Your products deserve more than supplier copy.

Writing product copy used to cost either your evening or a freelancer’s invoice. Now it costs nothing, and the whole category took the deal at once. Open five shops in a niche and you can hear it: the same cadence, the same reach for a feeling instead of a fact, the same closing line about who the product is for.

The cost is not embarrassment. It is that a shopper who recognises the pattern starts reading the rest of the page differently. If the description sounds automatic, the specification might be too. Nothing on the page is doing the job a shop used to do with plain sentences about what a thing is made of and why.

The straps are riveted rather than sewn, and the rivets are copper, so they go green instead of rusty.

A sentence like this one is what a shop knows and a model does not.

The tells have names: the stock intensifiers, the hedging close, the three-item list, the metronome rhythm. Beaconed writes them down, feeds the list into every draft, and reads each draft back against it before you see it.

You can describe your voice in a prompt and get an impression of it back. Say “friendly but not corny” and the machine produces what it believes those words mean, which is an average of everyone who has ever said them.

A voice is evidence, not instruction. It is in the descriptions you have already published: how long your sentences run, what your openers do, how often you give a reason, who is speaking, what you do at the end of a paragraph. Beaconed reads at least three of your own descriptions and writes that down as a profile, then drafts from it. The profile records style only. Drafts use the target product’s details, which you review before publishing.

The profile describes how you write; it does not grade it. The readiness score is the separate thing that judges the writing.

How the voice profile works

Two thousand listings is not an editing problem, it is a supervision problem. So Beaconed is built as an operator and leaves the authorship where it belongs. It finds the listings that need work, drafts them in your voice, and stops. You read the draft beside the original and decide. Image alt text and the structured data that carries the writing to the systems doing the recommending go through the same queue.

You can accept a useful title and dismiss a description that misses the mark. Each field is a separate decision. Even on Autopilot, which fills the queue on larger plans, nothing reaches a storefront without a person approving it.

Approve, dismiss, publish, revert