Not sure where to start? Take a guided tour.
AI Site Grader
AI Query Scanner
llms.txt Generator
In short
Four findings, strongest first- 43.3%
average agreement when 8 AI models answer the same question. Models agree on the top brand less than half the time.
Chapter 02 - 4.0%
of queries get the same top recommendation from all 8 models.
Chapter 07 - 14.6%
of queries are high divergence, with less than 25% agreement between models.
Chapter 02 - 798K
valid model comparisons across 44,088 reports, including Google AI Overviews.
Method
Each AI model is its own channel.
When someone asks ChatGPT, Claude, Gemini or Perplexity for a product recommendation, they expect consistent answers. But our analysis of 797,644 comparisons shows AI models agree less than half the time.
This isn’t a bug. It is a fundamental characteristic of how different AI systems are trained, what data they’ve seen and how they read intent. For brands, it means your visibility on ChatGPT tells you nothing about your visibility on Claude.
We analysed 797,644 valid comparisons across 8 major AI models, including Google AI Overviews. Whether your team calls that program AEO or GEO, the measurement problem is the same. See the AEO vs GEO comparison for the terminology boundary and shared metrics.
Models agree less than half the time.
The agreement problem
When you ask different AI models the same question, they usually disagree. More than half of all queries (60%) have less than 50% agreement between models on the top brand recommendation.
Only 4.0% of queries achieve perfect consensus - all 8 models recommending the same brand. This is rare, and typically happens only for brands with overwhelming category dominance.
Why this matters
Some models move together.
OpenAIClaudeGeminiAIOGrokDeepseekMetaPerplexityModel clusters emerge
Some models tend to agree with each other more often, forming implicit clusters. Claude and Deepseek show the highest correlation at 35%.
Claude+Deepseek35%Meta+Perplexity10%Why this matters
Not every model names a brand.
Not all models show up
Some AI models are more likely to provide brand recommendations than others.Meta appears in 95.0% of queries, while Google AIO only shows up in 56.5%.
This matters because if a model rarely provides recommendations in your category, your visibility strategy for that model may need a different approach.
Why this matters
MetaAIOMeta95.0%OpenAI85.4%Grok83.0%Gemini82.2%Deepseek80.9%Claude79.9%Perplexity79.4%AIO56.5%Comparisons agree most, general queries least.
Comparison queries = highest agreement
When users ask to compare specific brands ("Nike vs Adidas"), models agree 50.4% of the time. But "best of" and general queries cause the most disagreement - exactly the queries where brands have the most opportunity.
This makes sense: comparison queries have clearer context, while open-ended recommendations leave more room for interpretation.
Why this matters
Eight models, eight different answers.
These real examples show complete disagreement - 8 different AI models recommending 8 different brands for the same query. This is the reality brands need to understand.
OpenAIClaudeGeminiAIOGrokDeepseekMetaPerplexityOpenAIClaudeGeminiAIOGrokDeepseekMetaPerplexityOpenAIClaudeGeminiAIOGrokDeepseekMetaPerplexityWhen every model agrees.
Strong dominance + clear context = agreement
When ALL 8 models agree on a recommendation, it is typically for queries where:
- •A single brand has overwhelming category dominance
- •The query is highly specific or niche
- •There's clear category definition with limited alternatives
Why this matters
What this means for brands.
The old playbook is broken.
Optimizing for "AI" as a single channel doesn't work. Each model sees a different web. Here's what that means for your strategy.
How we measured it.
OpenAIClaudeGeminiAIOGrokDeepseekMetaPerplexityHow we measured agreement
We analyzed 44,088 brand visibility reports from Trakkr, each containing responses from up to 8 major AI models for the same set of queries.
For each query, we identified the #1 recommended brand from each model, then calculated what percentage of models agreed on the same top brand. A 43.3% average agreement means less than half of models typically agree.
Comparisons were filtered to include only queries where at least 5 models provided a valid brand recommendation, ensuring statistical significance.
