Profound: personal context changes ChatGPT and Muse Spark shopping recommendations
What this answers
- Does personal context change what ChatGPT recommends?
- Yes, per Profound's test of 2,240 answers: lists for a user with relevant saved context overlapped about half as much with lists for a user with no saved context as repeated lists for the same user overlapped with each other.
- Does Meta's Muse Spark recommend fewer products when it knows the user?
- In Profound's test, Muse Spark named an average of 4.5 distinct options per answer with relevant context versus 6.5 without it; ChatGPT's averages were nearly unchanged.
Profound published a study on Oct. 29 testing how saved personal context changes AI shopping answers. It used seven synthetic users (built from real conversation and browsing patterns, with memory files modeled on those stored by Meta’s Muse) plus an empty-memory control, and ran 14 consumer-electronics prompts ten times each on ChatGPT Instant and Muse Spark 1.3 with web search on, for 2,240 answers, on Oct. 19. Repeated answers for the same user overlapped by about 42% on ChatGPT and 18% on Muse Spark; lists for a user with relevant context and a user with none overlapped roughly half as much, per Profound.
When relevant context was available, 19% of ChatGPT’s search queries and 54% of Muse Spark’s referenced it. Across 13 prompts, Muse Spark named an average of 4.5 distinct options per answer with relevant context versus 6.5 without, and cited 0.7 web domains per answer versus 1.4; ChatGPT’s averages were nearly unchanged (2.5 versus 2.7 domains). The profiles and their memory files are synthetic, and Profound reports no results from real users’ stored memories.
Why it matters: If assistants tailor shopping lists to what they remember about each person, there is no single ranking to appear in, and a model that shortens its list with context leaves fewer slots to win.
The record: ProfoundChatGPTMeta
Posted to the wire October 29, 2026. Edited by Joe Balewski.