The State of AI Citations, Q3 2026

By Abiot Y. Derbie - Founder, RankEcho - Published July 10, 2026 - CC BY 4.0

Method (before findings)

We publish the method before the results because that is what separates a study from marketing. Everything here is reproducible: the full prompt battery is listed below and the raw per-citation data is a downloadable CSV.

• Battery: 48 prompts across 8 commercial categories (SaaS, Ecommerce, Health, Finance, Legal, Home Services, Education, Travel), spanning four buyer intents (category, comparison, problem, brand). No prompt mentions RankEcho or any site we operate.
• Engines: anthropic, perplexity, gemini, google_aio. Four engines - we did not have an OpenAI key at run time and say so plainly rather than implying coverage we lack.
• Sampling: every prompt x every engine, run twice, 24h apart (July 9, 2026 and July 10, 2026). 384 observations. Run 0 measures the cross-sectional questions; run 1 exists to measure stability.
• Statistic: cited-domain sets are compared with Jaccard similarity (|A∩B| / |A∪B|). Cross-engine agreement is the mean pairwise Jaccard per prompt; stability is the mean Jaccard between the two runs of the same prompt and engine.
• Publishing rules: engine-infrastructure redirects (e.g. unresolved grounding URLs) are excluded from every domain statistic and reported separately. Skipped runs are excluded from denominators and their reasons are disclosed. Answers with no sources are kept as data, not dropped.

Finding 1 — The long tail dominates. There is no oligarchy.

The single most-cited domain is Reddit - but community sites are only 5% of all citations. The ten most-cited domains together account for just 18%; the rest is a very long tail. 81% of citations point to ordinary independent websites rather than to big platforms, review aggregators, or news outlets.

What it means: "get cited by the big platforms" is the wrong strategy. AI engines pull from a wide field of ordinary pages, which is good news for smaller sites - visibility is winnable without being a household name.

Finding 2 — The four engines barely agree with each other.

On identical prompts, the engines cite mostly different sources. Counting only cases where two engines both returned sources, mean overlap is 0.195 - about 20%. Counting an absent answer as non-overlap (the experience a brand actually has), it falls to 0.138. Either way, roughly four of five cited domains are unique to one engine. And it replicated: agreement was 0.195 on day one and 0.201 on day two.

What it means: there is no single "AI visibility." Being cited by Perplexity tells you little about whether Gemini cites you. Any tool that reports one blended score is hiding this - visibility has to be measured per engine.

Finding 3 — Google's two AI surfaces agree with each other least of all.

The lowest-agreeing pair in the entire study is Gemini and Google's AI Overview at 0.129 - two systems from the same company, answering the same question, agreeing less than any cross-vendor pair. The highest pair is Google AI Overview and Perplexity (0.295), which makes sense: both lean on live web-search retrieval. So Google's AI Overview resembles Perplexity more than it resembles Google's own Gemini.

Caveat, stated up front: the AI Overview appeared for a minority of queries, so this pair rests on 15 comparisons. We flag it as a strong signal, not a settled fact, and will retest at larger scale next quarter.

Finding 4 — A third of citations change within 24 hours — but Gemini is far more volatile than the rest.

Asked the same question a day apart, engines returned overlapping-but-shifting sources: overall stability 0.671. But the average hides a wide split. Perplexity and Claude held about three-quarters of their sources steady; Gemini changed nearly two-thirds of its cited sources from one day to the next.

What it means: a single snapshot is not a measurement. If a third of citations turn over in a day - and most of an engine's, for Gemini - then a one-time "visibility score" is reporting a moment, not a position. This is the entire case for continuous tracking.

Finding 5 — Google shows an AI Overview for a minority of commercial queries — but cites generously when it does.

Google's AI Overview appeared for only 40% of our commercial prompts. On the rest, Google showed no AI Overview at all. But when it did appear, it cited about 7.7 sources - right in line with Perplexity and Claude. The often-quoted "Google AI Overview cites very little" is an averaging artifact: it is absent often, not stingy when present.

Limitations

Read these before citing anything above.

• 48 prompts across 8 categories is a probe, not a census. US-English, consumer-commercial intent only.
• Two runs, 24 hours apart, one week. Stability over longer horizons is unmeasured.
• Four engines. No OpenAI/ChatGPT this round - a real gap we will close next quarter.
• Finding 3 (Gemini vs AI Overview) rests on 15 paired observations. Directionally strong, not definitive.
• We measured which domains are cited, not whether a brand was named in prose without a citation. That "mentioned but not cited" question is deferred to Q4, when we will capture answer text.
• AI engines change without notice. This is one snapshot of one week. That is precisely why we are publishing it as a dated, repeating series.

The data

Everything above comes from one downloadable file: every prompt, every engine, every cited domain, with its source-type classification and position, for both runs. Reuse it freely under CC BY 4.0 - a link back is all we ask. If you rerun the battery yourself, the prompt list below is the whole instrument.

Download the dataset (CSV)

Method note: this report front-loads its answer and marks up its own findings with Dataset and FAQ schema - the same extraction-friendly structure our product recommends. We eat our own cooking.

Frequently asked questions