Why does ChatGPT sometimes cite no sources?

Because it did not search the web for that answer. In Prefer's September 2026 run, 3 of 50 prompts were answered this way, since short definitional or opinion questions are often answered from the model's training, which produces no citations. When ChatGPT does search, it usually shows several sources.

Does a higher source count mean a better answer?

Not necessarily, and Prefer tracks whether your brand is among the cited sources, not the raw count. More sources often just means a broader, research-style question. A precise factual answer may cite one authoritative page. Judge the answer by whether the cited sources are credible and whether your brand is among them, not by the raw count.

Do other AI engines cite the same number of sources?

It varies by engine and by question, so there is no single number that holds everywhere. Prefer measures it across five engines on the prompts that matter to you, which is more reliable than assuming a fixed count.

How many sources does ChatGPT cite in a typical answer?

About three to four: 3.8 on average across 47 ChatGPT answers in Prefer's measured run, from zero to nine. Why the count changes with the question.

Javed Khatri Co-founder, Prefer

5 September 2026 Updated 2 October 2026 2 min read Answer

The short answer

How many websites does ChatGPT link to in an answer?

About three to four. In Prefer's September 2026 run of 47 web-grounded ChatGPT answers, it cited an average of 3.8 sources each (177 citations across 134 unique URLs), with a typical answer citing 3 and a range from 0 to 9. Prefer tracks whether ChatGPT cites your site.

Key takeaways

In Prefer's 2026-09-02 measured run of 50 tracked prompts on ChatGPT with web search on, 47 answers were web-grounded and together cited 177 sources across 134 unique URLs, an average of 3.8 per grounded answer. (Prefer AEO loop run, 2026-09-02 (ChatGPT))
In the same run, the median grounded answer cited 3 sources and the range was 0 to 9, so the count varies with the question rather than being fixed. (Prefer AEO loop run, 2026-09-02 (ChatGPT))
Of the 50 tracked prompts, 3 were answered from the model's memory with no web search at all (ungrounded), and one grounded answer cited zero sources, which is why some ChatGPT answers show no citations. (Prefer AEO loop run, 2026-09-02 (ChatGPT))

What our measured run found#

In Prefer’s measured run, 47 web-grounded ChatGPT answers cited 3.8 sources each on average. On 2 September 2026, we sent 50 tracked buyer prompts to ChatGPT with web search on. 47 of the answers were grounded in a live web search; those 47 answers cited 177 sources in total, drawn from 134 unique URLs. That works out to about 3.8 sources per grounded answer.

Averages hide the spread, so here is the shape of it:

The typical answer cited 3 sources. That is the median, and it is the number to picture when you imagine a normal ChatGPT answer with citations.
The range ran from 0 to 9. Some grounded answers leaned on a single page; the most source-heavy pulled from nine.
Some answers cited nothing. One grounded answer showed no sources, and a further 3 of the 50 prompts were answered straight from the model’s memory with no web search at all.

So “how many sources” has no single fixed answer. Three to four is the honest center of gravity, with plenty of variation on either side.

Why the count changes from answer to answer#

The number of sources tracks the question, not a setting. A narrow factual question needs one good source or none. A comparison or research question (“best tools for X”) pulls from many, because ChatGPT is assembling a shortlist from several pages. When you see a long citation list, it usually means the question invited breadth.

Two practical takeaways follow:

Do not chase the citation count as a metric. What matters is whether the sources ChatGPT trusts for your buyers’ questions include you. Being one of three cited sources on a high-intent question beats being absent from a nine-source answer. Our study of who gets cited in AI search shows which kinds of sources tend to make the list, and our chapter on how AI engines choose their sources explains why.
Measure it on your own prompts. The only count that matters for your brand is how many sources ChatGPT cites on the questions your buyers actually ask, and whether your domain is among them. General averages set expectations; your own prompt set sets strategy.

That is exactly what an audit measures: for the questions that matter to you, which sources ChatGPT and the other engines cite, and whether you are one of them. Run a free AI visibility audit to see your own numbers.

Sources