ChatGPT Search Is Citing Fewer Sites Under GPT-5.3, Data Suggests
Apr 6, 2026
ChatGPT Search is citing fewer websites per answer since GPT-5.3 Instant became the default, according to data reported by Search Engine Journal on April 6. Publishers tracking AI referral traffic have been seeing the change land in their analytics over the past two weeks.
The mechanism is straightforward. Earlier ChatGPT Search iterations tended to cite four to seven sources per answer. GPT-5.3’s default behavior skews toward two or three citations, with the rest of the answer synthesized from retrieved content that doesn’t get surfaced as a visible source. Worse for publishers. The ones who do get cited pick up more of the traffic, but the total pool of cited-and-clicked sites shrinks.
For teams doing AI visibility work, the concrete implication. Citation frequency becomes a zero-sum metric faster than it was a month ago. Being one of three sources in an answer is rarer than being one of six, and the competition for those three slots gets meaner on the same prompt volume.