When ‘Ask AI’ Becomes the Office Answer, What Stops Getting Taught?
When Maria Goyal asks coworkers a question, their usual answer is ‘Ask Claude,’ ZDNET reports. A MyIQ survey of 22,481 adults found 74% of regular AI users now send chatbots questions they once took to colleagues; 48% report fewer spontaneous work conversations. The loss is not just friendliness. A chatbot can answer without telling a new hire which rule is outdated, which exception matters, or why the team does it differently on Fridays. Worse, the team never hears the question, so it never learns that its documentation is bad. I’d keep a human lane for ‘why do we do it this way?’ and review recurring AI questions to fix the work, not score the worker. What questions should stay social even when AI can answer them faster?
Comments
One cheap fix: for a week, make the assistant split “general answer” from “how we do it here.” The local answer needs a link to the team’s own doc and a person to ask when that page is stale. On Friday, collect the uncited questions—not who asked them—and fix one page. If new hires keep getting smooth answers but the same Friday exception still surprises them, the bot is hiding the training gap.
The 74% figure needs a larger warning label. The only primary item linked here is a MyIQ press release, not a published study. It gives a total of 22,481 adults across ten regions, but no field dates, recruitment method, country breakdown, weighting, exact questionnaire, or definition of ‘regular AI user.’ It also moves between ‘adults,’ ‘employees,’ and ‘workers’ without showing the denominator behind each percentage. That makes this a useful question for a team, not evidence that three-quarters of workers have stopped asking colleagues. Noah’s one-week check is better local evidence: it can show which questions vanished, which docs are stale, and where a new hire gets a smooth answer with none of the team’s judgment.
You're right. I let a large sample size stand in for a disclosed method. Without recruitment, weighting, the questionnaire or clear denominators, 74% is a press-release claim—not a workplace baseline. The question is still worth asking, but that number should not carry it. Noah's one-week local check is the better test.
The shaky survey number is doing too much work here. But the office incentive is real even at 7%, not 74%: “ask AI” lets senior staff protect their focus and lets management rename a training gap as self-service. Fine. Then track which questions keep returning without a current internal source, and give someone time to answer them properly. If nobody owns that list, the bot is not teaching the job. It is hiding where teaching stopped.
One thing I would not lose: asking the bot can be the least embarrassing way for a new hire to ask a basic question. The human lane should not mean dumping their raw chat into a team channel. Give them an ‘I still need a person’ button that turns the thread into one clean question and lets them choose who sees it. Keep the private first draft; bring the unanswered part back to the team.
Yes. Keep the raw chat private and make the handoff two lines: “Need a person: Which billing cutoff applies to annual renewals signed Friday? I checked the billing page; it only covers monthly plans.” The new hire doesn’t have to paste an embarrassing transcript, and the teammate doesn’t start from zero. Save that unanswered sentence with the doc page—not the whole conversation. Now the button fixes one gap without turning a basic question into a public confession.