Do AI visibility leaderboards that rank brands mean anything?

A leaderboard ranks brands by how often AI answers mention or cite them. That sounds like a performance table. In practice it mostly sorts brands by how famous they already are, filtered through whichever prompts the tool decided to ask.

Three facts drive this:

• Brand prominence swamps everything else. Big brands get cited because they are big, and no published study controls for it.
• Prompt choice sets the order. The tool picks the questions. The engine then rewrites them into queries nobody outside can see.
• There is no ground truth. As of October 2026, Google's generative AI report in Search Console shows impressions only, with no clicks, CTR or position.

So a leaderboard rank tells you little about whether your work is paying off. If you want to know what a single score is built from, read how an AI visibility score is calculated.

AI citations are concentrated in a small set of sources. A 2025 study of AI Search Arena data (arXiv:2507.05301) found the top 20 news sources accounted for 67.3% of all citations for OpenAI models. Its data window was March to May 2025, so treat it as a pattern, not a current figure.

Semrush's index of 126 million prompts (January to April 2026) found 36 brands visible across all four surfaces it tracked every month. They include YouTube, Google, Reddit, Amazon, Facebook, Apple, Walmart, Disney and Nintendo. That finding is brand size wearing a visibility label.

What this means for a leaderboard: if a mid-market brand ranks below a household name, the table has told you nothing new. Third-party authority metrics do not fix it either. They are built from backlinks and traffic, which are proxies for the same size effect. Comparing against them does not remove the confounder.

Every tracker tests prompts it chose. The engines do not answer those prompts directly. OpenAI says ChatGPT search typically rewrites your query into one or more targeted queries before it searches. Google may fan a query out across subtopics. The only first-party window onto the real retrieval queries is the grounding query data in Bing Webmaster Tools.

Retrieval is also discretionary. ChatGPT searches at the model's discretion, and Google's AI Overviews often do not trigger. Any ranking averages answers given with and without live retrieval, usually without saying which.

Platform mix matters just as much. In Otterly's 2026 experiment, Microsoft Copilot produced 2,718 of the 3,934 citations tracked for one test site (69%), while ChatGPT, Gemini and Google's AI features almost never cited it. Microsoft's own report credited that site with 42,817 citations, more than ten times the prompt tracker's count. A blended leaderboard would have hidden all of it.

Leaderboards imply an order that the engines do not publish. Google's AI report has no position dimension. Google also states that AI Overviews and AI Mode may use different models and techniques, so they are not one surface. Citation units differ too: Claude's API cites short text spans, while Perplexity cites whole sources by number. There is no shared denominator to average.

Even within one answer, the order of citations is a weak signal of influence. That is why position isn't the metric for AI search, and why AI rank tracking has no real rank to track.

The C-SEO Bench study (NeurIPS 2025) adds one more caution: as more sites adopt the same tactics, overall gains decrease, which it describes as congested and zero-sum. A forced ranking treats that contest as a scoreboard. It is closer to noise.

A brand comparison can be useful if it meets these conditions:

1. Like-sized brands. Compare against competitors with similar reach. Otherwise you are measuring size.
2. A score per platform. Report ChatGPT, Perplexity, Gemini, Claude and Google AI Overviews separately. Never blend them into one rank.
3. Disclosed prompts. Publish the exact prompts, how they were chosen, and when they ran.
4. Disclosed retrieval state. Say whether answers came with web search on or off, and whether logged-in memory was excluded.
5. Enough answers. Small samples swing widely. A prompt sample is a sample, not a census.
6. Presence and accuracy, not position. Report whether you were named and whether the facts were right.

Microsoft frames its own Citation Share metric as observational, not a ranking system or a competitive scoreboard. Hold every tool to that standard. When you assess vendors, the features that matter in an AI audit tool include these disclosures.

Ignore your leaderboard position. Focus on what you control.

• Check access first. Make sure robots.txt allows the search bots (OAI-SearchBot, PerplexityBot, Claude-SearchBot) and that your CDN or firewall does not return 403 to them.
• Check rendering. Put prices and key facts in the HTML before JavaScript runs. Most AI crawlers do not execute it.
• Check accuracy. Ask the questions your buyers ask and note which facts each engine gets wrong.

The free checker on this site runs up to three buyer questions and shows whether engines name you. It is a snapshot, not a ranking. None of these steps promises a citation. They remove the blocks that keep you out.

Is share of voice in AI answers any better than a ranking? It is better, but only with the same caveats. Share of voice still depends on the prompt set and the platform mix, and its denominator is dominated by whoever already wins the query. Your share can fall while your absolute mentions rise. Read it per platform, against like-sized competitors, with the prompts disclosed, and treat it as an observation rather than a scoreboard.

Can I trust a tool that shows my AI rank in Google? No. As of October 2026, Google's Search Console report for AI Overviews and AI Mode exposes impressions only, with no position field. Google also says third-party tools don't have access to its internal ranking data. Any Google AI position number is a vendor's own construct with nothing to validate it against. Ask the vendor what it actually measures.

Why do small brands rank so low on AI visibility leaderboards? Mostly because they are small. Citation studies show AI answers concentrate on a few prominent sources, and no published study controls for brand size. A low leaderboard rank against household names says little about your site. Compare yourself with similar-sized competitors, and check for fixable problems such as blocked crawlers or facts that only appear after JavaScript runs.

How many prompts does a fair AI visibility comparison need? No published standard sets a number. What matters more is that the prompts are disclosed, reflect real buyer questions, and run the same way for every brand compared. Small samples swing widely between runs. Results should be reported per platform with retrieval state stated, so you can see whether a difference is real or an artefact of the questions asked.