Performance by Topic: where AI recommends your brand and where you disappear
An average visibility score tells you how much you show up, not where. Performance by Topic splits your category into topics and measures each one separately, against every competitor you declared.
Short answer
Performance by Topic is the Refractia module that splits a brand’s category into topics, such as price, trust, speed or support, and measures each one separately: AI incidence (your share of the mentions), your position against each declared rival, sentiment, and the split between chat engines and AI search. It shows which parts of the conversation AI recommends you in and which you’re missing from, something the overall average hides.
Why the average hides where you lose
People don’t ask AI about “your category” in the abstract. They ask which option is cheapest, which one they can trust, which delivers fastest or which has the best support. Each of those questions is a separate competition, and AI can recommend you in some and not name you in others.
Illustrative example
A fictional digital bank has an average AI incidence of 30%. Broken down by topic: 55% on fees, 40% on ease of use, 20% on security and 5% on customer support. The 30% looks acceptable; the 5% on support says that when someone asks which bank responds best, AI recommends others. The data is invented for this example.
With the average alone, the team doesn’t know what content to write or which rival to write it against. With the topic breakdown, it knows both.
What it measures in each topic
Topics come from your brand’s category, which Refractia builds in Identity & DNA by reading your website, and each topic is measured separately with questions in your market’s language. For each topic, the module shows:
- AI incidence. Your share of brand mentions in that topic’s answers, with the sample in view (k of n).
- Position against each rival. Whether you’re above or below each declared competitor in that topic, not just against the average of all of them.
- Sentiment per topic. The tone used to describe you in that topic, from 0 to 100: 70 or higher is positive, 40 to 69 mixed, below 40 negative. A brand can be praised for its price and criticized for its support in the same scan.
- Chat and AI search separately. How much of your incidence in the topic comes from chat engines and how much from Google AI Overviews.
- Comparative radar. Your footprint across all topics next to a rival’s, to see at a glance where you overlap and where you don’t.
Chat and AI search, topic by topic
The split between chat and search is more useful per topic than in the total. Chat engines (ChatGPT, Claude, Gemini and Meta AI) answer from what the model knows; Google AI Overviews summarizes what ranks on the web for that search in your country. In a given topic you might appear only through search, because the engine paraphrases pages of yours that rank well, while the chat models don’t associate you with that topic yet.
If you appear only through search, your presence in that topic depends on those pages and on who else publishes about it. If you appear only in chat, the model knows you but other pages win the search. Each case calls for a different action, and the average blends them.
How teams use it
- Content priorities by topic. Topics with low incidence are the first candidates for the content plan, which turns into tasks in Strategy and content. The Demand module adds whether your site covers those topics.
- Which rival owns which topic. Position by rival shows, for example, that one competitor dominates “price” and another “trust.” That decides who each piece is written against.
- Messaging and service. A topic with recurring negative sentiment can be a communication problem or a problem with the service itself. The data doesn’t solve it, but it shows where to look.
- Briefs for agencies and PR. “Move up on support against Rival B” is a more useful brief than “improve AI visibility.”
- Public figures. For leaders, candidates or executives, the module measures issues on the agenda instead of product attributes.
Performance by Topic is one of the capabilities built for enterprises, alongside product lines and measurement in several countries. If your brand has several lines, each line has its own topics.
Frequently asked questions
What is AI visibility by topic?
Your presence in AI answers measured separately for each topic in your category, such as price, trust or support, instead of a single average. It shows which topics AI recommends you for and which it barely names you in.
How do I know which topics AI recommends my brand for?
Group your category’s questions by topic and measure your share of mentions in each group against your competitors. Performance by Topic does that breakdown across five engines and marks your position against each rival.
What is AI incidence?
Your share of brand mentions in the answers for a topic. It’s shown with its sample, for example 12 of 40 mentions, so you can see how much evidence is behind it.
Where do the topics come from?
From your brand’s category, which Refractia builds from your website in Identity & DNA. They are the dimensions people use to compare options in that category.
How is it different from overall share of voice?
Overall share of voice is a single number for the whole category. Performance by Topic breaks it into topics and adds position by rival, sentiment and the chat-versus-search split for each one.
Keep reading
Start by finding out what AI says about your brand
Your customers already ask AI what’s best in your category; today you don’t know who it recommends. We’ll show you what ChatGPT, Claude, Gemini, Google and Meta AI say about your brand and your competitors. Free, no commitment, and we reply within one business day.