AI Visibility Benchmark (Research)

Growth Torque B2B AI Visibility Benchmark

Most claims about AI visibility cannot be checked. We publish the method so you can inspect it, repeat it and challenge it. Published data is the kind of original source other sites and AI tools can cite.

  • 40Buyer questions per company
  • 5Question types
  • 4AI tools tested
  • 3Runs per prompt, per tool
  • 480Answers per company per cycle
  • MonthlyRepeat runs, so trends are reported

Study design

How was the study designed?

Each company in the study received a basket of 40 buyer questions in five types.

  • 10questions

    Category

    Example pattern
    "Best [category] providers for [industry]"
  • 8questions

    Comparison

    Example pattern
    "[Vendor A] vs [Vendor B] for [use case]"
  • 8questions

    How-to

    Example pattern
    "How to choose a [service] supplier"
  • 8questions

    Proof

    Example pattern
    "Is [vendor] reliable for [use case]?"
  • 6questions

    Brand

    Example pattern
    "What does [company] do?"

Run protocol

What is the run protocol?

Same wording, every tool, every run, logged and double-coded.

  1. Every prompt is run 3 times in each of 4 tools (ChatGPT, Gemini, Perplexity, Claude). That gives 480 answers per company per cycle.
  2. Prompts use identical wording across tools and runs.
  3. Each answer is logged with the tool, model version, date and whether web search was on.
  4. Two reviewers code each answer for mention, citation, accuracy and position. Disagreements are resolved by a third.
  5. Runs are repeated monthly so trends, not single snapshots, are reported.

Metrics

Which metrics are reported?

Mention rate, citation rate, accuracy and position score are combined into the AI Visibility Index:

AVI = 0.40 × mention rate + 0.25 × citation rate + 0.20 × accuracy + 0.15 × position score.

Limits

What are the limits of this study?

Read every figure with these three caveats in mind.

  • AI answers vary between runs, tools and model versions, so every figure is a snapshot with a date.

  • Tools with live web search may answer differently from tools without it.

  • Correlation is not proof: pages that rank well may differ from competitors in many ways at once.

Do it yourself

How can you repeat this for your own company?

Four steps, using the same method we publish here.

  1. Write 40 buyer questions using the five types above.

  2. Run each three times in each tool and log every answer.

  3. Score mention, citation, accuracy and position.

  4. Calculate AVI and repeat monthly.

Our free audit does this for you and includes a competitor comparison. For the step-by-step version, read What Is a Prompt Basket?

FAQ

Frequently asked questions

Want the step-by-step version?

The guide walks through building and scoring your own prompt basket.

Read the Prompt Basket Guide

Why run each prompt three times?

AI tools can give different answers to the same question. Three runs per tool reveal how stable a mention is, and the average is more reliable than a single answer.

See how your company scores

Our free audit runs this method for you and includes a competitor comparison.

  1. A prompt basket for your marketBuyer questions across the five types
  2. Your AI Visibility IndexCompared with your competitors
Book a Call