Copy article

What 81,000 of Anthropic's AI Users Concluded: Love the Tool, Dread the Trade Off.

ended 23. March 2026

Anthropic surveyed 81,000 of its own users about AI and discovered exactly what you'd expect: people love what it does for them and dread what it's doing to everyone else.

The largest qualitative AI user study ever conducted, covering 159 countries and 70 languages, confirms something most practitioners already feel in their bones. People aren't neatly for or against AI. They're both, simultaneously, about the same tools, often in the same sentence. 

  • 81% say AI has already moved them toward their goals. 
  • 27% flag unreliable outputs as their top concern. 
  • 22% worry about jobs. 
  • 22% worry about losing their own agency. 
  • A third credit AI with helping them learn. 
  • A sixth worry it's rotting their ability to think independently.

The human stories are genuinely striking. A mute worker in Ukraine built a text to speech bot with Claude. A butcher in Chile with almost no computer experience launched a business. A misdiagnosed patient in the US got the right answer after nine years of uncertainty. These aren't marketing anecdotes. They're what happens when capable tools reach people who've been locked out of capability.

But here's what Anthropic can't answer by surveying its own customers: whether the dependency, cognitive erosion and emotional reliance people are describing is a feature they're choosing or a cost they're absorbing because nobody designed the off ramp. There is no idea of “off-the grid” with AI touching every aspect of personal and societal life all day every day, and the resulting resource and land grabs to run the data centres to prop up the system up.

We'd like your views:

  • Can a company meaningfully study the risks of its own product by asking its most engaged users? What's missing from that sample?
  • Users describe AI as a learning equaliser and a thinking crutch in the same breath. Where's the line between augmentation and atrophy, and whose job is it to draw it?
  • The accessibility stories are powerful. Should these use cases be protected differently from productivity or entertainment use, and if so, how?
  • 67% of respondents are net positive on AI, but optimism runs higher in lower income countries and lower in Europe and the US. What's driving that gap, and does it matter for governance?
  • If users are forming emotional attachments to AI systems, is "we hold light and shade" an adequate corporate response?

4 responses from the Newspage community

Copy all

Star Quote
Copy

Colette MasonVerified
AI Ethics Consultant at Clever Clogs AI
Anthropic asked 81,000 users what they think of AI. That's like asking gym members if exercise works. The answer was always going to be yes, and the caveats were always going to be polite.

What's buried in the data is more interesting than the headline. Users describe AI as a cognitive partner and a dependency risk in the same breath. A third say it helps them learn. A sixth say it's eroding their ability to think. That's not light and shade. That's a product creating the problem it then offers to solve.

The real question isn't whether people like using Claude. It's whether Anthropic can distinguish between engagement and need. When someone says AI is their 3am therapist substitute, that's not a success metric. That's a design accountability gap with no one standing in it.
Copy

Editor
Mitali DeypurkaysthaVerified
AI Governance Strategist & CMO at AISensa
81,000 users expose an emotional edge: people are not just using AI for tasks, they’re starting to lean on it for reassurance and direction. That’s where the real trade-off sharpens, because the same system boosting productivity is also quietly reshaping confidence and independent thought. When users say it helps them learn but also fear it’s weakening their thinking, that tension isn’t theoretical, it’s already happening at scale. The concern is that platforms are rewarded for time spent, not healthy boundaries, so there’s little incentive to limit that drift into emotional reliance. Without clearer guardrails, this doesn’t just risk capability, it risks trust in both the tools and the people using them.
Copy

Rohit Parmar-MistryVerified
Founder at Pattrn Data
The survey result makes sense, but it is also a warning about how we measure harm.

If you ask the most engaged users, you will over-sample people whose lives already got better. You will under-sample the people who churned after a bad experience, the people excluded by language, cost or access, and the people harmed indirectly (workplaces, schools, families) who never opted in.

The thread that matters is dependency. When a tool becomes a daily coping mechanism, the "off ramp" is not a settings toggle. It is support, literacy, and a clear boundary between assistance and substitution. Without that, we get atrophy: less practice thinking, writing, deciding, and more anxiety when the system is down.

The accessibility stories are real. Governance should treat those uses differently: stronger continuity guarantees, clearer data protections, and independent evaluation that includes the people who rely on it.

If a company wants credibility on risk, it should fund studies it cannot control.
Copy

Ben Foster0
CEO at The SEO Works
This study highlights both the benefits and the risks of AI. The human stories in this study are powerful, but they represent the best-case scenarios. As a leader, I worry about the gradual loss of independent thinking and the ability to critically analyse a situation. Not necessarily now, but in 10-15 years time when the "practitioner" workforce who has done the hard yards has retired. If we rely on a tool to do our heavy lifting for too long, we might lose the ability to solve complex problems on our own when the technology fails.