How do I compare my brand's AI visibility against competitors?

A competitor comparison is a controlled test. Change only the brand, and keep everything else fixed.

1. Pick like-sized competitors. Choose brands with similar reach, category and market. Our playbook on benchmarking against like-sized competitors covers how to pick them.
2. Write one prompt set. Use the questions buyers actually ask: "best X for Y", "X vs Y", "how much does X cost". Use the same wording for every brand.
3. Run each prompt on each engine separately. ChatGPT, Perplexity, Claude, Gemini and Google AI Overviews are different systems. Keep their results in separate columns.
4. Fix the settings. Run logged out, with memory off, from the same location. Note whether each answer showed citations.
5. Record two things per answer: which brands were named, and which URLs were cited.

The result is a grid of prompts by engines. That grid tells you where a rival wins and you don't, which is the only part you can act on.

Citations in AI answers are concentrated on big, well-known sources. A study of AI search citation patterns (arXiv 2507.05301, data from March to May 2025) found the top 20 most frequent news sources accounted for 67.3% of all citations for OpenAI models. That study covered news sources, but the pattern matters for brands too: prominence and citation volume travel together.

The evidence treats brand prominence as a strong association with no publisher lever. Nothing you ship this quarter makes you as famous as a category leader. So if you benchmark a mid-market brand against the biggest name in the space, the gap you see is mostly the size gap.

What it means: use large brands as context, not as the yardstick. Use like-sized brands for the comparison you'll actually make decisions on. If a peer of similar size gets named and you don't, that gap is far more likely to have a fixable cause.

Filter your grid to rows where a competitor is named or cited and you are absent. Those are your gap prompts. The step-by-step version is in our competitor prompt gap analysis playbook.

Then split each gap by how the competitor got there:

• Named with no citations. The engine likely answered from what the model already knows. That knowledge is frozen at a training cutoff, and nobody can change it this quarter.
• Named and cited. The engine retrieved pages and drew on them. This is the case worth investigating, because retrieval depends partly on access you control.

To separate the two reliably, run the same prompt with live search on and off. Our guide to testing answers with search on and off shows how.

One caveat: your prompt is not the query the engine runs. OpenAI says ChatGPT search typically rewrites your query into one or more targeted queries. Treat your prompt set as a proxy for buyer intent, not a copy of what the engine searched.

For each cited URL on a gap prompt, ask three questions.

Whose page is it?

If it's a third-party page (a review site, a directory, a comparison article), the engine is citing other sites rather than the competitor's own. Being named on the sites engines cite is associated with appearing in the answer. It's an association, not a guaranteed effect, but it tells you where to get listed or corrected.

What specific fact does it state?

Look for the sentence the answer leans on: a price, a number, a one-line definition. Stating a specific fact on a fetchable page gives an engine a sentence it can quote. Check whether your equivalent page states the same kind of fact plainly.

Can AI crawlers read your version?

• Rendering. Vercel found ChatGPT and Claude crawlers fetch JavaScript files but don't execute them. If your pricing only appears after JavaScript runs, those crawlers may never see it.
• robots.txt per bot. OpenAI documents that OAI-SearchBot controls ChatGPT search and GPTBot controls training use, and each setting is independent.
• Firewall or CDN. A firewall can return 403 to a crawler that robots.txt allows.
• Google eligibility. Google says a page must be indexed and eligible for a snippet to appear as a supporting link in AI Overviews or AI Mode.

Don't copy the competitor's schema or page structure. No credible evidence shows markup, chunking or an "AI-friendly" style gets pages cited.

A few conditions decide whether your comparison means anything.

• Keep engines apart. Google says AI Overviews and AI Mode may use different models and techniques. One engine can cite a brand often while others barely mention it, and a blended number hides that.
• Note retrieval state. As of October 2026, ChatGPT searches at its own discretion and Claude.ai search is a user toggle. Any average mixes answers with and without live search.
• Know which metric you read. Share of voice is relative, so it can fall while your own mentions rise. See citation rate vs citation share and our step-by-step share of voice method.
• Expect tools to disagree. Different prompts, settings and engine versions produce different numbers. Here's why AI visibility tools disagree.
• Use first-party data where it exists. Bing's AI Performance report shows the grounding queries Copilot actually ran for your site, though it doesn't show competitor citations.

Next move: if you want a quick baseline before building the full grid, the free checker on this site shows whether engines name you for up to three buyer questions. Then fix the access gaps on your gap prompts first, because those are the ones you control.

What counts as a like-sized competitor? A brand that sells to the same buyers, in the same market, with roughly similar reach: comparable traffic, press coverage and review volume. The point is to hold brand prominence roughly constant so the comparison shows differences you can act on. A household name in your category is useful context, but its lead mostly reflects size, not page-level choices you could copy.

Can I see how often competitors are cited in Bing's AI Performance report? No. Bing's AI Performance report covers your own verified site. Microsoft describes its Citation Share figure as observational and says it does not expose competitor domains. You still get something no prompt tracker has: the grounding queries Copilot actually ran when it cited your pages, which is a good input for your prompt set.

How many prompts do I need for a fair comparison? No published study sets a number. What matters more is coverage and consistency: include the main buyer questions for each product line, use identical wording for every brand, and rerun on a fixed schedule. Answers vary between runs, so repeat each prompt a few times before treating a single absence as a real gap.

Should I copy the structure or schema of competitor pages that get cited? No. The evidence does not support markup or page structure as a citation lever. Ahrefs' matched test found AI Overviews citations fell 4.6% after pages added schema. Copy the substance instead: if a cited competitor page states a clear price or definition, make sure your page states yours plainly, in HTML that crawlers can read without running JavaScript.