ChatGPT Search appears to be citing fewer sites in each answer, and that matters more than it sounds. Resoneo's analysis of 27,000 comparable responses found that after the move to GPT-5.3 Instant, the average number of unique domains cited per answer fell from 19 to 15. The average number of URLs also dropped, from 24 to 19. On top of that, crawl logs analyzed by Oncrawl showed less crawling. Put together, the signal is clear: fewer sites are sharing the citation space inside ChatGPT answers, so visibility is getting more concentrated.
The short version is simple: ChatGPT Search is citing fewer sources per answer than before. Resoneo's data points to two declines after the transition to GPT-5.3 Instant.
Those two numbers describe slightly different things. A domain is a website, while a URL is a specific page. When both numbers fall, it suggests answers are drawing from a narrower set of websites and pages.
That may sound like a modest shift, but it changes the economics of AI visibility. If an answer used to spread attention across 19 domains and now spreads it across 15, four potential winners disappear from the average response.
Crawl is how often systems request and revisit pages on the web. Oncrawl's log analysis showed a drop in crawl activity, which matters because pages that are crawled less often have fewer chances to be discovered, refreshed, or reinforced as citation candidates.
Lower crawl does not automatically mean lower quality answers. A system can crawl less and still return strong results if it is simply more selective. But reduced crawl does mean fewer pages are likely competing for attention at any given moment.
For example, if a brand updates a product page, publishes a new study, or fixes outdated facts, slower or narrower crawl behavior can delay the moment that information becomes visible in AI-generated answers.
It means AI search is becoming less forgiving. When fewer domains are cited per answer, average content is less likely to get a seat at the table.
A practical example: if two competitors publish similar guides, the one with cleaner structure, clearer claims, and easier crawl access may capture the citation while the other disappears entirely from the answer. In a tighter citation market, near-equal pages no longer share visibility as often.
When answer engines cite fewer sources, the real risk is not just lower traffic. It is losing narrative control. If your brand is mentioned less often, or only through third-party pages, customers start learning about you from a narrower and less predictable set of sources.
This is where BotRank's AI Visibility feature matters. It lets teams run reusable prompts across models like ChatGPT, compare visibility over time, and inspect which sources keep appearing in answers. Just as important, it extracts entities, sentiment, keywords, and top cited pages so you can see not only whether you show up, but how you are being described and who is occupying the space around you.
That kind of monitoring becomes more useful when citation slots shrink. If the number of cited domains per answer is falling, you need prompt-level evidence of when you are being displaced, which competitors are replacing you, and which pages are actually driving the answer set.
The goal is not to publish more pages. The goal is to make the right pages easier to discover, trust, and reuse.
One useful example is a comparison or category page. If it contains precise definitions, a concise summary near the top, and well-structured supporting detail, it has a better chance of being reused than a page that hides the answer behind marketing copy.
Per answer, yes: the data suggests fewer sites now share the citation surface inside ChatGPT Search. But that does not mean every brand is equally disadvantaged.
Brands with strong topical pages, clean technical access, and content that earns reuse may actually benefit from this concentration. When the model chooses fewer sources, the winners can become more durable. The problem is for teams that are not measuring it. A tighter citation pool can quietly widen the gap between brands that are consistently selected and brands that are merely indexable.
There is also an important limit here. Fewer citations do not prove that all answers are worse, nor does lower crawl by itself prove why the citation count fell. What the data does show is a narrower distribution of visibility, and that is enough to change how SEO and GEO teams should operate.
Because citation opportunities shrink fast when averaged across thousands of answers. A reduction of four domains per answer means fewer brands, publishers, and pages are getting exposure in the same response space.
A domain is the website, while a URL is a specific page on that website. When both counts fall, it suggests the system is relying on a narrower set of websites and a narrower set of pages.
Not always. A system can crawl less and still perform well, but lower crawl reduces the chances that new or updated pages enter the citation mix quickly.
Start with the prompts that matter to revenue, then track which brands, sources, and pages appear in the answers over time. If you cannot see those shifts, you cannot tell whether your visibility is improving or quietly disappearing.
GEO is the practice of improving how your brand appears in AI-generated answers. If fewer sites are cited, GEO needs to focus on technical accessibility, citation-friendly content structure, and ongoing visibility tracking across models.
The takeaway is blunt: when ChatGPT cites fewer sites, winning one of those slots matters more. If you want to see whether your brand is gaining or losing ground in that tighter answer set, measure it directly with BotRank instead of guessing from traditional rankings.
