Our newsletterBundling backfired for GPT-5 in the phone category
Creatives
- Reader Note:
- This is one model in one product category inside the simulation, not a general rule for all AI agents.
- Evidence:
- Experimental
- Vendor interest:
- None
- Sample:
- Repeated simulated choice experiments using 4 AI models (GPT-4.1-mini, GPT-5, Gemini 2.5 Pro, Gemini 2.5 Flash Lite) across 4 product categories, with a target of 1,000 choice iterations per product per model; each choice screen showed 8 product cards.
Reader Note:
Evidence:
Vendor interest:
Sample:Good for
Backing budget and pricing decisions in similar situations.
Not for
- Situations far from the one tested, without checking again.
- The source never says when this data was collected. It could be considerably older than the publication date suggests.
- Nobody counted this directly — it's a calculation. The publisher worked it out itself, or took the ad platform's word for what its ads achieved.
Related cards
IdeaCreativesTrend
Only 5% of Meta ads reach winner status
Trend
Go deeper on this topic
Our Substack turns the data behind these cards into full breakdowns: what changed, why, and what to do about it.
Open full graph