How to read an AI visibility report

Check the questions, models, and completed answers

Twenty questions sent to three models creates 60 question-and-model pairs. That is the planned scope. The number of answers used in a metric can be smaller if responses were excluded or analysis has not finished.

Read the run date, model selection, and completion notices before interpreting a result. A report built from questions about price answers a different business question from one built around product quality. Its percentages describe the captured sample; they do not estimate every conversation taking place in an AI product.

Braign runs selected models through an API. A model's consumer app can have different search tools, instructions, or personalization. Record the exact model and search settings when you share the findings. See how to choose an AI visibility question set for a practical starting point.

Overview: read the four metrics together

The headline tells you how many included answers mention your brand at least once. The four measurements below it answer different questions.

The report's standard share-of-voice view counts every recorded occurrence. Repeating a brand within one answer can increase mention share, while that answer still counts only once toward presence. The worked comparison shows why the two percentages can be far apart.

Named first uses the answers with recorded order. Positive when mentioned excludes unknown tone; judged mentions include positive, neutral, and negative verdicts. Neither denominator is automatically the total number of answers. Read the counts printed below each percentage.

For example, 18 positive verdicts among 20 judged mentions produces a 90% positive rate. If the brand appears in only 20 of 100 answers, that positive rate still sits alongside limited presence. These numbers are illustrative, not a Braign customer result.

Use questions and keywords to locate the explanation

The question breakdown shows where the overall result came from. Look for a group of buyer questions where your brand is absent, then open the captured answers. A question about a use case may produce a different competitor set from a general category question.

The keyword summary helps you find recurring language in the analyzed answers. Treat a phrase as a clue to inspect. It does not supply Google search volume, prove buyer demand, or tell you which wording caused a recommendation.

If a phrase such as “refillable packaging” keeps appearing, find the questions and responses behind it. Check whether the language describes your brand, a competitor, or the category before turning it into a content brief.

Sources & citations: inspect the pages behind the counts

This section groups the cited sources and surfaces pages worth reviewing. You can compare domains, inspect individual URLs, and distinguish your own properties from competitor and third-party sources.

A page cited in an answer that mentions your brand does not necessarily mention your brand itself. Open it before recording a coverage claim. Website ownership may be inferred until the report owner confirms it, and a third-party source is not automatically earned media.

The domain citation mix and the page list also count different things. Two pages on one domain can contribute two page appearances but only one appearance for that domain within the same answer. The citation guide explains how to review those sources without overstating the evidence.

Competitors: find a gap you can investigate

Start with the brands ahead of you by mention share, then select a rival. The report compares answer presence by model, question group, and cited domain, and shows the questions with the largest gaps.

Use those question-level gaps to choose a small set of answers for review. A rival's overall lead may be concentrated in a use case you do not serve. Or it may expose missing product information on a question your sales team hears every week.

Write down the question, the captured answer, the relevant source, and the proposed change. The competitor analysis guide walks through that process.

End the review with a decision and its evidence

A useful report handoff can be short: “Our brand is absent from these four comparison questions. Two cited pages omit the relevant product facts. We will review those pages and update our comparison page.” Include links to the answers that support the statement.

For a recurring report, review the matched questions and models before describing a gain or loss. Comparing runs explains why a missing answer must not become a visibility loss.

Reports are private by default. You can turn on link sharing or export HTML or PDF. Review the saved settings, source reviews, and marketing activity before sharing; those annotations are visible in a shared report.