The Assumption Gap: How People Are Setting Their Own Rules for AI
AnswerLab’s new Signature Report, built from 334 interviews across seven areas of daily life, reveals the gap between what AI products assume people want and what people actually do, and the workarounds people build to close it.
Grounded in Real Human Experiences
What you'll learn
This report maps how people are setting their own terms with AI across every major context of daily life, with more behavioral specificity than check-the-box surveys can offer. The findings further build on three years of AI and user experience research by AnswerLab, a research-powered experience strategy firm, across hundreds of research engagements with companies including Google, Meta, American Express, PayPal, and Airbnb.
Insights Include:
- Why people keep a human approval step before AI can act, and what that means for agentic product design
- The hidden verification tax users are running alongside every AI tool, and why that safety net fades over time
- What tasks people keep from AI because they’re too personal
- How people actually detect AI-written messages, and where their tolerance runs out
- What teens say about AI changing how they think, said with more clarity than most adults in the study
- How AI adoption is filling gaps where healthcare, education, and support systems fall short
- Six questions for your next roadmap review, plus three predictions for product, design, and experience leaders
FAQ
How can companies improve AI adoption?
Start with the workarounds. AnswerLab’s research found people are already writing their own rules for AI: keeping the final approval, checking its work, and protecting tasks that carry meaning. Companies that design for those rules, with clear recourse when something goes wrong and AI that signals when it’s unsure, earn more of the work.
Do people trust AI to make decisions for them?
People trust AI to research, compare, and surface options, but they want to make the final call themselves. According to a study from AnswerLab, none of 50 shoppers would let AI make a new purchase on its own, at any price. Without being asked, many wrote their own rules for what AI may do, like dollar limits, repeats only, and a trial period first.
How do people know if a message was written by AI?
According to research from AnswerLab, people caught AI-written messages less by how polished they were than by how generic they sounded and how little they sounded like the sender.
Do people verify the answers AI gives them?
Yes. Because nothing tells people when an AI answer can be trusted, roughly 80% of participants across five of AnswerLab’s studies described checking AI’s work. However, AnswerLab's study found that as AI builds a track record, people check it less, just as a missed error starts to cost more.
What tasks will people never give to AI?
Tasks tied to meaning, identity, and care. AnswerLab’s research found that most respondents wanted to choose gifts themselves, and people kept tasks like cooking and meal prep as a way of caring for their families.