ChatGPT Search Is Citing Fewer Sites, Data Shows via @Sejournal, @MattGSouthern

Why It Matters

The narrowing citation surface reduces the number of sites competing for visibility inside each ChatGPT answer, amplifying traffic potential for those still cited and prompting marketers to reassess analytics around the March model transition.

Key Takeaways

• ChatGPT citations dropped 20% after GPT‑5.3 Instant rollout.
• Unique domains per response fell from 19 to 15.
• Crawl volume on websites decreased, some pages unvisited.
• Sites cited now receive larger share of answer surface.
• Future GPT‑5.4 may expand citation breadth again.

Pulse Analysis

The recent transition to GPT‑5.3 Instant marks a subtle but significant shift in how large‑language models retrieve and surface information. Whereas earlier iterations performed broader web fan‑outs, the new default limits the number of distinct domains consulted per query, effectively tightening the citation funnel. This behavior mirrors broader industry trends where AI providers balance latency, cost, and relevance by curating a more focused set of sources, a strategy that can improve answer consistency but also concentrates exposure among a select group of sites.

For SEO practitioners, the 20% drop in unique domains per response translates into a narrower pool of opportunities to earn AI‑driven referral traffic. Websites that continue to appear in ChatGPT answers may see a disproportionate boost in visibility, while others risk falling off the citation radar entirely. Marketers should therefore audit their referral logs around early March, identify which pages retained citation frequency, and consider reinforcing those assets with structured data, clear attribution, and high‑quality content to maximize capture of AI‑generated traffic.

Looking ahead, the upcoming GPT‑5.4 Thinking model promises to re‑introduce broader search fan‑outs and targeted site operators, potentially widening the citation surface once again. This oscillation suggests that AI citation patterns will remain fluid, driven by model objectives and cost considerations. Companies should adopt a monitoring framework that tracks citation volume, domain diversity, and crawl depth over time, enabling rapid tactical adjustments as OpenAI iterates its search‑augmented offerings.