Two Thirds of Buyers Want an AI to Deal With Sellers. Only 6% Have Tried It

Two surveys of US consumers landed in the past four weeks, and read together they say something more useful than either does alone. Capacity published The Agentic Consumer Report on 27 October, drawn from 1,000 consumers. Lorikeet fielded its own survey of 1,966 consumers between 30 September and 8 October, census-matched, ±2.2 points.

The headline is not that AI agents are about to start calling your counter. It is that they are already briefing the person who does.

Demand is huge. Delegation is not — yet

Lorikeet found 66.4% of consumers interested in an AI agent that deals with companies on their behalf. Actual use is a fraction of that: 13.9% say an AI tool has dealt with a company for them, and only 6.3% say it contacted the company itself. Asked what they want right now, 53.8% would still rather handle it themselves and 32.9% want the AI to advise them rather than act.

Machine-to-machine ordering is coming, but it is not landing this quarter. What has landed is the advised buyer.

The buyer arrives already armed

Capacity found 55% of consumers have used an AI tool to help with a service request — and of those, 44% hit information from AI that contradicted a company representative or its policy. Twenty-three percent now ask representatives to explain or cite company policies. Fifty-eight percent say they would be more likely to push back after a request was denied if they had used AI first.

For a parts seller that lands squarely on the details that are hardest to publish and easiest to get wrong: fitment across model years, core charges and how they come back, freight on a long item, warranty claim steps, which electrical parts are non-returnable at all. A shop manager with a truck on a lift reads a confident AI answer about a core charge, calls the counter, and hears a different number. Whoever can cite the real policy in thirty seconds keeps the order — which matters to an independent distributor as much as to a dealer network.

Lorikeet’s comfort data points the same direction. Buyers are most willing to hand an AI the job of getting a better price, at 55.5%, well ahead of making a complaint at 41.8% or making a booking at 38.5%. Price and terms are where the AI gets used first.

Where parts sellers should start

Three things, in order.

Write the policies down as plain text on your own site: returns window, core charge and how it is refunded, freight rules, warranty claim steps, the non-returnable list. Not a PDF, not a phone-only answer. Text is what an assistant reads and quotes, and right now it is quoting somebody.

Then take the ten questions your counter answers every day and check that every person and every channel gives the same answer. That 44% contradiction figure is mostly not AI inventing things — it is businesses whose own answers differ depending on who picks up.

Last, close the hours gap. Trucks break at night, and a buyer holding an AI’s answer will not wait until Monday to test it. Something that answers calls, texts and chat with your live pricing and stock, in English or Spanish, and hands a warranty dispute to a person, turns that advised buyer into an order instead of a callback. The sale you lose to a wrong AI answer never appears in your reporting. It appears as a part bought somewhere else.