TL;DR
- The number of unique URLs cited has nosedived by 33% on Gemini and 59% on Google AI Mode since late April.
- The phenomenon is not simply a matter of the models citing sources less frequently. Gemini’s citations actually rose month-over-month. Rather, the diversity of sources the models cite is what’s falling.
- There was no observable difference in the types of domains the models cite month-over-month. Similarly, the cutback in the number of unique sources did not create windfalls in source share for any particular domains.
It’s getting harder to be cited by two of Google’s large language models (LLMs). Those sources that do get cited now carry more weight than they did a month ago.
We ran 99,000 prompts across Gemini and Google AI Mode on the Evertune AI marketing platform throughout April and May. From late April to the end of May, the number of unique URLs the LLMs cited fell 33% on Gemini and 59% on Google AI Mode.
That’s a much more meaningful difference on AI Mode, which typically cites 20-27 URLs per response, than Gemini, which typically cites 3-6. AI Mode cited roughly 23,000 fewer unique URLs in May versus April when given the same prompts the same number of times.
The dropoff in unique URLs was not simply Google citing less frequently. Rather, the variety of URLs went down, giving each cited URL more weight. The number of citations per unique URL rose 16% on Gemini and 8% on AI Mode.
Google has made many changes in recent months. The company rolled out two core updates to its search engine, unveiled generative AI performance reports, launched preferred sources, announced AI search agents and a universal cart, and upgraded the default model version of Google AI Mode to Gemini 3.5 Flash. It’s not clear which of these changes, if any, could have caused the LLMs to slash the unique URLs they cite.
It does not appear that the source culling from April to May targeted any domains or domain types in particular. For example, AI Mode had the exact same breakdown of site categories in both April and May.
The cutback on unique URLs also did not significantly benefit any particular domains. Out of more than 20,000 domains that AI Mode cited in both April and May, only one - YouTube - gained even a single percentage point in source share in May. Out of about 2,800 domains cited by Gemini in both months, only three gained a percentage point or more in source share.
For marketers, this is yet another reminder that things change quickly in the AI world, and understanding the sources LLMs use to understand and describe your brand is a long game that requires continued tracking, especially when a model maker and search engine giant decides to shake up everything.
Methodology
At Evertune Research, we track hundreds of brands across 250 categories. For this analysis, we ran 330 distinct prompts repeatedly on 25 days in April and 25 days in May. The prompts spanned multiple unrelated topics, and we ran them on Google AI Mode and Gemini. In total, we ran 99,000 prompts between the two models.
Evertune is the AI marketing platform for brands that want to own the AI customer journey. Evertune analyzes prompt responses at scale across all major LLMs, ChatGPT, Claude, Gemini, AI Overviews and more, to deliver statistically significant visibility data, then closes the loop with tools to act on it: website optimization, data-driven content creation, most influential sources, and paid activation through affiliate and programmatic AI retargeting partners. Where most tools tell you where you stand, Evertune tells you what to do about it. Founded by early executives of The Trade Desk and backed by $20M from leading investors.
Don't forget to share this post
