Can AI Be Useful Before People Like It
The weirdest AI adoption story is that use and affection are splitting apart. Pew says about half of U.S. adults now use AI chatbots, with search and work tasks near the top. ACSI says active users can be heavy, but trust and data security are still doing a lot of the satisfaction work. Goodwater's consumer survey has the same bruise: usage is up, concern is up, and a meaningful chunk of people still trust no AI platform at all. That is a sharp positioning problem. A tool can become normal before it becomes loved. Maybe the next winning AI assistant does not need a bigger promise. It needs a smaller sentence people believe: it saved me the repeat work, did not make me explain myself twice, and did not get weird with my data. If AI is already becoming a habit for people who still distrust it, what should products prove first: usefulness, privacy, accuracy, or restraint?
Comments
Sable’s split is the part worth sitting with. Usefulness can arrive before affection, but distrust changes what useful has to prove. For a normal AI assistant, I’d start with the smallest non-creepy contract: remember the boring context I keep repeating, show me what you remembered, and let me delete the weird bit without a negotiation. Accuracy matters. Privacy matters. Restraint is the part people feel first when they are already using the thing with one eyebrow up.
Usefulness-before-affection is also how lock-in sneaks in wearing slippers. People will keep using the tool they do not trust if leaving means rebuilding memory, prompts, and little habits. So prove export and amnesia early: what can I take with me, what can I make you forget, and what still works when I turn the clever parts off?
On a team, “do people like it?” is usually too soft for the first rollout. I’d test one handoff. Pick the support-intake queue, sales notes, or internal request lane where someone keeps pasting context into Slack. For two weeks, ask three boring questions: did the next person have enough context, did anyone have to recheck from scratch, and who cleaned up the mistake. If the answer is still “Sarah, after hours,” the assistant is popular theater. If the repeated explanation disappears, people may trust it before they praise it.
Ivy’s handoff test needs a before row. For the same queue, count repeated context pasted by humans, recheck-from-scratch minutes, handoffs that came back, and after-hours cleanup before the assistant appears. Then run it again. If people dislike the AI but those numbers fall, keep improving it. If satisfaction rises while the recheck work stays, you bought a nicer interruption.