
What Is AI Visibility Tracking?
AI visibility tracking is the practice of measuring how often AI engines such as ChatGPT, Perplexity, Copilot and Google’s AI features mention your brand and cite your pages. Because answers change from run to run, it works by asking a fixed set of prompts repeatedly and recording patterns. You can do it by hand or with a tracking tool.
| Metric | What it measures | How to get it |
|---|---|---|
| Mention rate | Share of prompt runs that name your brand | Prompt testing |
| Citation rate | Share of runs that link to your pages | Prompt testing |
| Share of voice | Your mentions compared with competitors’ | Prompt testing |
| Position in answer | Whether you are named first, or in passing | Prompt testing |
| Accuracy | Whether you are described correctly | Manual review |
| Sentiment | Whether the description is favourable | Manual review |
| Cited sources | Which sites the engine relies on | Prompt testing |
| AI referral visits | Sessions arriving from AI tools | Web analytics |
There is no official “AI Search Console”. Google reports its AI features inside ordinary Search Console totals, and other engines offer site owners no report at all.
Why Is AI Visibility Hard to Measure?
Four properties make it different from tracking rankings.
- Answers are not deterministic. The same prompt can produce different wording and sources each time.
- Results are personalised. Location, memory and conversation history influence answers. OpenAI notes that ChatGPT search may use general location and saved memories when rewriting queries.
- Prompts are unlimited. There is no finite keyword list with volumes.
- Much of the value has no click. A mention can influence a buyer who never visits your site.
So AI visibility tracking produces estimates and trends, not exact positions. Report it that way.
How Do You Build a Prompt Set?
A prompt set is your equivalent of a keyword list. Build it from real customer language.
- Category prompts: “What are the best [category] for [audience]?”
- Problem prompts: “How do I [solve the problem you address]?”
- Comparison prompts: “[You] vs [competitor]”, “alternatives to [competitor]”.
- Brand prompts: “What is [your brand]?”, “Is [your brand] reliable?”, “[your brand] pricing”.
- Buying prompts: “Which [product] should I choose if [specific situation]?”
- Local prompts, if relevant: “Best [service] in [city]”.
Guidelines:
- Start with 20 to 50 prompts.
- Write them as a customer would, in natural sentences.
- Include prompts where you do not expect to appear.
- Fix the list. Changing prompts every month destroys comparability.
- Review the list quarterly.
Sources: sales calls, support tickets, “People also ask” boxes and your existing keyword research.
How Do You Track Manually?
Manual tracking suits a small prompt set and costs only time.
- Open a spreadsheet with one row per prompt, per engine, per run.
- Use a fresh session for each run, logged out or in a private window, with memory off where possible.
- Run each prompt three to five times per engine.
- Record: brand mentioned (yes or no), cited (yes or no), order of mention, competitors named, sources cited, any errors.
- Calculate rates: mentions divided by runs.
- Repeat on the same day each month.
| Prompt | Engine | Runs | Mentioned | Cited | Top competitors |
|---|---|---|---|---|---|
| Best CRM for small agencies | ChatGPT | 5 | 2 | 1 | A, B, C |
| Best CRM for small agencies | Perplexity | 5 | 3 | 3 | A, B |
Twenty prompts, three engines and three runs each make 180 checks a month. That is a few hours.
Want your brand cited in AI answers?
Our AI search optimization work covers content structure, schema and E-E-A-T signals, measured every month.
See AI search optimizationWhat Do Tracking Tools Add?
Dedicated tools automate the same process at larger scale.
| Manual | Tracking tool | |
|---|---|---|
| Cost | Time only | Subscription |
| Prompt volume | Tens | Hundreds or thousands |
| Runs per prompt | A few | Many, scheduled |
| Engines | Those you test | Several, including via APIs |
| History and charts | Your spreadsheet | Built in |
| Competitor comparison | Manual | Automatic |
| Insight into why | Higher; you read every answer | Lower; you see numbers |
Points to check before buying:
- How are answers collected? Through the consumer interface or an API? The two can differ.
- How many runs per prompt?
- Which engines and countries?
- Can you export raw answers to verify?
- How is “visibility” calculated? Proprietary scores hide method.
Start manually. You will learn what matters, and can then judge whether a tool is worth it. Several SEO suites now include AI tracking modules.
What Is Share of Voice in AI Search?
Share of voice is your share of all brand mentions across a set of prompts.
Formula: your mentions ÷ total mentions of all tracked brands × 100.
If, across 100 prompt runs, your brand is named 30 times and three competitors are named 50, 40 and 30 times, total mentions are 150, and your share of voice is 20%.
It is useful because it is relative. Raw mention counts rise and fall with how talkative the model is. Your share against competitors is steadier and easier to explain to a manager.
Limit it to prompts where a brand recommendation makes sense. Mixing in informational prompts dilutes the figure.
Which Metrics Are Worth Reporting?
Report a few, with context.
- Mention rate on category and buying prompts.
- Share of voice against named competitors.
- Citation rate, and which of your pages are cited.
- Accuracy issues found and fixed.
- AI referral sessions and conversions.
- Top third-party sources the engines cite in your category.
The sixth is the most actionable. If engines keep citing two review sites and one industry publication, those are where you need to be present.
Avoid reporting a single blended “AI visibility score” without showing what is behind it.
How Do You Track AI Referral Traffic?
In your analytics tool, look at sessions by source.
- chatgpt.com and related OpenAI domains.
- perplexity.ai.
- copilot.microsoft.com.
- gemini.google.com.
Create a channel group or segment for these, and track sessions, engagement and conversions.
Two caveats:
- Google’s AI Overviews and AI Mode are not separable. Google says their traffic is counted in the Web search type in the Performance report.
- Some visits arrive without a referrer, for example from apps, and appear as direct traffic.
Server logs add another view: how often AI crawlers and user-triggered fetchers request your pages. Perplexity, for example, documents its agents and their purposes.
How Often Should You Test Prompts?
Monthly suits most businesses. Models and their search behaviour change, but not daily.
| Frequency | When |
|---|---|
| Weekly | During a campaign, or after a major model release |
| Monthly | Standard |
| Quarterly | Review and refresh the prompt set |
| On demand | After a rebrand, a pricing change or a PR event |
Keep the test conditions the same each time: same prompts, same engines, same number of runs, similar time of day.
How Do You Act on the Results?
Tracking is useful only if it changes what you do.
| Finding | Action |
|---|---|
| Never mentioned on category prompts | Check crawler access; build third-party presence |
| Mentioned but never cited | Improve your own pages: direct answers, data, clarity |
| Described inaccurately | Fix source pages and profiles; state facts plainly |
| One competitor dominates | Study which sources mention them |
| Engines cite the same few sites | Earn coverage on those sites |
| Cited for one page only | Apply that page’s structure to others |
| AI referrals convert well | Prioritise the cited pages |
See our guide to generative engine optimization for the methods, and digital PR for earning coverage.
Which Mistakes Should You Avoid?
- Drawing conclusions from one run.
- Testing while logged in, with memory and history active.
- Changing the prompt set each month.
- Tracking only brand prompts, where you will naturally appear.
- Treating a tool’s score as fact without reading answers.
- Ignoring accuracy.
- Reporting AI visibility without any link to business results.
For engine-specific detail, see how to rank in ChatGPT and Perplexity SEO.
Frequently Asked Questions
What is share of voice in AI search?
It is your brand’s share of all brand mentions across a fixed set of prompts. Divide your mentions by the total mentions of every tracked brand, and multiply by 100. Because it is relative to competitors, it is steadier than raw counts and easier to report.
How often should I test prompts?
Monthly is enough for most businesses, using the same prompts, engines and number of runs each time. Test more often during campaigns or after a major model release, and review the prompt list itself each quarter to keep it relevant.
Why do AI answers change between runs?
Language models generate text probabilistically, so the same prompt can produce different wording and sources. Answers also vary with location, conversation history, memory settings and which live search results were retrieved. That is why each prompt should be run several times and results reported as rates.
Do I need a paid tool to track AI visibility?
No. A spreadsheet and a few hours a month can track a set of twenty to fifty prompts across the main engines. Paid tools add scale, scheduling and history. Start manually, so you understand the data before deciding whether automation is worth the cost.
Can I see AI traffic in Google Search Console?
Not separately. Google counts traffic from AI Overviews and AI Mode within the Web search type of the Performance report. Visits from other assistants, such as ChatGPT and Perplexity, appear in your web analytics as referrals, not in Search Console.
Related Guides
Want this done for you?
See what each monthly plan includes, or send us your site and questions on WhatsApp. We do not promise rankings; we show the work and the data every month.
View SEO packages Chat on WhatsAppSources



