As AI takes on drafting, summarising, analysing and coding, the human role is shifting. Making things matters less. Checking them matters more.
Checking is a real skill. It takes effort, and the evidence shows people are not naturally good at it.
People know checking matters
Microsoft's 2026 Work Trend Index found 86% of AI users treat output as a starting point. Quality control was the top skill people said they needed, followed by critical thinking.
Intent is not the problem. Time and habits are.
What happens without it
In the Harvard and BCG study of consultants using GPT-4, people harmed by AI shared a habit. They accepted its output with too little scrutiny.
Microsoft Research and Carnegie Mellon found the same in 2025. The more people trusted the AI, the less critical thinking they applied.
The 2026 workslop study of 962 US desk workers shows the cost. Unchecked, low-quality AI work gets passed to colleagues, who then have to fix it.
Why checking AI is hard
AI output is polished, so errors do not stand out. The model sounds confident even when wrong.
Errors are often plausible: a slightly wrong figure, a source that does not exist, a conclusion that almost follows. Spotting them takes the effort AI tempts us to skip.
Stanford's AI Index 2026 found developers are sharing less about how models work. Less transparency makes checking harder.
Build this capability in your organisation
Growth Performance designs bespoke programmes that turn insight into measurable behaviour change.
Make checking the default
Good intentions are not enough. Busy, trusting people will accept fluent output unless the system expects checking.
Name the failure first. State what would make the output wrong, then check for that.
Trace sources. Follow every figure and citation back to its origin.
Think before you look. Form a view before seeing the AI's answer. A 2026 study in Frontiers in Psychology found checking your own thinking weakened the slide into dependence.
Pause on high stakes. Add a deliberate stop before important outputs go out.
What this means for leaders
Checking is part of Adaptation: people directing, combining and checking AI in real work. Teach it, give it time and reward it.
Thank people who catch AI errors. Count checking hours in your plans, because AI that produces more output also creates more to check.
Try this
For your team's most important AI-assisted output this week, ask a named person to write two sentences. What did they check? What would they expect to see if it were wrong?
Then use the Checking Capacity Calculator to see whether your team has the hours to check properly.