# How to audit competitor AI visibility

> How to run an AI visibility audit on competitors and map which questions each rival owns across AI engines. Read the guide.

**Category:** Playbook · **Published:** 2026-06-14 · **Updated:** 2026-10-10 · **Canonical:** https://www.zumihq.com/blog/competitor-ai-visibility-audit

## Reading brief

- **Decision:** Which competitive visibility gap deserves the first intervention.
- **Evidence:** Compare mention and prominence patterns by query and engine, then inspect the sources supporting each competitor.
- **Action:** Run a fixed query set, map the gaps and their source drivers, and prioritize one defensible query cluster.


An AI visibility audit measures which competitors AI engines mention, and for which questions. The method is simple enough to run by hand before buying any tool. The output is a map of the questions each rival owns, and that map is sharp enough to shape a real GEO plan.

Most teams start this work after a prompt. A prospect mentions a competitor appeared in ChatGPT. A founder checks Perplexity and finds a rival mentioned in the category answer.

A search for the brand's own category turns up no brand at all. That is a poor place to start. A planned audit is better.

## Key takeaways

- An AI visibility audit needs only the engines, a set of questions, and a spreadsheet. One afternoon of work shows where each rival stands.
- Track four signals per competitor: mention rate, share of voice per engine, average position, and citation share. They stay separate, because one combined score hides which signal moved.
- Run each question more than once per engine. AI answers vary between runs, so a single answer shows a possibility, not a rate.
- Competitors that show up on every engine have strong entity signals and wide third-party coverage. Those that show up on one engine but not others have tuned for that one channel.
- The most useful output is not a ranking. It is a query-ownership map: the questions competitors own that the brand does not, and why they own them.

## What should a competitor AI visibility audit measure?

The audit tracks four signals per brand: mention rate, average position, share of voice per engine, and citation share. Together they show how often a rival appears and where it lands in the answer. They also show which engines favor it and which sources back each mention.

| Signal | What it measures |
|---|---|
| Mention rate | How often the rival appears in answers to the question set |
| Average position | Where it lands in the answer: first, second, or last |
| Share of voice per engine | How its showing differs across ChatGPT, Gemini, and Perplexity |
| Citation share | Which outside sources the engine draws on when it mentions the rival |

A rival that shows up in seventeen of twenty questions has a very different reach than one that shows up in four. Position matters because people read AI answers like prose. The brand mentioned first gets the most attention.

Share of voice varies because ChatGPT, Gemini, and Perplexity [share only a small fraction of their cited domains](/blog/how-ai-engines-rank-brands). A rival strong on Perplexity but absent from ChatGPT has a clear profile: real SEO but weak entity signals. That points to a clear fix.

Review sites, analyst reports, press, and the rival's own content each point to a different kind of spend. Citation share shows which one is paying off.

## How should the question set be built?

A good question set has three layers, shown in the table below: category, use case, and comparison.

Twenty to thirty questions across these layers give a clear picture without eating a week.

| Layer | Example questions | What it shows |
|---|---|---|
| Category | "What is the best [category] tool?" | What buyers ask first, and which brand is the default |
| Use case | "Best [category] for agencies" | Where rivals own a niche |
| Comparison | "[Competitor A] vs [Competitor B]" | How engines frame the field, at the highest intent |

## How is the audit run?

Run each question on ChatGPT, Gemini, and Perplexity. Log each answer in a spreadsheet: the question, the engine, the brands mentioned in order, and the cited sources. A summary tab rolls up mention rate, average position, and share of voice per brand per engine.

Each engine gets its own tab. The summary tab is the AI visibility report: the four signals per brand per engine, the ownership map, and the run count behind every figure.

### How many times should each question run?

Answers to one question change from run to run, so a single answer is one sample, not a rate. Run every question several times on each engine and record how many of those runs mention each brand.

No run count is universal. Pick one, use the same count in every audit, and report it beside the results so a later audit stays comparable.

## What does a query-ownership map look like?

The map lists each question and the brand that most often appears in its answers. The example below is illustrative, using fictional brands, and is not measured data. Each question ran five times on one engine, and a brand "owns" a question when it is mentioned in at least three of the five runs and in more runs than any rival.

| Question | Northlane | Corvex | Ledgerra | Owner |
|---|---|---|---|---|
| Best [category] for agencies | 1 of 5 | 5 of 5 | 2 of 5 | Corvex |
| Best [category] for enterprise | 4 of 5 | 3 of 5 | 1 of 5 | Northlane |
| [Category] pricing comparison | 0 of 5 | 4 of 5 | 3 of 5 | Corvex |
| How to choose a [category] tool | 2 of 5 | 2 of 5 | 2 of 5 | No owner |
| Best [category] for small teams | 0 of 5 | 1 of 5 | 5 of 5 | Ledgerra |

Northlane is the brand being audited. It owns one question, and rivals own three that it does not. One question has no owner, which makes it the cheapest opening.

The next step is the best guess at why each rival owns its questions. Corvex's five of five on agencies points to agency-focused coverage that Northlane lacks, and the cited sources in those five answers show where that coverage lives.

Run the audit monthly and the snapshot becomes a trend line. A shift in a rival's mention rate or position is the first sign it is spending more on GEO, or losing ground.

## What does the audit data reveal about competitors?

Rivals that show up on every engine have strong entity signals and wide third-party coverage. Rivals strong on Perplexity but absent from ChatGPT often have good SEO but weak entity signals. Rivals strong on ChatGPT but absent from Perplexity often have a strong place in training data but old content that live search skips.

| Pattern | Likely cause | Fix |
|---|---|---|
| Strong on all engines | Strong entity signals and wide coverage | Keep the basics strong |
| Strong on Perplexity, weak on ChatGPT | Good SEO, weak entity signals | Entity work and earned citations |
| Strong on ChatGPT, weak on Perplexity | Strong in training data, old content | Fresh content for the same questions |

The output that matters most is not a ranking. It is a list of the category questions a rival owns that the brand does not.

Add a best guess at what the rival did to own them. That guess drives where the GEO budget goes.

## Why doesn't the audit produce one AI visibility score?

A single score would blend four different questions into one number. A rival could lead on mention rate, trail on position, and win on citations, and the score would show only the net.

The signals stay separate so each gap points at its own fix. [Why one visibility score hides the decision](/blog/geo-metrics-beyond-mention-rate) covers the reasoning.

To run the same audit on the brand's own visibility first, see the [AI visibility baseline guide](/blog/baseline-ai-visibility).

Closing the gaps usually starts with content. [How to write content that AI engines cite](/blog/how-to-write-content-ai-engines-cite) covers what to change on the page.

Zumi is an AI Search Intelligence Platform that runs this same audit on a schedule, per engine, instead of by hand once a month. [Book a demo](/book-demo) to see a real competitor set broken down by the four signals.