Our AI Citation Share Across Google AIO, Perplexity, and Claude

Most "AI SEO" content is speculation dressed up as strategy. This is not that. Below is a direct measurement of how often one site, ours, gets cited across three answer engines for the same 23-query seed set, plus a structural look at everyone else who shows up in those answers.

The citation share numbers

We ran the same seed set of queries against three engines and logged whether kompella.io appeared as a cited source in the answer.

Perplexity cites us in roughly seven out of every ten relevant queries. Claude and Google AIO land almost identically, both just under half. That gap is worth sitting with: it is not a small variance. Perplexity's retrieval and citation behavior is different enough from Google's and Claude's that a site optimized for one is not automatically optimized for the other two. If you are only checking one engine's citation behavior, you are missing the other two thirds of the picture.

Claude and Google AIO landing within a fraction of a point of each other, 47.8% against 47.6%, is the other thing worth noting. Those two systems are built entirely differently. At this sample size that is either coincidence or a sign that both are pulling from a similar layer of the web's authority signals, and 23 queries is not enough to tell you which. We are flagging it as something to watch across more runs rather than a finding to build a theory on.

Treat every percentage here as what it is: the citation rate on this specific seed set, on this specific date, for these three engines. A 23-query seed set is a starting instrument, not a definitive index of anyone's AI visibility. It will move.

None of these numbers are "we rank number one." They are "out of the queries we tested, this fraction of answers named us as a source." That is the metric that matters for AI visibility, and it is the metric almost nobody is publishing with real figures attached.

What the citation landscape actually looks like

Citation share only means something in context. Here is the context, measured from the same run.

The field is wide, not narrow. Across the full query set, 252 distinct domains got cited somewhere. This is not a two-horse race between you and one obvious rival. It is a sprawl.

The top of that sprawl is not dominant. The single most-cited domain across the whole set appeared in 37 citations spanning 15 of the queries, at an average rank of 9.3. Even the domain that shows up most often is not showing up in most queries, and when it does show up, it is not landing near the top of the citation list. Being "the most cited" in this landscape still means missing the majority of queries and averaging a rank deep in single digits. Getting cited a lot and getting cited prominently are two different wins, and the domain at the top of this field has one of them, not both.

The distribution has a long tail. Most of the 252 domains that get cited anywhere appear only once or twice across the entire set. A handful of domains show up repeatedly; the rest are one-off citations, probably because they happened to have the one page that matched the one query. This is the standard shape of a fragmented information market: a few domains with some recurring presence, and a very long tail of incidental appearances.

Some of the most-cited domains are not practitioners. Of the ten most-cited domains in this landscape, three are general platforms or job boards rather than firms or individuals actually doing the work being asked about. That is a meaningful chunk of the top-ten citation real estate going to sites that are not answering the query from expertise, they are answering it from being large, general, and already indexed everywhere. If you are building content strategy on the assumption that AI engines only cite domain experts, that assumption is wrong for at least a slice of the top results.

Why this matters more than your Search Console dashboard

We also pulled the raw Search Console numbers for the same site over the four weeks ending August 14, 2026: 35,024 total impressions, 122 clicks, a 0.35% CTR. Split by traffic classification, human impressions were 17,152 with 51 clicks (0.30% CTR), and bot/agent impressions were 2,736 with zero clicks. The polluted share of classified impressions, meaning traffic identified as bot or agent activity rather than a person, was 13.8%.

That 13.8% is not noise you can ignore. It means roughly one in seven of the impressions Search Console would otherwise have you believe are "reach" are not humans at all, they are crawlers and agents. If your CTR math is not separating human from bot impressions, your real click-through performance is better than your blended number suggests, but your real audience is smaller than your total impressions suggest. Both things are true at once, and conflating them is how teams end up either over- or under-investing in content that AI engines are actually reading, even if no human ever clicks.

There is a blunter way to put this. A CTR this low is what happens when your content gets pulled into an AI answer, summarised and cited, without a human ever needing to open your page. The impression counts. The click mostly does not. If your reporting stack still treats organic clicks as the primary success metric, you are structurally blind to the channel actually consuming your content, which is the same shift behind agent traffic reshaping how content gets read. Citation share, not CTR, is the number that tells you whether an engine trusts your site enough to name it.

This is the coverage behind these numbers: 120 pages, 500 distinct queries, 1,298 page/query rows in the snapshot. That is a big enough sample to say the CTR and pollution figures are not a fluke of one page or one week.

What to do with this if you are not us

1. Stop measuring AI visibility with one engine. A near-15-point-plus gap between Perplexity and the other two, in the same query set, means single-engine tracking will mislead you about your actual position.
2. Don't assume "most cited" means "cited often." The top domain in a 252-domain field was still absent from the large majority of queries. A landscape this fragmented rewards being present across many narrow queries more than it rewards owning one broad one.
3. Separate bot impressions before you trust your CTR. A blended CTR across human and bot/agent traffic understates human engagement and overstates total reach at the same time.
4. Expect job boards and platforms to eat some of your citation share. Practitioner content is not competing only against other practitioners. Some of the top-cited slots go to large general platforms by default.

The habit underneath all four: track citation share per engine, per query, over time, and do not blend them into one "AI visibility" score. A site that is strong on Perplexity and weak on Google AIO needs a different fix than a site that is weak everywhere, and an average erases exactly the information you would act on. That is the same measure-first discipline behind the AI-native operating model we run internally.

If you want a second set of eyes on how your own site's AI citation behavior compares, that is a conversation worth having before you spend another quarter guessing.

---

Need help thinking this through? Book a 30-min call — no pitch. Book a Free 30-Min Strategy Call →

Data and method

Every figure above is read directly from a measurement artifact committed to the kompella.io repository. None of them is an estimate, an industry average, or a round number chosen to sound credible.

.claude/aieo-snapshot.json — snapshot 2026-08-18 05:03 UTC

.claude/seo-snapshot.json — snapshot 2026-08-17 04:31 UTC The snapshots are regenerated on a schedule, so these numbers are a dated reading rather than a standing claim. Quote them with the date attached.