Fast to Adopt, even though Trust is Low
Half of U.S. adults now use AI chatbots. A quarter use them daily, some almost constantly even though most believe AI is moving too fast, will weaken their data security, and is more likely to harm society than help it.
Pew’s February 2026 survey captures a step-change: AI has shifted from occasional tool to routine behaviour.
Around four-in-ten people use chatbots for information search, 38% of employed adults use them at work, and usage is increasingly habitual, not experimental.
At the same time, adoption of AI-enabled devices and AI summaries in search results shows the same pattern: AI is becoming ambient.
This is what infrastructural technology looks like in real time. Like search or smartphones, AI is embedding itself beneath conscious choice powering how people access information, make judgements, and complete tasks.
Unlike previous waves, this is arriving alongside unusually high scepticism: roughly two-thirds say it’s advancing too quickly, and around seven-in-ten expect it to make their personal information less secure.
WHY IT MATTERS
The critical dynamic is adoption without belief.
People are not waiting to trust AI before integrating it into work and daily life. Behaviour is scaling ahead of meaning-making. Why?
In organisations, this produces fragile transformations: employees using AI to accelerate outputs while remaining uncertain about accuracy, appropriateness, or risk, creating hidden variability in how decisions get made.
Is this because it is 'good enough', a short cut that makes it easy, or because it has been legitimised by their organisation, it isn't their 'fault' if it is wrong?
WHAT TO WATCH FOR
→ The quiet normalisation of AI-mediated judgement.
→ AI-generated search summaries and tools become defaults informing choices
→ The shift from using AI to 'thinking through AI' first.
→ Declining friction before accepting an AI-generated answer.
LIMITATIONS
This is self-reported survey data from U.S. adults, which captures stated behaviour and perception not their actual dependency, depth of use, or decision impact. “Use” could mean anything from occasional queries to workflow reliance. Broad categories like “chatbots” or “AI features” flatten important differences in capability, context, and risk.
SOURCE
BESCI AI OPINION
Can two things be true at once. Yes.
We love AI, it makes our life easy
We don't trust AI
The level of trust is healthy, but not enough for us to stop using it.
Humans often do behaviours they know hurt them, yet enjoy (smoking, drinking, eating too many cakes, driving polluting cars). That isn't new.
What will be fascinating to study is the level of trust and the tipping point, based on how easy AI makes it for us.
Every time we click into a website, social media app or a blog, we leave a digital trail that gives someone the opportunity to sell and market to us. We trade that data for the information or enjoyment that we receive.
Is AI any different? Are we becoming immune to the trade off - the accuracy is 'good enough'.
We might rationalise it that humans make mistakes too.
There is probably a factor of expertise in here too.
If you are an expert baker and the recipe that AI gives you has a weakness, you will spot it and distrust in your specialist field.
If you are an amateur, then you may follow it, get an average and 'good enough' result and be happy.
One to watch for sure. It will be harder and harder to spot the weaknesses AI brings. Our expectations will become lower.
