November 8, 2026 · Commerce
Your next buyer may be an agent, and it will not read your homepage
Meta and Amazon spent the week fighting over who sits between a shopper and a catalogue. If it stops being you, a product page becomes a database row — and most B2B pages are not readable as one.
The two most interesting AI announcements this week were not about models. They were about a checkout.
On Thursday Meta introduced Spaces, a shared workspace where a team can work alongside its Muse agent. A day earlier, Amazon — which had just blocked Muse from reading its catalogue — began letting customers in the US text Alexa to shop. Block the other agent, ship your own. That is the entire strategy in one week.
As a platform fight it is ordinary. As a signal about where purchase decisions get made, it should bother anyone who sells things online.
What an agent actually looks at
A person browsing a product page absorbs things a machine ignores: the photograph, the tone, the line about being a family business since 1998, the fact that the page loaded before they got bored. An assistant looking for a supplier absorbs something much narrower — facts it can compare.
Price. Minimum order. Lead time. Incoterm. Certification. Stock. Carton dimensions. Whether you ship to Rotterdam.
Most B2B catalogues are built for the first reader and nothing else. Price hides behind "request a quote". The MOQ lives in a PDF attached to a different page. Lead time is a sentence in an email from 2023. The spec table calls the same attribute "Material" on one SKU and "Fabric" on the next.
To a human, that is friction. To an agent, it is a wall. And a wall does not produce a complaint. It produces silence, and a shortlist with your competitor on it.
The fair objection, and why it does not hold
Agents are bad at this today. They misread variants, invent shipping costs, and confidently quote a discontinued model. All true, and I would not build a business on agent traffic this year.
But the reader changed before the interface did. Procurement teams already paste supplier pages into a chat window and ask for a comparison table. Search engines already parse product markup to build shopping results, and that is where a lot of first contact happens. The agent is less a new audience than a stricter one.
There is a defensive reason too. Amazon just demonstrated that whoever owns the assistant can close the door on a rival's. If the buying layer consolidates into two or three assistants, the sellers who keep their place are the ones whose data can be read without negotiation.
Five things that take an afternoon
Publish a price band. You do not have to publish your best price. "USD 4.20–5.80 per unit at 500–2,000 units, FOB Ningbo" is enough for an agent to keep you in the comparison, and enough for you to defend in a call.
Write MOQ and lead time as sentences. Not in an image, not in a PDF, not behind a form. One line on the product page: "MOQ 500 units, 25–30 days production, 18 days sea freight to Rotterdam."
Use one name per attribute. If one page says Material and the next says Fabric, a parser sees two unrelated fields. Consistency is dull and it is the whole game.
Say what stock actually means. "In stock" that means "we can probably make it" is the fastest way to lose a buyer who planned a shipment around it.
Mark it up properly. Product and Offer schema with price, currency, availability and unit. Google Shopping reads the same thing, so the work gets paid twice.
What machine-readable actually looks like
It is less technical than it sounds. A block on the page, written for humans, that happens to stay consistent:
- Model: NX-240
- Material: 304 stainless, 1.2 mm
- MOQ: 500 units
- Price: USD 4.20–5.80 FOB Ningbo, 500–2,000 units
- Lead time: 25–30 days production, 18 days to Rotterdam
- Certification: CE, RoHS
- Carton: 40 × 30 × 25 cm, 12 kg
Every line is something a buyer asks in the first email. Put it on the page and you have answered it before the email exists — and a human reading it benefits just as much as a parser.
Where this does not apply
Custom manufacturing is safe for a while. When every order is a drawing and a negotiation, no assistant replaces the sales engineer, and pretending otherwise wastes money.
The shortlist stage is not safe, though. Even in bespoke work, buyers now ask an assistant who makes something similar before they ask you. Being absent from that first pass costs you the meeting, not the order — which is worse, because you never find out.
What it does to a marketing budget
If the decision moves earlier — into a comparison an assistant assembles before anyone visits a site — then money spent at the top of the funnel buys less than it did. The page that still matters is the one with the numbers on it.
That is not an argument for gutting brand work. Buyers still need a reason to trust you once you are on the shortlist. It is an argument for moving specification work off the sales call and onto the page, where both kinds of reader can find it.
The test worth running this week
Open two assistants. Ask them to quote something you sell: a model number, a destination port, a quantity. Do not help them.
Whatever comes back — right, wrong or empty — is your real product page. If the answer is "I could not find pricing", you have just watched a buyer leave without saying anything.
The fix is not a redesign. It is a dull afternoon with a spreadsheet and a content editor. Do it now and you are early. Do it in two years and you are cleaning up someone else's shortlist.
We publish the pricing model, minimum commitment and export-readiness notes for every tool in the library, and we flag it when a vendor changes one without changing the headline price.