How to measure brand visibility in AI: methods, metrics and Share of Voice

Your sales director asks: "Does ChatGPT recommend us?" It sounds like a simple question. In practice it is one of the hardest questions in modern B2B marketing, because AI shows no rankings, there is no position one, and every answer is different. Measuring AI visibility requires a different approach from conventional SEO. This guide covers how to do it systematically, from manual tests to professional tooling.

Looking for instructions for a specific platform? There are separate step-by-step guides: how to check brand visibility in Perplexity and how to check brand visibility in ChatGPT. If you want the result without doing the work yourself, consider an AI visibility audit.

Why measuring AI visibility is difficult

The fundamental problem is non-determinism. ChatGPT, Perplexity and Gemini do not return the same answer to the same question asked twice. Language models have a built-in randomness setting and generate a slightly different narrative each time. A company that appeared for one query may not appear for exactly the same query a minute later.

The second problem is the absence of external data. Unlike Google Search Console, where you have impressions, clicks and positions, AI systems provide no analytics about how often your company appears in answers. You have to measure it yourself, by asking questions and recording results.

The third problem is model diversity. ChatGPT, Perplexity, Gemini and Claude are four different systems with different knowledge bases, different data sources and different answer-generation logic. A company visible in ChatGPT can be entirely unknown to Perplexity, and the reverse. Comprehensive measurement requires testing each model separately.

Manual methods, step by step

Step 1: define the test question set

Before opening ChatGPT you need to know what to ask. The test set should mirror real query patterns - the questions your prospective buyers actually put to AI. We recommend a minimum of 15 to 25 questions across three categories:

• Category questions: "Which [industry] companies do you recommend?" - these test general category recognition
• Problem questions: "Who can help me with [specific problem]?" - these test visibility against customer pain points
• Comparison questions: "[Company A] vs [Company B] - which is better?" - these test position relative to competitors
• Specialist questions: "Which [specialisation] companies operate in [industry or region]?" - these test niche visibility

Step 2: how to test ChatGPT

A testing protocol for a single question: (1) Open a new session with no history. (2) Ask the question without extra context. (3) Save the complete answer. (4) Record whether your company appeared (yes or no), in which position it was named (first, second, third or later) and how it was described (neutrally, favourably, with which attributes). (5) Repeat the same question three times in separate sessions. (6) Calculate the presence rate: how many of the three attempts named the company.

Step 3: how to test Perplexity

Perplexity works differently from ChatGPT: it actively searches the web and cites sources. That makes monitoring visibility in Perplexity partly similar to monitoring SEO positions. The protocol: (1) Ask the question in Perplexity. (2) Check the result two ways - whether the company appears in the answer text, and whether your page appears in the Sources section. (3) Record both indicators separately. (4) Test in every language your buyers use, since source coverage differs by language.

Step 4: how to test Gemini and Claude

Gemini: test questions the same way as in ChatGPT. Gemini has a stronger connection to the Google Knowledge Graph, so companies that rank well in Google appear here more often. Note the differences between Gemini and ChatGPT - they indicate where to concentrate the work.

Claude: test and record whether the answer used web search. Feature availability and the model used can depend on the plan, the account and the date. Treat results with and without web access as separate cohorts.

Example test questions for different B2B categories

Test questions have to reflect the real queries your buyers use. Some examples for common B2B categories:

For each category prepare at least 15 question variants - different phrasings, synonyms, questions with and without a location, and questions in each language your buyers use. A broader test set gives a more credible picture of visibility.

Metrics to track: Share of Voice in AI

Conventional SEO metrics do not translate directly into AI visibility measurement. You need new indicators. These are the key metrics worth tracking regularly:

How to calculate Share of Voice in AI

An example for company X: the test set is 20 questions. Company X appeared in the answers to 12 of them, so AI SoV = 12/20 = 60%. For comparison: competitor A at 9/20 = 45%, competitor B at 14/20 = 70%. The reading: company X sits mid-field, behind the leader and ahead of competitor A. Analysing the differences between models can show where to concentrate: if SoV in Perplexity is 80% but only 40% in ChatGPT, the problem most likely lies in on-site content optimisation.

Third-party tools for monitoring AI visibility

Manual monitoring is labour-intensive and hard to scale. Several tools now automate part of the process:

An AI visibility report template

Systematic monitoring needs a structured report. Here is a template you can adapt:

• Section 1 - Executive summary: company AI SoV versus the previous month, AI SoV of the top three competitors, the key changes and the recommended actions
• Section 2 - Results by model: a table of AI SoV for ChatGPT, Perplexity, Gemini and Claude, the month-on-month change, and the best- and worst-performing test questions
• Section 3 - Narrative analysis: how AI describes the company (quotes from answers), which attributes are attributed to it, and which questions produce the best descriptions
• Section 4 - Competitor benchmarking: AI SoV across all monitored companies, position in the narrative relative to rivals, and qualitative notes
• Section 5 - Source tracking (Perplexity): which pages are cited as sources, whether the company's site is among them, and which content is cited most often
• Section 6 - Trend and recommendations: an SoV chart over time (at least three months), the top three recommendations for the coming month, and an estimate of their impact

How to benchmark visibility against competitors

Competitor benchmarking is one of the most important elements of AI visibility monitoring. Without market context, "60% AI SoV" says nothing - it could be excellent or poor, depending on what competitors achieve. How to benchmark properly:

1. 1Choose three to five direct competitors of similar size, with a similar offer, in the same niche
2. 2Use the same test question set for every company
3. 3Test all of them within the same time window, ideally the same week
4. 4Calculate AI SoV for each company on each model
5. 5Identify where your company leads and where it falls behind
6. 6Tie the differences to specific features of competitors' sites and content - what do they have that you do not?

When to commission a professional diagnosis instead of doing it yourself

Self-service monitoring is possible and sensible for smaller companies, or as an initial orientation. A professional diagnosis makes sense when you need deeper analysis and a benchmark. A rough split:

The key argument for professional measurement is consistency. A one-off manual test is a snapshot. Credibility increases when you record complete answers, repeat an agreed set and apply the same coding rules to your brand and to the benchmark.

How often to measure

AI systems are updated - the knowledge base behind ChatGPT, Perplexity and the rest changes over time. New content published on the web reaches training data or is fetched live. AI visibility is therefore dynamic. The minimum monitoring frequency is monthly. If you are actively working on optimisation, check every two weeks. After a significant action such as a major article, a PR campaign or a new case study, check after four to six weeks to see whether anything has moved.

ChatGPT Search is a separate category to monitor

ChatGPT Search is a search mode with real-time web access, available to all users including free accounts. It is a fundamental change from the earlier ChatGPT, which relied solely on training data. ChatGPT Search behaves much like Perplexity: it searches the web, cites sources and generates an answer from current content.

How to test ChatGPT Search: (1) Sign in to ChatGPT. (2) Enable search by clicking the globe icon, or ask a question ChatGPT recognises as needing current information. (3) Ask a category question. (4) Check two things: whether your company appears in the answer text, and whether your page is among the cited sources. (5) Repeat three times in new chats. The result is the percentage of answers containing your brand.

How to test Grok

Grok is xAI's model, available through the X platform and the Grok app. It has become one of the fastest-growing AI systems, and is particularly popular among users active on X. Presence in Grok matters if your customers or partners are active on that platform.

Grok has one distinctive property: real-time access to content published on X. That means your brand's activity there - posts, quotes and discussions - directly shapes what Grok knows about your company. The testing protocol: (1) Open Grok on the web or in the X app. (2) Ask a category question, as in ChatGPT. (3) Record the result. (4) Check with DeepSearch mode enabled, which can produce different results from the standard mode.