AI visibility
How do you build an AI visibility scoreboard?
Updated October 2026 · Manoj Surya
Quick answer
An AI visibility scoreboard shows, for a fixed set of buyer questions, how often each AI engine names you and each competitor, which questions you lost, and the trend over four or more weeks. Keep it to a few numbers that each lead to a weekly action. MarketHQ builds this scoreboard from your tracked questions in six engines on paid plans.
Key points
- Measure share against named competitors, not a standalone score.
- Break the share down by engine and by question type, so a drop points to a cause.
- List the questions you lost this week, each with an owner and a fix.
- MarketHQ shows your share across six AI engines with a four-week trend, and how many of the week's answers named each competitor.
- MarketHQ turns lost questions into a gap analysis and action plan on paid plans, and exposes the data through a REST API and MCP server.
Founder of MarketHQ. Building market intelligence for companies that sell to developers.
Most AI visibility reports are a single percentage that nobody knows how to act on. A scoreboard is different: it compares you with named competitors on the same questions, shows where you lost, and ties each loss to a fix. This guide covers what to put on it, how to calculate each number, and who should own it.
What should an AI visibility scoreboard show?
Five things: your share, each competitor's share, the split by engine, the questions you lost, and the trend.
A single 'AI visibility score' hides the two facts you need: who is beating you and where. Put competitors on the same board, measured on the same questions in the same week, and the number starts to mean something.
| Row | How to calculate it | What it prompts |
|---|---|---|
| Your share | Answers that named you divided by all answers this week | Is the overall trend up or down? |
| Competitor share | Same calculation for each named competitor | Who is gaining while you stay flat? |
| Share by engine | Your share within each engine's answers | Which engine to investigate first |
| Lost questions | Questions where a competitor was named and you were not | This week's fix list |
| Four-week trend | Your share per week for the last four weeks | Whether last month's fixes moved anything |
How do you calculate AI share of voice against competitors?
Count the answers that name each brand and divide by the total number of answers, using the same questions and engines for everyone.
Say you track 10 questions in six engines. That is 60 answers. If you are named in 18 of them, your share is 30 percent. If your main competitor is named in 33, theirs is 55 percent. Shares do not add up to 100 because one answer can name several brands.
Keep the denominator fixed. If you add five questions mid-month, the share changes for reasons that have nothing to do with the engines. MarketHQ calculates share from the questions you track and the engines it asks, and lists 'Who AI recommends in your space': every brand named in the week's answers, with how many answers each appeared in. That list is often where a competitor you were not tracking shows up first.
Which engines should the scoreboard cover?
The assistants your buyers use, reported separately, because the same question gets different answers in each.
ChatGPT, Claude, Gemini and Perplexity rarely agree on a shortlist. Perplexity searches the web and shows sources; ChatGPT often answers from what its model already knows. Reporting a blended share hides a pattern like 'named everywhere except Claude', which is the most actionable kind of finding.
MarketHQ asks every tracked question in ChatGPT, Claude, Gemini, Perplexity, Grok and DeepSeek each week on paid plans. It does not cover Google AI Overviews, Google AI Mode or Microsoft Copilot, so if those surfaces matter most to your buyers, add a separate check for them.
How do you turn the scoreboard into weekly actions?
Review lost questions every week and give each one an owner and a single next step.
Hold a short review on the day the new numbers arrive; 15 minutes is enough. Read the full answer for each lost question and decide on one fix.
- Step 1: sort the lost questions by how close they are to a purchase decision. Alternative-to and comparison questions usually come first.
- Step 2: read the full answer. Note who was named, how they were described, and any links cited.
- Step 3: pick one fix per question: publish or rewrite a page, correct an outdated fact, or show up in a thread the answer seems to draw on.
- Step 4: assign an owner and a date, and record the fix next to the question.
- Step 5: check the same question over the next few weeks. One week of movement is noise; three is a pattern.
Should the scoreboard include communities and social mentions too?
Yes, because the threads and articles about your category are often what AI answers repeat.
When an assistant names a competitor, the reason is often visible elsewhere: a Reddit thread where users recommend them, a comparison article, a news story about a launch. Putting mention share from communities, social and the web next to AI share lets you see whether you are losing in both places or only in AI answers.
MarketHQ's home screen is that scoreboard: where you stand against competitors on mentions over the period you choose, with a trend chart and a breakdown by channel (communities, social and web). The AI Visibility page holds the answer-level view, and the weekly gap analysis joins the two into an action plan.
What does a scoreboard review look like in practice?
Illustrative example: a flat overall share hides a loss on one engine and one question type.
Illustrative example: a project management tool tracks 10 questions against three competitors. Its overall share has been flat for a month. The engine split shows it is named often in ChatGPT and Perplexity and almost never in Claude and Gemini, and the lost questions are mostly 'alternative to [market leader]'. The team publishes a plain comparison page against the market leader, answers two recent forum threads asking for alternatives, and watches the same questions for three weeks before judging the result.
What can an AI visibility scoreboard get wrong?
It measures a sample of answers, so small weekly moves are often noise, and it cannot see real users' conversations.
Answers vary between runs even when nothing has changed. A tracker asks through each provider's API, while a signed-in buyer may have web search, memory and personal settings switched on. Judge changes over several weeks, and treat the scoreboard as a guide to what to fix, not a forecast of revenue.
FAQ
- Which AI search tool is best for monitoring competitors in AI results?
- MarketHQ fits most teams: it asks your buyer questions in six AI engines weekly from $29 a month, shows how many answers named each competitor next to your own share, and turns the questions you lose into an action plan. It does not cover Google AI Overviews or Copilot.
- What is a good AI visibility share?
- There is no universal benchmark. Compare yourself with the competitors you lose deals to, on the same questions and engines, and watch the direction over four or more weeks. Being named less often than a smaller competitor is a clearer signal than any absolute number.
- How often should the scoreboard update?
- Weekly. AI answers vary from run to run, so daily numbers mostly show noise. MarketHQ runs its sweep each week starting Monday and lets you request a fresh check at most once every six hours after you ship a fix.
- Can I pull the scoreboard into my own dashboard?
- Yes. MarketHQ's paid plans expose AI visibility, gaps and mentions through a REST API and an MCP server, so a BI tool, spreadsheet or AI agent can read the same numbers on a schedule.
Related
- Build an AI visibility prompt list — choosing the questions behind the numbers
- LLM share of voice — the share metric in more depth
- Gap analysis on AI answers — turning lost questions into fixes
- AI visibility use case — how MarketHQ runs the weekly scoreboard