State of AI Search: How the Engines Cite

Key findings

1. Does the engine cite at all?

Before "how many sources," there's a prior question: does the answer cite anything? For most engines the answer is almost always yes — but not for Gemini, which returns a source on well under half its answers. Google AI Overview sits lower too, because it only cites when it appears at all.

2. How many sources per answer — and the quarter's compression

When they do cite, engines differ ~5×: ChatGPT backs an answer with about a dozen sources, Gemini with two or three. And the number is moving — the chattiest engines compressed sharply over Q2–Q3 2026, converging toward the ~10 band, while the already-lean engines barely moved.

Grok is excluded: it cited an extraordinary ~54 sources per answer through mid-May, then left the tracked panel, so it has no comparable quarter-end figure.

3. What AI actually cites

Pool the cited domains across the conversational engines and a clear pattern emerges: the top of the list is social and user-generated platforms, not traditional publishers. YouTube leads by a wide margin, followed by Google and Reddit.

Caveat, stated plainly: cloro's monitored prompt mix skews commercial, so product and retail domains (IKEA, Shopify, Nike, and similar) are over-represented here versus a neutral query set. The platform sources — YouTube, Google, Reddit, Instagram, Facebook — lead regardless of that skew, and are the durable signal.

Each engine has a citation personality

4. How concentrated is each engine?

Concentration is the share of an engine's citations that comes from just its top 10 domains. High concentration means the engine leans on a small, predictable set of sources; low means it spreads across a long tail. Google AI Mode is in a class of its own — more than half its citations come from ten domains, most of them Google's.

What this means

"AI search visibility" collapses six very different behaviors into one phrase. On Gemini, the fight is often to get a citation to exist at all. On Google AI Mode, the citable surface is dominated by Google's own properties — a hard set to break into from outside. On ChatGPT and Perplexity, there are a dozen slots per answer, but the number is shrinking, so the bar to earn one keeps rising. And across all of them, the sources that show up most are platforms — YouTube, Reddit — not the publisher homepages most SEO strategies target.

The practical takeaway: pick the engine that matters for your audience and optimize for how it actually cites, because a strategy tuned to ChatGPT's dozen publisher-and-forum citations will miss on Gemini's sparse answers and again on AI Mode's Google-owned set.

Methodology

All figures come from cloro's production monitoring corpus via Redash.Citation presence (share of answers with ≥1 source) and citations per answer are read from the populated source-count field: presence over a recent 3-day window (~4,600–6,100 answers per engine), and the per-answer trend from the MAR-149 weekly panel over a 120-day window (Mar 9 – Jul 6, 2026).Cited domains and concentration are extracted from each answer's source list over a 3-day window (9,400–25,100 citation instances per engine). Google AI Overview is excluded from the domain and concentration cuts because its citations are organic result links rather than an answer-level source list. Because the customer prompt mix shifts and skews commercial, read the levels as directional and the cross-engine differences and trends as the signal.

More studies from cloro's monitoring corpus are in the research index. For how each engine formats and places its citations, see LLM Citations; for whether blocking AI crawlers costs you citations, the AI Crawler Blocking Report.