What 81,000 People Want From AI Is a Better Growth Strategy Than Another Founder Tweet Thread
Anthropic’s 81,000-interview study reveals what users actually want and fear, with hard percentages around professional excellence, unreliability, job anxiety, and emotional support that many AI companies still ignore.
The attention-grabbing version: a shocking number of AI companies are still building for the version of users they imagine on X, not the ones who told us directly what they want.
If you want one of the most useful AI datasets of 2026 that is not a benchmark, Anthropic’s March 18, 2026 feature on 81,000 user interviews deserves far more attention than it gets.
Because this is not another leaderboard war.
It is a giant reality check.
Anthropic says nearly 81,000 people across 159 countries and 70 languages participated, making it the largest and most multilingual qualitative study of its kind.
That scale matters on its own.
But the really useful part is what people actually said.
The clearest demand signals
Anthropic reports that the top things people want from AI include:
- Professional excellence: 18.8%
- Personal transformation: 13.7%
In other words, many users are not mainly fantasizing about abstract AGI theater. They want:
- more effective work
- less routine cognitive drag
- more time
- more support
- help changing aspects of their lives
That sounds obvious until you look at how many AI products are still designed.
Too many are optimized to impress the builder, not to reduce friction in a user’s actual day.
The biggest fears are even more instructive
Anthropic says the biggest concerns people expressed included:
- Unreliability: 26.7%
- Jobs and economy: 22.3%
That is one of the most commercially important AI findings of the year.
Because it means a large share of users are not blocked by lack of curiosity.
They are blocked by:
- fear the system is wrong
- fear the verification burden is too high
- fear the economic consequences are personal, not abstract
If your product ignores those anxieties, your growth strategy is probably weaker than your dashboard says.
Why unreliability is still the category’s biggest tax
The 26.7% unreliability figure should be humiliating for the industry.
Not because models are useless.
Because the user sees the cost differently than the lab does.
A model that gets things mostly right but still:
- hallucinates
- cites fake sources
- forces constant verification
- argues when it is wrong
can still feel exhausting.
That is the hidden tax.
And it explains why some very capable systems still do not earn durable trust.
Why the jobs number changes marketing too
The 22.3% jobs-and-economy concern also matters because it tells you the emotional environment your product is entering.
Users are not all showing up to AI with optimism and clean curiosity. Many are arriving with labor anxiety.
That changes what product messaging, onboarding, and value framing should sound like.
If your growth strategy screams:
- automate everything
- replace more labor
- do more with fewer people
you may get attention, but not necessarily affection.
That is a real tradeoff.
The companies that will be liked, not merely noticed, are the ones that show:
- control
- usefulness
- honesty about limits
- better human outcomes
Why this study is a product brief in disguise
Anthropic says it used Claude-powered classifiers to categorize what people want, fear, and feel. That turns the study into something product teams can actually use.
It is basically a giant answer to:
what emotional and practical jobs are users hiring AI to do?
That is more valuable than another generic “AI trends” post because it helps teams decide:
- what to build
- what to say
- what to de-risk
- what to prioritize in UX and trust
The thing founders should hear but may not want to
If users are loudly saying they care about reliability and economic effects, then a growth strategy built mostly on novelty, loudness, and feature theatrics is fragile.
People may click it.
They may not love it.
And the user asked for both: traffic and approval.
Approval is harder.
Approval usually comes from making people feel more capable without making them feel manipulated, disposable, or fooled.
The blunt takeaway
Anthropic’s 81,000-interview study matters because it gives the AI industry something it often lacks: a large-scale human reality check. People want AI to help them work better and live better. They also worry it is unreliable and economically destabilizing. Those are not side notes. They are product requirements disguised as user sentiment.
The companies that ignore them may still get clicks.
The companies that build around them are more likely to get trust.