New research published in June 2026 by AI platform Evertune has revealed a striking shift in how Google’s large language models (LLMs) are selecting sources. Between late April and the end of May 2026, the number of unique URLs cited by Google AI Mode dropped by 59%. Over the same period, Gemini saw a 33% reduction. Across 99,000 prompts run on both platforms, the data is unambiguous: the pool of sources being cited is shrinking, and the brands that do get cited are gaining disproportionate authority.

For marketers and SEOs, this is not just a technical footnote. It represents a fundamental shift in how AI-generated answers are constructed, and it demands a rethink of visibility strategy across Google’s growing AI surfaces. Here is what happened, what it means, and what you should do about it.

What the Research Found

The Evertune research team ran 330 distinct prompts repeatedly across 25 days in April and 25 days in May, spanning multiple unrelated topics on both Google AI Mode and Gemini. The results were clear. In the week ending 6 April, AI Mode was typically citing between 20 and 27 unique URLs per response. By the end of May, that number had collapsed by 59%, meaning AI Mode cited roughly 23,000 fewer unique URLs in May than in April when given the same prompts.

Gemini, which typically cites just 3 to 6 URLs per response, showed a 33% reduction in unique sources. The scale is smaller, but the direction is identical. Both Google AI Mode and Gemini are becoming considerably more selective about which content they surface.

Crucially, this is not simply a case of the models citing sources less frequently overall. Gemini’s total number of citations rose month-on-month. What fell was the diversity of sources. The number of citations per unique URL rose 16% on Gemini and 8% on AI Mode. In other words, the same sources are being cited more often, while the overall pool of cited domains is getting smaller.

Why Fewer Sources Means Higher Stakes

If overall adoption figures look uneven, the picture among young people is far more uniform and far more strikinWhen an AI model cites fewer unique URLs but keeps total citation volume similar, each URL in that shrinking pool carries more weight. Being cited in Google AI Mode or Gemini is no longer a statistical probability based on volume of content. It is an outcome that depends on meeting a tighter set of quality signals, and it is becoming more valuable as a result.

For global brands operating across multiple markets, the implications extend further. A contraction in cited sources does not just reduce your odds of appearing. It concentrates AI-generated visibility among a smaller set of authoritative, trusted voices. Brands that have invested in generative engine optimisation (GEO) and authority-building will hold their position or gain share. Brands that have not will find themselves increasingly absent from AI-generated answers, regardless of their traditional search rankings.

The Evertune data also showed that no particular domain type benefited from the contraction. The breakdown of site categories cited by AI Mode was identical in April and May. Out of more than 20,000 domains cited by AI Mode in both months, only YouTube gained even a single percentage point in source share. This was not a cull that rewarded any one sector. It was a general tightening of standards.

What Changed at Google

The timing of the citation drop coincides with a period of significant activity from Google. Between April and May 2026, the company rolled out two core algorithm updates, launched preferred sources in AI Overviews and AI Mode, introduced generative AI performance reports in Google Search Console, announced AI search agents, upgraded the default model powering Google AI Mode to Gemini 3.5 Flash, and launched a universal shopping cart.

It is not yet clear which of these changes, individually or in combination, triggered the citation contraction. The preferred sources feature, which allows publishers to signal content they want surfaced in AI responses, is a plausible factor. The upgrade to Gemini 3.5 Flash may have introduced different weighting for source selection. The core updates may have adjusted the underlying quality signals feeding into both the search index and the LLMs.

What the data does confirm is that the change happened quickly, it was significant in scale, and it affected both AI Mode and Gemini simultaneously. For brands tracking AI visibility, this is a reminder that the landscape can shift materially in a matter of weeks, not quarters. 

What This Means for Your GEO Strategy

Generative engine optimisation is no longer an emerging discipline. It is a core component of any credible search strategy in 2026. The Evertune findings reinforce what practitioners have been observing: AI models are not selecting sources at random. They are applying quality filters that prioritise authority, relevance, and structural clarity.

The key implication of a smaller citation pool is that the bar for inclusion has risen. Your content needs to do more than exist. It needs to demonstrate expertise, answer questions directly, and be structured in a way that AI systems can parse and trust. Google’s own 2026 guidance confirms that optimising for generative AI features is still, fundamentally, SEO. But it is SEO applied with greater precision and an understanding of how LLMs evaluate and retrieve information.

For brands operating internationally, the challenge is amplified. The same quality thresholds apply in every market, but what constitutes authoritative, culturally relevant content varies significantly between regions. A GEO strategy that performs well in the UK may not translate directly to Germany, Japan, or Brazil without careful localisation of both the content and the structural signals that AI systems read.

How Adapt Can Help

At Adapt Worldwide, our AI SEO service is built around the same principles that the Evertune research validates: authority, content quality, structured signals, and relentless measurement. We work with global brands to build and maintain citation presence across Google AI Mode, Gemini, and other AI surfaces, without losing the cultural and linguistic nuance that international search demands.

Our approach begins with research: identifying the prompts that matter in your category, auditing your current citation footprint, and mapping the gap between where you are and where you need to be. We then build or optimise content to meet AI quality thresholds, structured for extractability and aligned to user intent at every stage of the journey.

For brands operating across multiple markets, our in-market specialists ensure that GEO strategy is adapted for local context. What works in one language or culture may require significant reworking for another, and AI models are increasingly sensitive to the authenticity and accuracy of localised content.

We also provide the measurement infrastructure to track your AI visibility over time: AIO citations, brand mentions in AI responses, sentiment analysis, and traffic attribution from AI surfaces. When Google makes a change as significant as a 59% contraction in cited sources, you need to know about it fast, and you need a plan ready.

Explore our AI SEO, Content SEO, and Technical SEO services to understand how we build citation presence across AI and traditional search simultaneously.

Key Takeaways

• Google AI Mode cut the number of unique URLs cited by 59% between late April and end of May 2026. Gemini fell 33% over the same period.
• This is not about citing less often. It is about citing fewer distinct sources. Each cited URL now carries more weight.
• No domain type or sector benefited disproportionately. The contraction was broad-based.
• GEO performance depends on authority signals, content quality, technical structure, and prompt alignment. These are long-term investments, not quick fixes.
• Brands without AI citation tracking have no early warning system. Measurement is now a non-negotiable.
• For global brands, culturally attuned localisation of GEO content is a differentiator in every market.

Frequently Asked Questions

Source

↗ Evertune Research: Google Slashes Number of Unique URLs Cited in AI Mode, Gemini (June 2026)