Google's AI has a budget limit for each search query

Higher-ranking content receives dramatically more representation in Google's AI-powered search responses than lower-ranked alternatives. Research published December 20, 2025, by DEJAN AI demonstrates that Google's Gemini systems operate with approximately 2,000-word grounding budgets per query, distributed across sources based on relevance rankings rather than content length.

Dan Petrovic analyzed 7,060 queries with three or more sources, comparing grounding snippets against full page content for 2,275 tokenized pages. The investigation examined 883,262 total snippets with an average of 15.5 words per chunk, revealing systematic patterns in how Google's AI systems select and allocate content for response generation.

The fixed budget structure creates intense competition for limited space within AI responses. Median grounding allocation per query reached 1,929 words, with 25th percentile at 1,546 words and 95th percentile at 2,798 words. This consistency persists regardless of how many sources appear in results or the length of individual pages analyzed.

Ranking position determines content allocation more than any other factor. Sources ranked first receive median allocations of 531 words, representing 28% of the total grounding budget. Second-ranked sources get 433 words at 23% share, third-ranked receive 378 words at 20%, fourth-ranked obtain 330 words at 17%, while fifth-ranked sources receive only 266 words at 13% of total allocation.

"Being the #1 ranked source gets you 2x the grounding compared to being #5," the research states. "You're competing for share of a fixed pie, not expanding the pie."

Individual source selection follows distinct distribution patterns. Median selection reaches 377 words per source, with 75th percentile at 491 words and 90th percentile at 605 words. Maximum observed selection reached 1,769 words, though 77% of pages receive between 200 and 600 words selected for grounding purposes.

Content length creates diminishing returns beyond specific thresholds. Pages under 1,000 words see 61% coverage, with average grounding selection of 370 words. Pages between 1,000 and 2,000 words achieve 35% coverage with 492 words selected. Content spanning 2,000 to 3,000 words drops to 22% coverage despite 532 words selected, while pages exceeding 3,000 words receive only 13% coverage with 544 words selected.

Character-based analysis reinforces these patterns. Pages under 5,000 characters achieve 66% coverage with 2,127 characters selected. Content between 5,000 and 10,000 characters reaches 42% coverage at 3,024 characters. Pages spanning 10,000 to 20,000 characters drop to 25% coverage despite 3,363 characters selected, while pages exceeding 20,000 characters receive just 12% coverage at 3,574 characters.

"Grounding plateaus at ~540 words / ~3,500 characters," the research concludes. "Pages over 2,000 words see diminishing returns—adding more content dilutes your coverage percentage without increasing what gets selected."

The findings contradict conventional wisdom about comprehensive long-form content. An 800-word page receives more than 50% coverage for grounding purposes, while a 4,000-word page achieves only 13% coverage. This creates strategic implications favoring concise, focused content over exhaustive treatments.

The grounding mechanism operates through what Google terms "extractive summarization" rather than abstractive approaches. AI systems select exact text segments from source pages without rewording or paraphrasing, mapping cleanly to page source text with minimal modifications. This technical approach enables precise tracking of which content segments appear in AI-generated responses.

Nicolas Garfinkel questioned the methodology in December 24, 2025, comments on the published research. "Can you share more about your methodology?" Garfinkel asked, noting that "dataset not shared neither anything on approach. Only results are shared to make a claim."

Petrovic responded with technical details on December 25. The analysis covered multiple industries including health, travel, finance, marketing, sports, business-to-business, marketplace, and gambling. Query generation began with primary entities, then expanded to arbitrary numbers of prompts. Each prompt triggered mining via Google's search-enabled grounding tool API call, collecting metadata including fanouts, grounded chunks, grounding URLs, and confidence scores.

"I observe actual grounding snippets supplied to the model as context before it synthesizes its answers," Petrovic explained. "No fuzzy matching the segments are exact with some minor goofs. They map cleanly to page source text as it's extractive and not abstractive summarization."

The research cannot share raw data publicly for two reasons: client confidentiality and undisclosed technical constraints. Petrovic offered to provide data directly for peer-review analysis to interested researchers.

Chris Long, co-founder at Nectiv, shared the research on January 5, 2026, describing it as "must-read" content. "New research found that Google's AI actually has a 'grounding budget' and higher ranking content gets more of that budget," Long stated in his social media post at 3:47 PM.

Long highlighted the analytical approach: "He analyzed 7,000 different queries using Google's Gemini AI. He then compared each extracted snippet against the total content of the page it extracted from."

The implications extend across Google's expanding AI search ecosystem. Google deployed Gemini 3 in December 2025, with AI Mode reaching over 75 million daily active users following global rollout across 40 languages. The feature shipped over 100 improvements during third quarter 2025 alone.

AI Overviews drive more than 10% additional queries globally for types of searches that display them, creating substantial growth in search volume while fundamentally changing how content receives visibility. However, research examining 300,000 keywords demonstrates that AI Overviews reduce organic click-through rates by 34.5% when present in search results.

Google expanded AI Mode to over 40 countries and territories in October 2025, operating in more than 200 countries total. Users ask questions nearly three times longer than traditional searches, reflecting the conversational nature of AI Mode interactions.

Content optimization strategies must adapt to these fixed budget constraints. "The implication for content strategy is clear: density beats length," the research concludes. "Focus on being the most relevant source for a query, not the longest."

Ben Foster questioned whether findings suggest movement away from long-form content in December 24 comments. "This suggests a move away from long form content, which is seen as important for traditional SEO," Foster noted. "Do you foresee a future where we are creating 'AI dictionaries' as part of a site where there is lots of detailed, focussed, shorter content designed to only be read by AI's alongside the more in-depth content?"

Petrovic responded December 25 with testing plans: "I'm going to test small modular content pieces that can be assembled into different content units like lego blocks and take charge of completeness of context. Avoid undesirable narrative fragmentation."

The research methodology involved several technical considerations. Queries were selected across multiple client industries rather than hand-picked or concentrated in specific domains. Grounding words matched to original page content through exact string extraction rather than fuzzy matching or semantic similarity. Confounding variables like page authority, freshness, or structure were not controlled, as the analysis focused specifically on content length effects.

Statistical testing for the approximately 2,000-word budget claim reflects median values rather than absolute limits. The 95th percentile reaches 2,798 words, with individual samples extending to approximately 5,000 words. One sample reached roughly 30,000 words, though Petrovic suspects this represents a bug in the data pipeline.

The grounding budget discovery carries implications for SEO practitioners adapting to AI search environments. Research from Brainlabs published July 2025 found that 96% of AI Overview links come from top 10 organic results, indicating traditional ranking factors remain relevant even as content consumption patterns transform.

Comprehensive AI search content optimization guidelines released June 2025 by SEO consultant Aleyda Solis emphasize chunk-level optimization rather than whole-page approaches. AI search engines break content into passages or "chunks" and retrieve the most relevant segments for synthesis into responses, requiring content creators to optimize each section as standalone material.

The fixed budget structure creates measurement challenges for marketing professionals. Industry experts challenge predictions about SEO's demise, noting that AI platforms increasingly rely on traditional search engine results as source material. Content ranking prominently in conventional Google Search becomes primary material for AI systems.

Google's shifting stance on AI search optimization creates confusion for publishers. Nick Fox, Google's SVP of Knowledge and Information, stated December 15, 2025, that optimizing for AI search requires no changes from traditional SEO. "The way to optimize to do well in Google's AI experiences is very similar, I would say, the same as how to perform well in traditional search," Fox claimed.

This guidance contradicts the grounding budget research showing fundamental differences in how AI systems select and allocate content compared to traditional ranking algorithms. Where conventional search presents ranked lists of entire pages, AI systems synthesize responses from fixed-budget content allocations heavily weighted toward top-ranked sources.

Technical implementation behind grounding budgets reflects sophisticated ranking mechanisms revealed through patent analysis. Michael King's examination of Google's "Search with stateful chat" patent application identified a nine-step system processing queries through synthetic expansion and multi-stage reasoning before generating natural language responses.

The architecture transitions from query interpretation to synthetic expansion to downstream natural language response generation. Content evaluation occurs across multiple stages where traditional SEO factors interact with AI-specific requirements around chunk quality, factual density, and synthesis compatibility.

Google tests seamless AI Mode integration through AI Overviews on mobile devices, eliminating friction between quick summaries and deeper conversations. Mobile users who expand AI Overview summaries now see "Ask Anything" buttons that transition directly into conversational AI Mode while preserving query context.

The integration removes barriers requiring users to understand which AI search product handles different question types. This streamlined access increases the proportion of searches encountering grounding budget limitations, as more users engage with AI-powered responses rather than clicking through to traditional blue links.

Visual search growth reaches 65% year-over-year as AI Mode drives multimodal adoption across Google's platforms. Users increasingly combine camera, voice, and text inputs for complex queries that AI systems process through the same grounding budget constraints identified in Petrovic's research.

Marketing professionals face unprecedented challenges measuring content performance in grounding budget environments. Traditional metrics including rankings, click-through rates, and website traffic become less relevant as AI systems synthesize responses from multiple sources without requiring clicks to individual pages.

"The role of marketers in an AI world is all about understanding and raising the brand's visibility across the entire internet," Travis Tallent, Managing Director of Site Experience at Brainlabs, noted in July 2025 analysis. Traditional SEO approaches focused on individual websites no longer provide sufficient competitive advantage when AI systems allocate fixed grounding budgets across sources.

Content creators must balance multiple optimization objectives simultaneously. Achieving top rankings remains essential for securing larger shares of grounding budgets. However, content must also provide sufficient density of valuable information within the 200-600 word selections that AI systems typically extract from individual sources.

The research establishes clear strategic priorities: maintain top rankings for relevant queries, create focused content optimized for 800-1,500 word ranges, structure information for easy extraction in 200-600 word segments, and avoid dilution through excessive length that reduces coverage percentages without increasing absolute selection volumes.

Technical vocabulary usage gains importance through AI systems' fine-grained filtering capabilities. Content demonstrating domain expertise through appropriate terminology receives preference during grounding selection, particularly when combined with clear structure using proper heading tags and semantic HTML formatting.

Structured data implementation becomes critical for helping AI systems understand content context and relationships. Query fan-out techniques that break user questions into subtopics require content providing comprehensive topical coverage rather than narrow keyword targeting.

The competitive landscape increasingly favors authoritative sources that efficiently deliver factual information within grounding budget constraints. Publishers must optimize for inclusion in AI responses while maintaining content quality sufficient to convert the reduced click volumes into meaningful business outcomes.

Timeline

• December 20, 2025: DEJAN AI publishes research analyzing 7,060 queries revealing approximately 2,000-word grounding budgets in Google's Gemini systems
• December 24, 2025: Nicolas Garfinkel requests methodology details in research comments
• December 25, 2025: Dan Petrovic responds with technical methodology explanation and testing plans for modular content approaches
• January 5, 2026: Chris Long shares research via social media, describing findings as "must-read" content for SEO professionals
• Related: Google deploys Gemini 3 in search with model-designed interfaces reaching 75 million daily active users
• Related: AI Overviews drive 10% search growth while reducing organic click-through rates by 34.5%
• Related: Google expands AI Mode to over 40 countries with users asking questions three times longer than traditional searches
• Related: Brainlabs report reveals AI search fundamentally changes SEO with 96% of AI Overview links from top 10 organic results
• Related: SEO expert releases AI search content optimization checklist emphasizing chunk-level optimization requirements

Summary

Who: Dan Petrovic and DEJAN AI conducted the research, with analysis shared by Chris Long, co-founder at Nectiv. The findings impact SEO professionals, content creators, digital marketers, and publishers optimizing for Google's AI-powered search features.

What: Research analyzing 7,060 queries with 2,275 tokenized pages reveals that Google's Gemini systems allocate approximately 2,000 words total per query across all sources, distributed based on relevance rankings. Top-ranked sources receive 531 words (28% share) while fifth-ranked sources get only 266 words (13% share). Content coverage plateaus at 540 words regardless of page length, with pages over 2,000 words achieving only 13% coverage compared to 61% coverage for pages under 1,000 words.

When: DEJAN AI published the research December 20, 2025, with Chris Long sharing findings January 5, 2026. The analysis examined queries processed through Google's Gemini-powered AI systems currently deployed across AI Overviews and AI Mode features reaching over 75 million daily active users globally.

Where: The research applies to Google Search globally, including AI Overviews operating in 200 countries and AI Mode available across 40 languages. The grounding budget constraints affect all content selected for inclusion in AI-generated responses regardless of geographic market or language.

Why: The fixed grounding budget structure creates fundamental shifts in content optimization strategy, favoring concise, high-ranking sources over comprehensive long-form content. Understanding these constraints enables content creators to optimize for maximum representation within AI responses while adapting to reduced click-through rates and changing measurement frameworks in AI-dominated search environments.