People are letting AI into their shopping, just not into their wallets. That's the pattern hiding inside a year of adoption data, and it's more interesting than either the hype or the backlash suggests.
Roughly 73% of consumers now use AI somewhere in the buying process. But only about 13% say they've actually completed a purchase after an AI assistant referred them. The gap between those two numbers is where agentic commerce currently lives.
The funnel has a shape now
Research published this month put usage at about 62% for product comparison, 23% at checkout, and 19% for post-purchase tasks like returns and tracking. Adoption is front-loaded — heavy where the work is annoying and the stakes are low, thin where money actually moves.
Look at what people ask AI to do and it makes sense. About 45% use it for product ideas, 37% to summarise reviews, 32% to compare prices. Those are all research chores: reading forty reviews, checking whether a price is genuinely good, remembering which model number has the feature you want. Nobody enjoys them, and getting one wrong costs nothing.
Checkout is different. Choosing the wrong size, the wrong seller, the wrong return policy costs money and time. So people take the wheel back.
What an agent is meant to do
The full promise goes further than research. A shopping agent is supposed to act for you — find, evaluate and buy without you searching, comparing or clicking between stores. It holds your preferences, scans thousands of products across dozens of retailers in seconds, and returns with a decision rather than a list.
Forecasts assume that's where this goes. One projection has the share of consumers using AI agents rising from 19% now to 46% by the end of 2026. AI platforms are expected to drive $20.9 billion in retail spending this year, close to four times last year's figure, and McKinsey has floated up to $1 trillion in US retail revenue influenced by agentic commerce by 2030.
The willingness is already there
Here's the part that complicates the sceptical read: about 70% of people say they're at least somewhat comfortable with an AI agent buying on their behalf.
So the holdup isn't fear. Comfort sits at 70%; completed AI-referred purchases sit at 13%. That's not a trust problem, it's an execution problem. The last mile of retail — accurate stock, correct variants, working payment, a clear returns path — is where agents keep stalling.
Which explains why the practical advice circulating among retailers right now is unglamorous: clean up your product data. Agents favour listings with complete, accurate, machine-readable information, because an agent can't ask a shop assistant what "medium" means on this particular brand. Ambiguity that a human shopper resolves with a shrug and a guess is a hard stop for software.
What this means if you're the one shopping
Use AI where it's already good and where the downside is small. Narrowing a field of options, summarising what reviewers actually complain about, sanity-checking a price — that's genuinely useful, and it's what most people are doing.
Be more careful when an agent is ready to transact. Check the seller, the return window and the final total yourself, because those are exactly the details that get flattened in a summary. Whether the agent charged you the right amount to the right merchant is not something you want to discover later.
The forecasts may well be right about where this ends up. But the current numbers describe something narrower and more sensible: people using a fast researcher, then making the call themselves. That's not a failure of the technology. It's a reasonable division of labour, and it may last longer than the projections assume.
Image: Erick Gielow, via Pexels
