Something reads your page before a person does.

Get read correctly.

I'm Ron Davis. Eight years inside AWS Marketplace, founder of an AI-powered GTM platform. By the time a buyer opens your page, an assistant has usually already read it and decided whether to name you at all. This page is three things: what that reader actually sees, how to check your own page for free in about fifteen minutes, and what to do when the check comes back thin.

01 — What a machine reads

I read 595 pages the way a machine reads them. Most of them said nothing it could use.

Not a survey. One corpus, read field by field, with the numbers that survive someone checking them.

// what I read, said plainly

When I say listing, I mean one specific thing. An AWS Marketplace listing is the page a software vendor writes to sell their product — vendors from two-person shops up to names you would know. They write every word on that page themselves. I read 595 of them the way a machine reads them.

That corpus is software vendors, not a random sample of the web, and I won't pretend otherwise. But nothing below is a marketplace problem. It is a writing problem, on a page the company controls.

81%
482 of 595
open with a superlative — leading, best-in-class, world-class. A word that changes nothing if you delete it.
37%
219 of 595
carry nothing on the page a machine can check. No number, no date, no named customer, no third-party record.
14
of 52 sellers
had an llms.txt file — the plain-text file that tells an assistant what your site is. The ones selling AI were no likelier to have one than the ones selling storage.

The reason isn't laziness. You cannot read your own page cold, because you already know what it means. The buyer arrives with none of that context — and the machine the buyer brings has less than the buyer did. It doesn't infer, it doesn't give benefit of the doubt, and it will not ask you a follow-up question.

Almost twenty years ago I helped a company write their first set of service offerings. They sold service catalogs for a living and had never written one of their own. I have been finding that same bug ever since, and this year I found it in my own site. The machine did not create it. It just stopped forgiving it.

02 — The free self-read

Three checks. About fifteen minutes. Nothing to buy and nothing to install.

Run them in this order on your own front page. Each one is written to be done exactly as it reads.

Read it out loud, in a private window.

2 minutes

Open your front page in a private or incognito window — no logged-in state, no autocomplete, nothing your browser remembers about you. Read it aloud from the top, every word, without scrolling ahead.

The sentences you skip are the ones you wrote for yourself. Out loud is the whole trick: silently, you read what you meant. Aloud, you read what is there.

Send it to one person who has never heard your pitch.

5 minutes

Send the URL — nothing else, no framing, no "so what we do is." Ask exactly one question: what does this company do, and who is it for?

Then say nothing until they answer. If they hesitate, a machine already did — and the machine had less to work with than they do.

Delete every sentence a stranger could not check.

10 minutes

Copy the first hundred words of your page into a blank document. Delete every sentence that contains no number, no date, no named customer, and no third-party record — certification, audit, review, filing. Adjectives do not count. Trusted, leading and enterprise-grade all get deleted.

Count what is left. 219 of the 595 pages I read had nothing left at all.

If all three come back clean, you do not need me, and I would rather you found that out for free. It happens — not often, but it happens.

03 — The audit, for any page

When the checks come back thin, the next question is which fix first.

That is the part you cannot do from inside your own head — it needs someone who has never heard your pitch.

The Second Buyer Audit is a written document, delivered in 10 business days. Not a call, not a scorecard, not a number out of a hundred. Six findings, and a fix list in the order to do them:

Where you get filed — which category an assistant puts you in when it answers, and who it names beside you.
Where your words go — title, meta description, H1 and the first hundred words, measured against what buyers search and what your competitors already own.
The evidence surface — named customers, dated numbers, third-party records, and what survives once the adjectives come out.
Agent-mode answerability — whether you carry the facts that answer the security questionnaire, the RFP row, the plain "does it do X."
A side-by-side — against whoever the assistant actually names when a buyer asks your category question. Not the competitors in your board deck.
A ranked fix list — sequenced by what shifts the read first. You can hand it to your team on day 11 and start.
// two tracks, one price

Any website, or an AWS Marketplace listing. You pick the track at intake. Same six findings, same price — the track only decides where I look.

$2,500 one-time. It credits 100% toward the 8-week sprint if you book within 30 days. The first three sprints go at a founding rate in exchange for a named case study.

Find out what the machine actually sees.

Run the three checks first — they are free and they are above. If the page comes back thin, the audit tells you which fix comes first.