A zero-citation AI answer runs 21 to 54 percent shorter. Whether it also goes blank depends on the engine.
An earlier post established that most AI responses cite something, ranging from 90.7 percent of ChatGPT's responses up to 100 percent of Perplexity's. A small share cite nothing at all. A separate post found that when a response names none of the six brands a panel tracks, it almost always still carries a citation, 95.7 percent of the time. Neither post asked what the uncited responses themselves look like: are they a normal-length answer that simply left its sources out, or a thin, short-circuited one? And does citing nothing predict naming nothing? We pulled every zero-citation response out of the same dataset to check.
Zero-citation answers are shorter on every engine that produces them
Across 270 live audit panels and 8,664 real non-error responses, 293 (3.4 percent) carry zero citations. That rate is not even across engines: ChatGPT 9.3 percent (121 of 1,300), Claude 5.0 percent (133 of 2,680), Gemini 1.7 percent (39 of 2,326), Perplexity 0 percent (0 of 2,358), the same per-engine split an earlier post on citation count already established. Perplexity's zero rate being exactly zero means there is nothing to compare on that engine; its fixed 5-to-10 citation window, documented previously, never lets a response through uncited.
| Engine | Zero-citation avg words (n) | Cited avg words (n) | Gap |
|---|---|---|---|
| Gemini | 303.2 (39) | 664.4 (2,287) | -54.4% |
| Claude | 311.0 (133) | 473.8 (2,547) | -34.4% |
| ChatGPT | 557.9 (121) | 705.9 (1,179) | -21.0% |
| Perplexity | n/a (0) | 251.4 (2,358) | n/a |
On every engine that produces a zero-citation response at all, that response is meaningfully shorter than a cited one from the same engine, not close to the same length. The gap ranges from ChatGPT's 21.0 percent to Gemini's 54.4 percent. That answers the first question: a zero-citation response reads like a thin, short answer the engine wrote without doing the retrieval work, not a normal-length answer that happened to omit its sources.
On Gemini, uncited also means blank. On Claude, it means the opposite.
The second question is whether an uncited response is also more likely to name none of the six brands a panel tracks, what an earlier post called a null response. Here the engines split into two different patterns instead of one.
| Engine | Zero-citation null rate (n) | Cited null rate (n) |
|---|---|---|
| Gemini | 38.5% (39) | 1.5% (2,287) |
| ChatGPT | 4.1% (121) | 0.1% (1,179) |
| Claude | 1.5% (133) | 10.5% (2,547) |
On Gemini, a zero-citation response goes null 38.5 percent of the time, against 1.5 percent for a cited response, a 25 times higher rate. ChatGPT moves the same direction on a smaller scale, 4.1 percent versus 0.1 percent. On those two engines, citing nothing and naming nothing travel together, consistent with a single retrieval failure: the engine did not find sources, and it also did not have a brand name to offer.
Claude does not fit that pattern. Its zero-citation responses go null only 1.5 percent of the time, lower than its cited responses at 10.5 percent. A Claude answer that skips citations is not more likely to come back empty. If anything it is less likely to. That means Claude's short, uncited answers are still doing the job of naming a brand almost every time, which looks less like a failed retrieval and more like the model answering from what it already knows rather than what it just looked up.
By category, the gap holds without a single category driving it
Zero-citation responses are not concentrated in one vertical. Project management software runs 3.9 percent (218 of 5,589), customer support 3.3 percent (63 of 1,937), HR software 1.1 percent (12 of 1,138). The spread tracks each category's engine mix more than anything else, not a category-specific retrieval problem.
Consistent with the two posts it builds on
Restricting this session's re-scoring to organic prompts only reproduces the do-null-responses-still-cite-sources baseline exactly: 488 null responses, 467 of them (95.7 percent) still citing at least one source, with the same per-engine split published there (Claude 99.2 percent, Perplexity 100 percent, Gemini 65.9 percent). That confirms this is the same dataset and the same brand-mention logic, read a new way rather than a different sample producing a coincidentally similar number.
Find out whether the engines your buyers use are retrieving real sources about your category, or answering thin and uncited more often than you would guess.
Get an AI Visibility Audit, $490Methodology
Data drawn from 270 live, search-grounded audit panels (project management, customer support, and HR software brands), each run across four AI engines: ChatGPT with web search, Gemini with grounding, Perplexity Sonar, and Claude with web search. We re-read every real (non-error) response from results.jsonl fresh this session, counted words by splitting the response text on whitespace, and used the raw length of the response's citations array to classify it as zero-citation or cited. Null status came from re-running engine/scorer.ts's brand-mention logic fresh this session against all six brands a panel tracks (the target plus five named competitors) for every response. No development-rail or fixture data is included; all responses came from live engine calls. Data collected June-August 2026.