We Measured AI Visibility Across 102 Brands and 5 AI Engines. Here's What the Data Shows.

What we measured

Most "AI visibility" studies are small: a few dozen prompts, one or two engines, a single run. Ours is the largest openly published measurement we know of, drawn from Ranqo's production tracking between March and May 2026. Every prompt was unbranded — a real category question with no brand name in it — because a branded prompt makes "does AI mention you" tautological.

A disclosure, up front: this is our own data, gathered on the platform we sell. That is a real limitation — we are a vendor measuring the market we serve. We have tried to counter it two ways. First, the full methodology, statistics, and confidence intervals are published openly on arXiv so anyone can check the work. Second, we are reporting the findings that are inconvenient for us too: that your own website barely moves the needle, and that the metric most tools love to sell — sentiment — is mostly noise.

One more honest note. Two of the five engines (Claude and Grok) were measured on a smaller, higher-stature slice of brands, so we do not over-read their individual numbers. Every cross-engine finding below pools all five engines.

Finding 1: AI visibility is a 73 / 44 / 11 ladder

Sort the 102 brands by how established they are — global household names, mid-market challengers, genuinely niche players — and visibility falls in clean steps.

Each step down the ladder costs roughly 30 points of unbranded visibility. A Tier 1 brand shows up in about 73% of category answers without anyone naming it; a niche brand, 11%. The gap is not subtle, and it is not mostly about your website — it is about how much the rest of the web already says about you. The paper reports this with bootstrap confidence intervals and a Kruskal–Wallis test; the short version is that the tiers are real, not sampling noise.

The practical read: stop benchmarking yourself against the giant in your category. Benchmark against your tier, and treat the tier above as a multi-quarter climb that runs through third-party coverage, not a homepage rewrite. For how to set that benchmark without fooling yourself, see our guide to measuring AI share of voice.

Finding 2: your own domain is 2.9% of citations

When these engines cite a source, where does the citation point? Almost never to you.

Across 149,912 citations, a brand's own domain accounted for 2.9%. Corporate and competitor pages — third-party sites describing the category, including your rivals' pages — took 75.2%. The rest is spread across video, media, community forums, reference sites, and review platforms. If your AI strategy is "optimize our own pages and wait," you are optimizing the 3% slice.

In AI search, your own website is the smallest lever you have. The brands that win are the ones other pages talk about.

This is the hardest pill for SEO-trained teams. The lever that moves AI visibility is earned presence on the pages models actually read: the category listicles, the comparison articles, the forum threads, the reference entries. Source Analytics exists to show you exactly which of those domains the engines cite about you, so you can earn your way in rather than guess. It also reframes a familiar SEO idea: the goal is not to rank your page, it is to become the brand those ranking pages name.

Finding 3: the listicle is the highest-leverage page

If third-party pages do the work, which kind of page does the most? We classified citations by content format. One format leads all the others.

Among citations we could classify by format, the ranked listicle — the "best X tools" article — took 35.7%, well ahead of the generic article in second place. In absolute terms that is about 21% of all 149,912 citations from a single format. One good list that ranks you surfaces you across a long tail of AI answers, because the models reach for the same well-structured comparison pages again and again.

You will see higher listicle figures quoted elsewhere; they usually measure a different denominator. Ours is a clean decomposition of content-type citations across 102 brands, so the formats are comparable to each other. The takeaway is the same either way: getting onto, and up, the credible ranked lists in your category is the highest-leverage content play in GEO. What is GEO covers the broader playbook, and Content Lab is how we help build the assets that earn those placements.

Finding 4: sentiment is mostly noise

Plenty of tools sell sentiment dashboards. Our data says be careful with them.

Between consecutive measurements, whether a brand was mentioned at all flipped about 6.8% of the time. How the brand was framed — positive, neutral, negative — flipped 45.5% of the time. Sentiment is roughly 6.7x noisier than mention. A single reading of "AI sentiment" is close to a coin flip; the signal only emerges as a trend across many runs.

Measure mention and position first, because they are stable enough to act on. Treat sentiment as a slow-moving trend, not a number to react to week over week. If a tool shows your sentiment swinging wildly, that is usually the measurement, not your brand.

Finding 5: visibility is 77.5% deterministic, so measure repeatedly

Given all that volatility, you might conclude AI visibility is chaos. It is mostly not.

Across brand, prompt, and engine combinations, 77.5% were deterministic: the brand was either always cited or never cited, run after run. Only the remaining ~22.5% sat in the stochastic middle. That is good news — most of your visibility is a stable, improvable state, not a dice roll — with one catch: you cannot tell which bucket a given prompt is in from a single run. Independent work makes the same point. The recent "Don't Measure Once" paper argues AI visibility should be treated as a distribution, not a one-off observation.

So the discipline is simple: measure the same prompts repeatedly, act on the deterministic majority you can actually move, and watch the volatile tail for genuine shifts. Earlier GEO research found targeted tactics can lift visibility by up to 40% on the right pages — but you only know they worked if you were measuring before and after.

What this means for your brand