: your domain is cited.
2. Earned citation: another site is cited while discussing your brand.
3. Profile citation: an official social or directory profile is cited.

This helps diagnose what kind of web footprint is supporting visibility.

A useful AI visibility score

A single score is attractive, but it can hide important detail. If you use one, make it explainable.

For example:

Component	Example weight
Brand mention rate	30%
Owned citation rate	25%
Recommendation prompt visibility	20%
Competitive share of voice	15%
Entity accuracy	10%

The exact weights should match the product or business objective.

A publisher may care more about citations. A SaaS company may care more about recommendation prompts. A local service business may care about geography and service intent.

The score should summarize the evidence, not replace it.

How often should you run the same prompts?

For most businesses, weekly or monthly tracking is more useful than constant checking.

The right cadence depends on:

• How often your content changes.
• How quickly the market changes.
• The number of prompts.
• API or tool cost.
• The variability of the assistants you monitor.

If you are testing a major content change, you can sample more frequently for a short period. For stable reporting, keep the cadence consistent so trend lines mean something.

Control the variables you can

Document the conditions of every run.

Useful fields include:

• Date and time.
• Platform or assistant.
• Model if exposed.
• Search or web retrieval enabled or disabled.
• Geography if configurable.
• Prompt text.
• Response text or structured extract.
• Mentioned brands.
• Citation URLs.

Without this metadata, comparisons become difficult later.

Connect AI visibility to search data

AI tracking is more useful when combined with normal search performance.

Google's Search Console provides reporting for generative AI features in Search, and Bing Webmaster Tools offers AI performance reporting for supported Microsoft AI experiences. Use those first party signals alongside your prompt testing.

The combination can reveal patterns such as:

• A page gaining search impressions and AI citations together.
• Strong AI mentions with weak click traffic.
• High rankings but no brand mentions in recommendation prompts.
• A third party page being cited more often than your own page.

That is much more actionable than one metric in isolation.

Diagnose why a competitor keeps appearing

When another brand is repeatedly surfaced, do not immediately copy its wording.

Investigate the evidence behind the pattern.

Ask:

1. Which pages are being cited?
2. What exact subtopic do those pages answer?
3. Are they supported by reviews or independent mentions?
4. Is their entity clearer across the web?
5. Do they have original data or product documentation?
6. Is their page more current?
7. Is their answer more specific to the prompt?

Then improve the weakest layer on your own site.

For strategy, pair this measurement framework with my GEO guide for 2026 and content strategy for AI search.

A simple reporting dashboard

A useful monthly dashboard can fit on one page.

Top row

• Total prompts tracked.
• Brand mention rate.
• Owned citation rate.
• Competitive share of voice.

Trend charts

• Mention rate over time.
• Citation rate over time.
• Recommendation prompt visibility over time.

Tables

• Top cited pages.
• Top prompts where you appear.
• High value prompts where competitors appear and you do not.
• New source domains mentioning your brand.

Action list

End with three to five content or authority actions for the next month. Measurement without a decision loop becomes reporting theatre.

Common measurement mistakes

Counting navigation prompts as wins

If the prompt is “What is Brand X?” and Brand X appears, that proves entity recognition, not category leadership.

Mixing prompt sets every month

You cannot compare trends if the questions keep changing. Maintain a stable core set and a smaller experimental set.

Treating generated answers as perfectly reproducible

They are not. Use repeated observations and trend direction.

Ignoring citations

A brand can be absent from the prose while its content still supports the answer. Measure both.

Ignoring business intent

A mention on a low value informational prompt should not automatically count the same as a recommendation during purchase evaluation.

Frequently asked questions

What is AI share of voice?

AI share of voice is the proportion of tracked brand visibility that belongs to your brand compared with a defined competitor group across a stable set of prompts.

Can Google Search Console track AI visibility?

Search Console can report visibility from Google's generative AI search features, but it does not replace cross platform prompt tracking for assistants outside Google.

How many prompts should a small business track?

A focused set of 30 to 100 good prompts is usually more useful than hundreds of weak prompts. Add more only when you can organize them by intent and review the results meaningfully.

Should mention sentiment be part of the score?

Only if it changes decisions. For many brands, recommendation context and citation quality are more actionable than generic positive or negative sentiment.

The takeaway

Treat AI visibility like a measurement system, not a screenshot collection.

Define a stable set of real customer prompts. Track mentions, citations, competitive share and context separately. Connect those results to Search Console and normal organic performance. Then use the gaps to decide what content, entity or authority work should happen next.

That turns AI visibility from a vague marketing idea into something you can actually improve.