What 8,008 Brand Score Changes Reveal About AI Reputation Volatility
The Volatility Data
8,008
Total changes
74
Improved (0.9%)
464
Declined (5.8%)
462
Neutral (5.8%)
The 6:1 negative-to-positive ratio is the most significant finding. Most brands that experienced a score change this week moved in the wrong direction.
Score change distribution
Weekly volatility comparison
Week 12
Week 13
2.6× week-over-week variation — likely driven by model update cycles.
Why Scores Change
Model updates (the biggest driver)
When AI models incorporate new training data, they re-evaluate brand entities against the updated corpus. If newer sources describe a competitor more favorably, or if industry categorizations have shifted, brand scores change even when the brand itself hasn't done anything differently. We see this in the weekly volatility data — W12's 724 deltas vs. W13's 276 suggest a model update cycle.
Content changes
When a brand publishes new content, updates structured data, or modifies their website, AI models that retrieve that data in real-time update their representation. We've tracked cases where a single structured data update on a brand's homepage moved their BAI score by 15+ points within a week.
Source authority shifts
Changes to Wikipedia, Crunchbase, G2, or other authoritative sources propagate to AI models on their next training or retrieval cycle. A Wikipedia edit can take 2–6 weeks to affect parametric knowledge, but retrieval-based models pick it up immediately.
Competitive displacement
AI brand perception is relative. When a competitor improves their AI presence, they may push you down in recommendation rankings even though your own data hasn't changed. In category placement queries, AI models have a limited number of 'slots' — improving one brand's position often means another brand drops.
The Default State Is Decay
The 464 negative vs. 74 positive split isn't a one-week anomaly. It reflects a structural reality: AI brand perception degrades naturally over time.
Think of it as information entropy. Your brand's AI representation was built from a specific set of training data at a specific point in time. Every day that passes:
- •Your competitors publish new content that AI indexes
- •AI models incorporate new training data with updated competitive landscapes
- •Your own content ages — what was current 6 months ago may not reflect your latest positioning
- •Third-party sources about you may become outdated or get edited
Without active maintenance, the accuracy of AI's understanding of your brand decays. Not because anyone is attacking you, but because the information ecosystem moves on and your brand representation doesn't update itself. The brands that maintain Incumbent status aren't the ones with the best initial score — they're the ones that treat AI brand reputation as an ongoing discipline.
Correlations Worth Watching
Caveats: these are based on one dataset and we're continuing to study them.
Higher BAI scores correlate with lower volatility
Brands scoring 80+ tend to have smaller week-over-week score changes. Strong, consistent signals across many authoritative sources resist noise from less reliable sources.
Category matters
SaaS/Cloud Software brands experience higher volatility than Retail/E-commerce brands, likely because the SaaS competitive landscape changes faster and AI models are constantly re-evaluating category placement.
Structured data brands are more resilient
When brands have Organization schema, SoftwareApplication schema, and an llms.txt file, their BAI scores are more stable. Structured data gives AI models a clear, authoritative signal.
What to Monitor and How Often
Run a full audit across all models and query categories. This is your comprehensive baseline check.
Track your BAI score for significant changes. A 10+ point swing in a week warrants investigation — something changed in either your sources or the model's training data.
If you update your website, structured data, or authoritative third-party profiles, check your BAI score within 1–2 weeks to see if the change propagated.
Related
Data source: Optimly AI Brand Directory delta tracking, week of March 22–28, 2026. BAI scores are based on queries across ChatGPT, Claude, Gemini, and Perplexity. This represents brands in our directory of 5,829; the full market may behave differently.