To track brand visibility in AI search, measure six KPIs against a fixed prompt set on a fixed schedule: AI share of voice, brand-owned citation share, AI visibility rate, accuracy rate, query coverage rate and cross-platform consistency rate. Record a baseline first, re-run the same prompts quarterly, and report each KPI by platform, because answers and cited sources change between runs.
Last updated: October 29, 2026
Summary: Tracking AI visibility is a process, not a single score. This guide sets out the six KPIs with their formulas, how to build a prompt set, how to record a baseline, how often to re-measure, and what to do when a KPI is low.
In this guide:
- The six KPIs to track
- How to build a prompt set
- How to record a baseline
- How often to re-measure
- What a tracking report should show
- What to do when a KPI is low
- Frequently asked questions
What Are the Six KPIs for Tracking Brand Visibility in AI Search?
The six KPIs for tracking brand visibility in AI search are AI share of voice, brand-owned citation share, AI visibility rate, accuracy rate, query coverage rate and cross-platform consistency rate. Each answers a different question, so no single number describes visibility on its own. The definitions come from 5W's guide to AI share of voice and share of model.
| KPI | Formula | Question it answers |
|---|---|---|
| AI share of voice | Brand mentions ÷ all brand mentions in the competitive set × 100 | How much of the AI answer names the brand? |
| Brand-owned citation share | Citations to brand-owned pages ÷ all citations × 100 | Do AI answers link to the brand's own pages? |
| AI visibility rate | Responses that mention the brand at least once ÷ all responses × 100 | How often does the brand appear at all? |
| Accuracy rate | Brand appearances with no material factual error ÷ all appearances reviewed × 100 | Does AI describe the brand correctly? |
| Query coverage rate | Priority prompts that produce at least one brand appearance ÷ all priority prompts × 100 | Which customer questions does the brand answer? |
| Cross-platform consistency rate | Average, across priority prompts, of the share of tested platforms that name the brand | Does visibility repeat across platforms? |
The figures in this example are an illustration, not company data. If a brand appears in 40 AI answers and 8 contain a material error, accuracy rate is (32 ÷ 40) × 100 = 80%. If the answers contain 400 citations and 52 point to the brand's own domains, brand-owned citation share is 13%.
How Do You Build a Prompt Set for AI Visibility Tracking?
A prompt set is a fixed list of the questions customers ask, written once and reused unchanged in every measurement round. Changing the wording between rounds changes the result, so a trend is only meaningful when the prompts stay the same.
A useful prompt set covers five types of question:
| Prompt type | Example pattern |
|---|---|
| Category | "What is the best [product type] for [use case]?" |
| Comparison | "[Brand A] vs [Brand B] for [use case]" |
| Trust | "Is [brand] a reputable [type of company]?" |
| Problem-based | "How do I solve [customer problem]?" |
| Branded | "What does [brand] offer and how much does it cost?" |
The example patterns are templates, not tested prompts. Name the competitors in the comparison prompts and keep the same competitive set for every round, because share of voice is calculated against it.
How Do You Record an AI Visibility Baseline?
A baseline records every setting of the first measurement, so that later changes can be separated from platform and model updates. Document the exact prompt wording, the platforms tested, the model or mode shown, the test date and location, the competitors included, and the rules used to score recommendations and accuracy.
Test on the platforms your customers use. 5W's AI Visibility Index tests ChatGPT, Claude, Perplexity, Gemini and Google AI Overviews, and it documents its prompts, platforms and testing dates in the methodology that accompanies each report. Google states that AI Overviews and AI Mode may use different models and techniques, so test each experience separately.
How Often Should You Re-Measure AI Visibility?
Re-measure AI visibility on a fixed schedule, such as quarterly, and again after a major announcement or change. An Ahrefs study of more than 43,000 queries, each with at least 16 recorded AI Overviews in a month, found that 45.5% of cited URLs changed between consecutive observations, so a single test describes one moment, not a trend.
5W's banking research reached a related conclusion about platforms. It found meaningful differences between AI platforms, so a single-platform view can mislead. Read the details in the banking AI search analysis.
What Should an AI Visibility Tracking Report Show?
An AI visibility tracking report should show each KPI separately and by platform, with the prompt wording, the test date and the competitive set beside every number. A combined score hides the cause of a change, because one percentage cannot diagnose a weak share of voice, a weak citation share and an accuracy problem at the same time.
A report should include four items:
- The six KPIs, each with a baseline, the current value and the change.
- A platform breakdown for each KPI.
- The cited sources behind the answers, split into brand-owned pages and third-party pages.
- A list of material errors found, with the platform and prompt where each appeared.
Label any percentage used as an example as an illustration unless a documented measurement produced it.
What Should You Do When a KPI Is Low?
Each low KPI points to a different action, so the fix depends on which number is weak. The table maps the KPI to the work and the team that usually owns it.
| Low KPI | Likely action | Usual owner |
|---|---|---|
| AI share of voice | Expand content for long-tail questions, add comparison pages and FAQs for specific buying scenarios | Brand marketing |
| Brand-owned citation share | Publish technical documentation, detailed product pages and original research on your own domain | Technical SEO and content |
| AI visibility rate | Check which prompts return no mention and which third-party sources the answers cite | Growth and PR |
| Accuracy rate | Find the page that carries the error and correct it at the source | Communications and SEO |
| Query coverage rate | Write pages that answer the uncovered priority questions | Content |
| Cross-platform consistency rate | Compare the sources each platform cites and close the gaps | PR and SEO |
5W's GEO audit checklist lists 20 checks to run once the weak KPI is known, and the AI visibility audit practice page describes how an audit is run.
Frequently Asked Questions
How do I track brand visibility in AI search?
Track brand visibility in AI search by building a fixed prompt set, recording a baseline, and measuring six KPIs on the same platforms every quarter. Report each KPI separately and by platform.
What is the best KPI for AI visibility?
No single KPI is best. AI share of voice shows how often the brand is named, brand-owned citation share shows whether answers link to your pages, and accuracy rate shows whether the description is correct, so a report needs all of them.
How many prompts do I need to track AI visibility?
There is no fixed number. Use enough prompts to cover each of the five prompt types for your priority products and customers, and keep the wording identical in every round.
How often should I check AI visibility?
Check AI visibility on a fixed schedule such as quarterly, and again after a major announcement. A single measurement describes one moment because cited sources change between runs.
Do I need a tool or an agency to track AI visibility?
A tool can run prompts and count mentions and citations. Fixing a weak KPI usually requires changes to pages, coverage and third-party sources, which is separate work from measuring.
