Gartner surveyed 3,566 business and consumer customers in February and March, and published the results in August. Half said their interactions are easier when a company uses generative AI. Yet 87% said it is essential that the company offer a way to reach a human agent.
Both numbers are true at once, and most support teams hear only the first. Here are four common objections to adding the human option, with an answer to each.
"The AI already handles most questions"
It may. Gartner's finding does not say the AI fails. It says customers want the exit even when it works.
That is a different problem from accuracy. A customer who never needs the person still weighs whether the person exists. The most common thing that would win over customers unwilling to use AI, Gartner reports, is the ability to switch to a human when needed.
"Humans cost too much"
They do, per conversation. So spend them where one conversation is worth the most.
A shopper asking where her parcel is does not need a person. A wholesale buyer with forty cases in the cart and one open question does. Gartner's analyst says AI should gather information and identify intent, and try to resolve only when its confidence is high. That leaves the people for the cases the AI is unsure about, which are also the ones with money attached.
"Buyers prefer self-service"
Many do, and Gartner's data agrees. Customers were about three times more likely to use a third-party AI tool than a company's own chatbot in their latest service interaction. In B2B, 74% of customers who use generative AI had used it to complete a task for them.
Buyers are already doing the research alone. What they have not given up is the right to ask someone when the research stops. Offering a person on a call next to the cart does not interrupt self-service. It finishes it.
"A human button invites every question"
Only if you put it everywhere. Gartner's advice is not to make AI a mandatory first step for every issue, and it does not follow that every page needs a big red button.
Place the option where the page runs out of answers: beside the cart, on the delivery estimate, next to a spec table with a missing line. Offer it once, quietly, and let the customer decide. Our post on when a chatbot should hand over to a person lists the moments worth watching for.
One rule that settles all four
Write a single sentence your team can test: a customer can reach a person from any screen where the AI has failed twice, in one click, without repeating the question.
Then check it. Ask someone outside the team to try. If it takes more than one click, or the person asks what the problem is again, the option exists on paper only. The survey gives no click count, so use the test as your own standard and tighten it each month.
Source
Gartner, "Gartner Survey Finds 87% of Customers Say Companies Using GenAI for Customer Service Must Provide Access to a Human Agent", press release of 4 August 2026 (3,566 B2B and B2C customers surveyed in February and March 2026; 87% call a route to a human agent essential, 50% say interactions are easier with GenAI, customers about three times likelier to use third-party GenAI tools than a company chatbot, and the analyst's advice not to make GenAI a mandatory first step). Gartner's own page blocked automated reading, so the figures were read from the Insurance Canada republication of 11 August 2026 and not from Gartner's page. The survey mixes B2B and B2C customers and does not measure live video. The objections and answers are editorial, not Gartner findings. Photos: Wikimedia Commons, CC0 (Telephone Exchange, Adelaide Railway Station, 1929, State Government Photographer; police telephone switchboard, East Riding Archives; Phone Helpdesk at the 38C3, Leonhard Lenz).
