ChatGPT Search Cites Fewer Sites

Recent data indicates that ChatGPT Search is citing fewer websites per response, a shift that coincides with the rollout of GPT-5.3 Instant as the default model. While the AI continues to explore each site in depth, it now draws on a smaller number of sources overall. This change has implications for both SEO professionals and website owners who track referral traffic from AI-generated content.

Drop in Unique Domains

French SEO consultancy Resoneo analysed this trend using Meteoria, an AI visibility-tracking platform. Over a 14-week period, Meteoria monitored 400 prompts daily, producing a dataset of 27,000 comparable ChatGPT responses. Their findings show that the average number of unique domains per response dropped from 19 before the GPT-5.3 Instant update to 15 afterwards. Similarly, the number of unique URLs per response fell from 24 to 19.

Interestingly, the ratio of URLs per domain remained consistent at one, suggesting that although ChatGPT now cites fewer websites, it still examines each referenced site thoroughly. The result is a smaller citation surface in each response, meaning the websites that are cited now dominate the answer more than before.

Independent Verification

The pattern was independently confirmed by Jérôme Salomon from Oncrawl. Analysing server logs across multiple websites, Salomon noted a clear drop in crawl volume by ChatGPT-User bots. Some pages are no longer being visited at all, while others are being crawled less frequently. This independent data aligns with Resoneo’s analysis, suggesting that the change is not just statistical noise but reflects a real shift in ChatGPT’s web-citing behaviour.

Why the Change Happened

Resoneo links the decrease in cited domains to GPT-5.3 Instant, which appears to rely more heavily on pre-trained knowledge rather than performing broad web searches. Essentially, the AI is now pulling from a smaller set of sources to generate answers. Earlier research has also shown that AI platforms cite sources differently from traditional search engines. An SE Ranking analysis of 129,000 domains found that referring domains were the strongest predictor of whether a site would be cited, with a threshold effect around 32,000 referring domains. Similarly, Search Atlas reported that the median overlap between Google rankings and ChatGPT citations was only around 10–15%.

Impact for Website Owners

For website owners and marketers, the implications are notable. A 20% drop in cited domains per response reduces the number of sites that gain exposure through AI-generated answers. While the depth of content examined on the remaining sites hasn’t changed, fewer domains now compete for visibility within each ChatGPT answer. For those tracking referral traffic, early March—the time when GPT-5.3 Instant became the default—marks an important date range to examine in analytics.

What’s Next?

Resoneo’s analysis also highlights GPT-5.4 Thinking, the next model update, which is expected to reintroduce broader search fan-outs and include site-specific operators to target trusted domains. However, these behaviours were not part of the GPT-5.3 Instant dataset, so it remains uncertain whether the citation surface will widen again or continue narrowing.

In the meantime, website owners should be aware that ChatGPT-generated responses are citing fewer sites overall, meaning that the domains which continue to be cited may receive increased visibility in each answer. Monitoring these changes is essential for businesses relying on AI-driven referral traffic.