What predicts whether an AI platform recommends your brand again tomorrow

I pulled 3,115 day-pair observations across 10 brands and 6 AI platforms and ran an XGBoost model and ordinal regression to figure out what actually shifts a brand's position in an AI response. Found that position is stickier than expected: 94% of absent brands stayed absent the next day, 71% of primary placements held, and 63% of secondary positions persisted. Only the "mentioned" tier was volatile, with just 24% retention.

That stickiness changes how to think about the work. Optimizing for a single-day spike is close to useless when most positions stay where they were yesterday; the strongest predictors of upward movement in our model were the things that take weeks to build. Very-positive sentiment carried a +1.60 coefficient, 7-day rolling SOV +1.49, and platform structure itself dominated everything else (Gemini +1.32, GAIO +1.16, Grok −3.60).

Excerpt length, which we saw yesterday as the #1 predictor of whether a brand appears at all, had zero correlation with position tier (+0.007) today. This suggests that excerpt length (which itself might be a proxy for internal representations) only predicts entry into the response; it does not predict ranking inside it.

Platform makes a big difference still. Perplexity averages 3.21 competitors per response and gives brands a primary slot only 6.7% of the time, ChatGPT returns single-brand answers 18.5% of the time, and Grok churns brands daily with 0% primary placement for the monitored set. So if you're an up and coming brand, Grok is the best shot at any visibility, but also the most competitive (Lines up with our own experiences - Sill showed up in Grok before any other platform).

Find out more here:
https://trysill.com/blog/ai-brand-position-study