Definition
Citations per response is the average number of source citations an AI search engine includes in a single answer, usually calculated only over responses where the engine actually searched the web. It describes the size of the citation inventory: how many sources, and therefore how many brands, can be cited in one answer.
The number differs a lot by engine. Promptwatch's average sources per response tracking shows ChatGPT citing about five sources per web-search answer, while Google AI Overviews and Perplexity each cite roughly ten. Perplexity is the most stable of the four engines tracked, staying close to ten day after day. Microsoft Copilot is the least stable: its average has swung from under two sources to nearly seventeen within a few weeks.
The count also changes when platforms ship new models or retrieval systems, sometimes overnight. Around the GPT-5.3 rollout on March 4, 2026, ChatGPT's average fell from about 6.4 sources per search-enabled answer to about 4.7 to 4.9 by late March, roughly 27% fewer citation slots, across GPT-5.3, GPT-5.4 and GPT-5-Mini at the same time (see the ChatGPT citation drop report).
Citations per response matters for GEO in three ways:
- It caps your opportunity. With five slots per answer, a page has to be one of the top few retrieved results for the prompt; with ten, mid-authority pages have a realistic path in.
- It shapes other metrics. Fewer sources per answer lowers every brand's citation rate and changes the denominator of citation share, so a platform-wide drop can look like a loss of visibility.
- It changes how much each citation is worth. A source in a five-citation answer gets more attention than one of ten, and source order matters more (see answer position).
To use it, track the average per engine over time and annotate model releases. When your citation numbers fall, check whether the engine's citations per response fell in the same week before changing content. Compare it with citation diversity, which counts how many distinct domains those citations come from: an answer with eight citations from two domains has a high count and low diversity. Also note that formats differ. Some engines attach inline citations to specific sentences, others list sources in a footer, and some also show pages they browsed but did not cite; decide which of these you count and keep it consistent.
Examples of Citations Per Response
- A team sets lower citation-rate targets for ChatGPT than for Perplexity after seeing that ChatGPT lists about five sources per answer against Perplexity's ten.
- An SEO lead explains a sudden March 2026 dip in ChatGPT citations to leadership by showing that average sources per ChatGPT answer fell about 27% in the same week.
- A marketer stops optimizing against Copilot weekly snapshots after noticing its average sources per answer swung between 2 and 17 within a month.
- An analyst counts only inline citations, not footer sources, and documents that choice so month-over-month comparisons stay valid.
