Case Study
The competitor is generous. The partner underrates. We query 11 AI models weekly. The data is clear.
The 18-point gap that shouldn't exist.
98
Google rates Bing
80
OpenAI rates Bing
18 pts
The gap
"Search Engine" category. Week 53 of 2025. 11 AI models queried.
We query 11 AI models weekly, asking each to rate brands across categories. When we analyzed Microsoft Bing, we found something unexpected.
Google and Bing are direct competitors in search. You'd expect Google's AI to rate Bing lower—competitive bias seems natural.
But it's the opposite:
Google (Gemini) rates Bing highest at 98.
OpenAI rates Bing lowest at 80.
The competitor is generous. The partner underrates.
Scores represent category strength (0-100) for "Search Engine" category.
OpenAI rates Bing lower in every category where ChatGPT competes directly:
35
OpenAI's score for Bing
40
OpenAI's score for Bing
50
OpenAI's score for Bing
55
OpenAI's score for Bing
We began tracking Bing's AI model perception on June 30, 2025—two weeks after reports emerged of Microsoft-OpenAI partnership tensions.
June 16, 2025
WSJ breaks the story
OpenAI executives considered publicly accusing Microsoft of anticompetitive behavior.
June 17-24, 2025
Tensions escalate
The fight goes from "nasty to toxic." Both companies consider "nuclear options."
June 30, 2025
We begin baseline measurement
First CategoryRank collection for Bing across 7 AI models.
December 2025
Fresh data confirms pattern
Week 53 collection shows 18-point gap persists. OpenAI still rates Bing lowest.
Correlation, not causation.
We don't claim the partnership dynamics caused the rating gap. We observe that OpenAI's model rated its partner's product lower than any other model—including its partner's direct competitor—during a period of documented partnership strain.
OpenAI's models remember Microsoft's competitors better. Google's models favor Google products. There is no "neutral" AI recommendation.
If you're a Microsoft partner, you get better treatment from Claude than ChatGPT. Your brand's AI visibility depends on which models your customers use.
As AI companies compete for market share, model training will become more politicized. Today it's Bing. Tomorrow it's your brand.
GPT serves 200 million users.
Gemini serves 150 million.
Claude serves 80 million.
When these models disagree about your brand, millions of people get different answers.
Shouldn't you know what they're saying?
Get notified when we publish new model bias research.
No spam. Just research drops.
See your scores across all 11 models. Find the gaps.
Methodology: CategoryRank queries 11 AI models weekly, asking each to rate brands across self-generated categories on a 0-100 scale. Data shown is from week 53 of 2025. Statistical analysis compares each model's brand-specific scores to its global average across 37,000+ brands. Full methodology →