AI engines differ sharply in how many sources they pull into a single answer. According to Semrush's 2026 AI Visibility Index: an analysis of 126 million US AI search prompts collected between January and April 2026 and released on 26 June 2026. ChatGPT cites about 15 sources per response while Gemini cites about 3. That gap changes your realistic odds of being one of the sources cited on each engine. This is a dated, fully-sourced reference to the numbers and what they mean for answer engine optimisation.
Last reviewed: 12 July 2026. All figures here are from Semrush's 2026 AI Visibility Index (Semrush is an Adobe company) unless noted. It is a single-vendor dataset. Large and useful, but vendor-reported and a snapshot of a fast-moving landscape, so read the direction as firmer than any single number, and cite "Semrush 2026 AI Visibility Index" with the date when you reuse a figure.
How many sources does each AI engine cite per answer?
The headline figures from the index, with the two engines Semrush reported source-count averages for:
| Engine | Avg. sources cited per response | Where it leans | Source |
|---|---|---|---|
| ChatGPT | ~15 | Community + reference (Reddit, Wikipedia) | Semrush 2026 AI Visibility Index |
| Gemini | ~3 | A narrower pool (Wikipedia, YouTube) | Semrush 2026 AI Visibility Index |
| Google AI Mode | Analysed in the index; per-response source count not reported here | . | Semrush 2026 AI Visibility Index |
| Google AI Overviews | Analysed in the index; per-response source count not reported here | . | Semrush 2026 AI Visibility Index |
The index covered ChatGPT, Gemini, Google AI Mode, and Google AI Overviews across more than 1,200 brands in 22 industries. Perplexity was not part of this index: a common misreading, so don't attribute a Perplexity source-count to it. The one-line summary: ChatGPT casts a wide net per answer; Gemini pulls from a short list.
Why does the source count matter for getting cited?
Because it sets how many citation slots exist in a typical answer. An engine that assembles ~15 sources per response has far more openings than one that pulls ~3, so on ChatGPT, broad and well-corroborated coverage of a topic has more chances to land, while on Gemini the few slots are contested and being in the narrow trusted pool matters more.
- More slots ≠ easier, exactly. A wider pool also means more competing sources per answer, so a ChatGPT citation is one of many rather than one of a few.
- Fewer slots raise the bar. Gemini's ~3-source answers reward being an established, corroborated source on the topic. See how to get cited in Gemini.
- The selection filter is the same everywhere. Regardless of count, engines keep passages that are extractable, evidence-dense, scoped, authoritative, corroborated, and fresh. The mechanics in how to get cited by AI.
The practical read: the source count tells you the shape of the opportunity per engine, but not a shortcut. This complements our source-share reference (which sources get cited most often) in the AI search statistics reference. Count is per-answer breadth; share is how often a given domain appears.
ChatGPT pulls about 15 sources into an answer and Gemini about 3, so the same topic is a wide-net contest on one engine and a short-list contest on the other. Being cited on one is no promise on the other.
Are brands visible on every engine, or just one?
Mostly just one. Semrush reported that across the whole study period only 36 brands maintained visibility on every platform it analysed. A group it called the "Universal 36," made up largely of very large platforms (Semrush's list includes the likes of YouTube, Reddit, and Amazon). Everyone else showed up on some engines and vanished on others.
That single finding is the strongest argument for measuring each engine separately. Because the engines pull different numbers of sources from different pools, a brand can be well-cited on ChatGPT and absent on Gemini. The brand-mention gap versus source gap problem, seen across engines. Semrush also noted that on Gemini specifically, the overlap between mentioned brands and cited domains can be as low as 30%. The ghost-citation pattern where you are named without your page being the source (more here).
Does an integrated SEO + AI-visibility workflow help?
Semrush's companion survey suggests it does. Organisations that ran SEO and AI-visibility as one integrated workflow reported far better outcomes than those keeping them siloed:
| Approach | Reported increased traffic or leads from AI platforms |
|---|---|
| Fully integrated SEO + AI-visibility | 81% |
| Siloed (managed separately) | 36% |
Source: Semrush 2026 AI Visibility Index companion survey. Self-reported, so directional. The result lines up with our standing position that GEO is not a separate discipline from SEO: the crawlability, structure, and entity work that earns rankings is the same work that earns citations. Treat the 81%-versus-36% gap as a reason to unify the two functions, not as a guaranteed multiplier.
How should you use these numbers?
Treat this as a dated reference, not a target. The figures are Semrush's own and a January-April 2026 snapshot; citation behaviour shifts month to month as models and retrieval pipelines change. So:
- Cite the source and date ("Semrush 2026 AI Visibility Index") whenever you reuse a figure, and hedge it as vendor-reported.
- Don't chase the count. Aim to be a clean, corroborated source on your topic; the per-engine count then tells you how contested each answer is.
- Measure per engine. Because visibility rarely carries across engines, track ChatGPT, Gemini, AI Mode, and AI Overviews separately. The discipline in how to measure AI visibility.
The number that ultimately matters is not how many sources an engine cites in general, but whether your brand is one of them for the questions your buyers actually ask. Tracked across every engine, over time rather than spot-checked. That measurement is exactly what Buffy Intel is built to provide.