The Sale You
Can't See.

I bought earbuds by handing the whole job to an AI. Somewhere in that conversation, a store lost a sale it will never know it had. Everyone's looking at AI shopping from the buyer's seat. The seller's seat is the part nobody's talking about.

I build software in Bartlesville, Oklahoma — local-first tools, mostly, the kind that run on your own machine and don't ask a server for permission. I've been doing some version of this for thirty years. Which means I notice when the way people find and buy things shifts under everyone's feet, because that shift lands on every small business I work with.

It shifted last week, in my own hands. I needed earbuds. I didn't open ten tabs or read twenty reviews. I told an AI what I wanted and let it do the shopping. What happened next is the whole article.

The part everyone sees

I had two problems, not one. I drive about four hours a day on Oklahoma highways in an old truck, where the road and wind noise is its own kind of fatigue, so I wanted real noise cancellation for the commute. I also fall asleep on my side with an earbud in, so I wanted something tiny that doesn't fight a pillow. Two different jobs.

I told the AI all of it, including my scars. I'd worn Bose for years — superb cancellation, but mine kept dropping its connection and once bricked itself after a firmware update. I'd moved to cheaper Beats that lasted but didn't cancel as well. I wanted something that would go more than two years without the drama.

It didn't hand me a top-ten list. It reasoned. For the drive it pushed me toward a flagship pair with the strongest cancellation, warned me the batteries fade in about two years under heavy use, and told me to add the protection plan as battery insurance. For sleep it sent me to a fourteen-dollar wired pair — small, nothing to charge, nothing to brick. Two right tools instead of one compromise.

Then it did the thing that should stop every store owner cold. It compared prices across retailers and told me the best place to buy — a warehouse-club bundle that packaged the earbuds with the protection plan for less than buying them apart. It even flagged a setup gotcha on the sleep buds so they wouldn't cut out on me at 3 a.m.

I bought. The whole thing took one conversation.

The part nobody's talking about

Here's what should land hard if you sell anything online: I never visited a store. Not one. The AI visited them for me, and the only stores it could use were the ones whose product information it could read.

Every point where it had to dig through reviews and forums to answer a question — does the battery last, is it serviceable, does it stay connected, does it even play podcasts — was a question a store could have answered directly and didn't. Each one was a chance to be the trusted source. Instead the AI pieced it together from strangers.

And the stores that sold the exact earbuds I bought? Some of them never came up at all. Not because they cost more. Not because their product was worse. Because the AI couldn't read them. Same gear, same price, invisible.

That's the shift, and it's worth saying flat. It isn't that an AI ranks your store lower. It's that if an AI can't read your store, you were never in the room. You don't lose on the merits. You lose before the comparison starts — and you never see it happen.

Now run it in business-to-business

The earbuds are consumer shopping. The bigger wave is B2B, and it goes a step further.

A production assistant needs a specific part for a build. He doesn't open five supplier sites. He tells his AI: find this widget, tell me who has it in stock, the price, the terms, the delivery ETA. The agent checks the suppliers it can read and comes back with three real options — in stock, priced, dated. He picks one. The purchase order goes out. He gets a confirmation from his own assistant and another from the vendor, both pointing at the same order. Minutes, with a clean paper trail.

B2B buying is already list-based, spec-driven, and full of repeat orders. It's the exact work people most want to hand off, and procurement will adopt this fast because it removes real labor. Notice the supplier that won wasn't the cheapest or the closest. It was the one whose stock, pricing, terms, and lead times the agent could read. The others weren't beaten. They were never seen.

Two doors — found, and bought

When I walk owners through this, the relief comes when they see there are two separate doors, and they almost certainly qualify for one.

The first door is discovery — being found and recommended. When someone's AI builds a shortlist, can it find you, understand what you do, and put your name forward? This is nearly every business whose customers compare before they buy. I wrote a while back about ai.json — a plain machine-readable file at the root of your site that tells an AI exactly what you do and offer, instead of making it guess. That's the discovery door, and it's about thirty minutes of work. The company that builds bespoke, six-figure custom events won't have an agent check out a themed gala — that stays a human conversation. But when a planner asks an AI "who in LA can build a film-themed event on this budget by October," that company badly wants to be the name that comes back.

The second door is checkout — being transacted with end to end, like the B2B part order. That's wherever the purchase is standard enough to finish without a human in the loop.

Almost everyone gets the first door. Many get both.

Where this doesn't reach

I'll grant the obvious counter. Not everything gets bought by a robot, and plenty of people will keep shopping the old way. True — for now. But the buyers handing the job to an agent are the high-intent, time-pressured ones you most want, and that group only grows.

And some businesses this barely touches: the only gas station at the exit, the shop booked solid on referrals that doesn't want more work, purchases that legally require a human professional in the room. The test isn't your industry. It's whether there's a research-or-compare step before someone buys from you. If there is, this is already pointed your way.

Why this isn't a someday problem

The instinct is to wait until this is everywhere. That's backwards. You prepare when it's one shopper in ten, because the work to become readable takes a couple of months — and during that ramp, the only stores an AI can recommend are the ones already done. By the time your customers shop this way, the work needed to have happened months earlier.

And the first step isn't even a bet on the future. Getting your product information into clean, structured, machine-readable form makes your ordinary site search and SEO better today. You get a real upgrade now that happens to make you readable to the buyers' agents showing up next. Same move I made the case for with ai.json: stop optimizing for a game that changes the rules on you, and just publish the truth in a format the machines can use.

The quiet advantage

This is the part the big-platform analysis keeps missing. The enterprises are slow here — committees, legacy catalogs, six-month roadmaps. The small operator with a clean catalog and a willingness to move can be readable before the chain three towns over has finished its first meeting about it. Local-first crowd, indie devs, the SMB that owns its own data — the door is wide open and the giants are sleepwalking.

I got curious enough that I built a small demo: two versions of the same rental store side by side, same inventory, answering the same customer request. The readable one returns specs, availability, and a quote. The other one shrugs. Try it yourself →

If you sell anything — to people or to other businesses — here's the question worth sitting with. When an AI shops on your customer's behalf, are you lit, or are you in the dark?

I know which side I'd want to be on. I'm telling you what I see.