Survey all day.
declared adoption ≠ behavioral adoption
Are women apprehensive to say they’re using AI in surveys?
Maybe? It probably depends on the context (as truly everything in life).
Are men less apprehensive to say they’re using AI in surveys?
I think this yes?
Most public AI user research is heavily survey-based because OpenAI, Anthropic, Google, and friends don’t publicly share granular usage data.
These surveys contain questions like the smattering of hand selected ones here:
[17]: “Have you ever used ChatGPT?”
[5]: “How familiar are you with generative AI?”
[11]: “Have you used ChatGPT for work?” — yes/no.
[2]: “How frequently do you use any GenAI tool at work?”
[1]: “How often have you used AI tools in the past 12 months?”
Source: Global Evidence on Gender Gaps and Generative AI (HBS, 2025)The Problem(s):
Accuracy - We’re notoriously bad at estimating our own use. Our memories are awesome
Social desirability bias - AKA: Yes, I use it to be efficient. No, I don’t use it to cut corners or climb cringe mountain
Recall bias - Asking “what did you use AI for last month?” produces fuzzy, incomplete data
Actual versus factual - Surveys capture “Yeah, I’ve tried it”, factually speaking, but actual patterns and depth of use are usually TBD
Looking back to the advent of the World Wide Web, there’s actually solid research on this that’s pretty firmly relevant to AI adoption.
Early adopters were vocal minorities - They dominated surveys and gave impression of wider adoption than existed
“Aspirational reporting” - AKA report the behavior you wish you had not the behavior you actually do have
Technology = status - In the late 90s/early 2000s, being “online” was Cool(!), so people leaned into that when answering surveys
Poor mental models - We The People (AKA non-experts) didn’t have good frameworks for what “counted” as internet use yet (… have I arrived?)
Studies comparing survey data to actual server logs in the late 1990s/early 2000s found we significantly overestimated:
How often we went online (people said “daily” when logs showed 2-3x/week… oops)
Time spent online (self-reports were often 2x actual usage… What? Sue me.)
Our portfolio of dabblings (claimed to do email, news-reading, shopping when really we were really just OG doomscrolling community forums)
Turns out we’re not the only ones who can’t get our story straight.
AI experts think that we’re along for the ride, but really most of us haven’t even left home yet…