Gruow

Thinking

AI is taking the questions we used to ask each other

AI is good at answers, and colleagues were never only answers.

Most of what people know at work never gets written down. It moves as questions. How do I do this? What does that acronym mean? Is this right? Each one is small, and each one carries a bit of know-how from one person to another.

That traffic is moving to chatbots. The questions still get answered, usually faster. What stops happening is everything else a question used to do: pass on context nobody documented, show who knows what, and build a little trust between two people.

The short version: AI is good at answers, and colleagues were never only answers.

The questions have moved

In an October 2026 Clutch survey of 669 US workers who use AI regularly, 45% said they go to AI first with a work question. 16% go to a coworker first, and 9% to their manager.

The questions that move first are the ones people would rather not ask out loud. 76% have asked AI something they’d be embarrassed to ask a colleague, and 44% turn to AI first after making a mistake. Of those who go to AI first, 72% say it’s for speed.

The shift isn’t limited to one country or one level. In ELMO’s survey of 1,068 Australian employees, run by YouGov in the first half of 2026, one in seven go to AI before a person. Among senior leaders and business owners it’s 26%. 28% of those who pick AI say they don’t want to bother a colleague.

The public version of the same habit is easier to measure. In a study published in PNAS Nexus, researchers compared Stack Overflow, the programmers’ question site, with similar sites where ChatGPT was restricted. Within six months of ChatGPT’s release, activity on Stack Overflow fell 25%. The posts that disappeared weren’t the weak ones: quality held steady, so good questions left too. By May 2025, the site’s monthly question count was back to where it was at launch in 2009.

What travelled with the question

A question to a colleague did more than get an answer. Four things went with it, and none of them come back from a chatbot.

Context nobody wrote down. The answer to “how do we do this here?” lives in people: what was tried before, which client hates what, why the process has that odd step. Workers know this. They trust AI more for writing (64%) and how-to questions (60%), but they trust coworkers more for company-specific knowledge (67%), and by a wide margin for judgment.

For judgment, feedback and career advice, workers trust coworkers more than AI

Share of workers who trust each source more, by kind of question (%)

Source: Clutch survey of 669 US workers who use AI regularly, October 2026. Redrawn by Gruow.

Knowing who knows what. Every question asked out loud tells a team where the knowledge sits and where people are stuck. In the Clutch survey, 51% notice fewer quick questions and check-ins from teammates. 38% see teammates solving the same problem separately, and 57% say AI has created silos on their team. Most still say their relationships are fine: 91% call them unchanged or stronger. The cost shows up in the work before it shows up between people.

Standing with each other. People take questions to AI to avoid looking incompetent. The research says that fear is mostly misplaced. In experiments published in Management Science, Alison Wood Brooks, Francesca Gino and Maurice Schweitzer found that asking for advice on a hard task makes people seem more competent, not less, in the eyes of the person they ask. The embarrassing question was often a small win for both people.

An answer the next person can find. A good answer on Stack Overflow helped everyone who came after. An answer in a private chat helps one person, once. The public record that today’s AI tools learned from is getting thinner.

We’ve seen this shape before

Two earlier studies show what happens to a network when help stops coming from people.

When Microsoft moved its US staff to remote work in 2020, researchers tracked the email, chat, calls and meetings of more than 61,000 employees. Time spent collaborating across groups fell by about 25%. People talked more with their inner circle, added new contacts more slowly, and their networks grew more cut off from each other. The authors warned that teams like this may have a harder time finding and sharing new information.

AI coding tools show the same pattern. Researchers from Harvard Business School, GitHub and Microsoft followed open-source developers from 2022 to 2024, comparing those who got GitHub Copilot free with similar developers who didn’t. With Copilot, developers did 12% more coding and 25% less project management. They also worked with far fewer people: an average of 5 collaborators on public projects, against 22 without it.

Neither study is about chatbots replacing colleagues directly. Both point the same way: when a tool covers what another person used to, the circle of people you work with gets smaller.

Who loses most

Newer people. They have the most questions and the most to learn from the answers. In the Clutch survey, 33% have noticed newer employees getting less mentoring, and 76% worry they’re missing what colleagues would have taught them. In the Copilot study, the least experienced developers moved furthest from shared work, cutting project management by up to 27% against 14% for seasoned ones.

The people who used to be asked. Being asked is how an expert becomes known, and how they learn where others get stuck. Fewer questions means fewer chances for that person to teach, to be credited, and to spot a problem early.

So what do you do about it?

Not stop people asking AI. For a how-to question, a chatbot is often the better first stop: faster, always there, and no one feels judged. The aim is to keep the parts of asking that a chatbot can’t do.

  • Ask AI for the how, people for the why. The trust data already splits this way. Steps, wording and syntax go to AI. Judgment, feedback and “how do we do it here” go to a person.
  • Write down what a chatbot can’t know. Company context only reaches AI if someone records it. In the Clutch survey, 61% of workers would ask a company-aware AI tool before a coworker. That only helps if the context it draws on is current.
  • Share answers back. When AI solves a problem, post the fix where the team can see it. It costs a minute and stops the next person solving it alone.
  • Make asking cheap for new people. Name someone they can ask anything, and protect time for it. The questions they’re embarrassed to ask are the ones that teach them most.

The questions will keep moving to AI, and most of them should. What’s worth keeping is the habit of asking each other the ones that matter.

Sources