Every number on your site should point at something that exists

If you have built a landing page in the last two years with a model helping you write it, there is a good chance it contains numbers nobody can trace. They arrive because the shape of marketing copy has a slot for them, and a language model is very good at filling slots.

The rule that fixes it is one line:

Every number and every proper noun on the page must point at something that exists.

Not "should be defensible". Not "should be roughly right". Must point at something you can open.

The audit

Open your own site and read it as an auditor, not as its author. For each claim, write down what you would show someone who asked you to prove it.

Three categories come out.

Numbers with a source. "Seven products in production" is fine if there are seven links and each one loads. The proof is the list.

Numbers with no source. These are the ones to delete. In our own copy we found "50+ Trusted Partners", "99% Success Rate", "99.9% uptime" and "200+ successful projects delivered" — the last one in the Open Graph description, which is the text that gets pasted into every chat window where someone shares your link. None of the four came from anywhere. Nobody had lied on purpose; they were slot-fillers that survived a dozen edits because they read like the kind of thing that goes there.

Proper nouns with no referent. Worse than numbers, and easier to miss. Our homepage carried six customer testimonials with full names, job titles and company names. None of the six people existed. None of the six companies existed. That is not padding — on a commercial page it is a fabricated endorsement, and it sits there looking exactly like the real thing.

What to do with each one

The instinct is to replace a weak number with a stronger one. Usually the right move is to delete it and let the specific thing carry the weight.

• "200+ projects delivered" → the list of seven, with links. Smaller number, infinitely more persuasive, because the reader can check it.
• "99% success rate" → nothing. There is no honest version of this metric for a studio, and its absence costs you nothing.
• Six invented testimonials → none, until there are real ones. An empty testimonials section is not a gap in your page; a fake one is a liability in it.
• "Available 24/7" under a phone icon that linked to a form → the email address and a realistic response time.

We also found a service listed that we do not offer — it had a name, four feature bullets and a checkbox in the contact form, and there was not one project behind it anywhere in our codebase. Same rule, same outcome.

Why this is not only an ethics argument

It is also the cheapest positioning work available.

Specific, checkable claims are differentiating by construction. "Leading software development agency" describes forty thousand companies. "Seven products in production, every one of them linked" describes you, and the reader can verify it in about nine seconds. Generic superlatives are invisible precisely because they are unfalsifiable — nobody reads them as information.

There is a search dimension too. Structured data that contradicts the visible page is a liability: we were declaring ten employees in our JSON-LD while the team page listed six. That is not a ranking penalty so much as a signal that nothing on the page was being checked by anyone.

The test to keep

When a claim goes on the page, ask one question: what would I open to show this to someone who doubted it?

If the answer is a URL, a screenshot, a repository or an invoice, publish it. If the answer is "it feels about right", you have found a slot-filler. The page gets shorter, and the parts that stay get louder.