Models
Why the answers differ#
Different training data and cutoffs. A model trained later knows about products an older one has never heard of.
Different retrieval. Some models search the live web for most queries; others answer from what they already know. Live retrieval favours recent and well-linked sources.
Different source preferences. One model leans on community discussion, another on structured review sites. Your presence in each source type decides where you do well.
Different answer shapes. A model that returns three options is a harsher filter than one that returns ten.
What to do about it#
Track the models your buyers actually use, and read them separately before you average them. A strong overall score can hide being invisible in the one assistant your enterprise buyers have deployed internally.
When a gap between models is large, the citation lists usually explain it. Being present in community sources and absent from review sites will look exactly like this.
Credits and coverage#
Each model checked on each prompt costs credits. Higher plans unlock more models and more daily credits — see Plans and Credits.
If you have to choose, start with the assistant your buyers name in sales calls, and add breadth once the first one is stable.