The number I refused to write

The request was not dishonest. It came from calendar pressure and a sincere belief that the order of magnitude was known. It was simply not workable.

Why that particular number could not be approximated

The organisation in question sells traceability. Its promise fits in one sentence: what we state about a product's origin can be verified.

Publishing at the top of its homepage a percentage that nobody in the structure can source amounts to demonstrating the opposite of the promise, in the first screen, ahead of every other argument. The first visitor who asks for the calculation method takes the credibility of the entire rest of the site with them.

That reasoning holds beyond this sector. A brand can round a growth figure without much risk. A brand whose product is verification cannot, because the approximate number does not attack its communication, it attacks its offer.

What I did instead

I looked for the data in public sources for the sector. Nothing publishable came out of it, in the sense that nothing carried a citable methodology and a date.

So I turned the request into a written question to the operations director, with the expected shape of the answer: the value, the period, the source, and who signs it.

And I left the space empty in the layout. Not a provisional figure in brackets, not a cautious range. Empty, with the open question beside it.

A gap is information, provided you hold it open

A slot left empty in a layout is uncomfortable, and that is exactly its use. It stays visible at every review, it sits on the agenda, and it eventually produces either the data or the decision to do without it.

A provisional figure, by contrast, disappears. It blends into the page, it survives the proofread, and it goes to production. Almost every false number I have had to remove from existing material started life as a placeholder nobody recognised as one any more.

I set a cut-off date: with no answer by that date, the key-figures banner is dropped and the page stays on provable facts alone. The date serves as much to obtain the answer as to authorise dropping it.

The same rule applied twice more on the same file

A mention of an administrative registration appeared on eight pages of the site. No document in the file supported it. It was removed from all eight, and not replaced.

And the reverse happened too, which matters just as much. An outside opinion recommended deleting a figures banner it believed to be fictional. The evidence existed. The opinion was set aside, with proof. The rule is not to strip everything out of caution, it is to write only what is held.

The part of the site that genuinely misled

On the same audit I found something worse than the missing number. A reporting form, presented by the site as the organisation's ethical guarantee, was in fact inoperative: the credentials it used were not real credentials.

It was the only place on the site that actively misled the visitor, and it carried no figure at all. It carried a promise of recourse.

That is the lesson, and it is wider than the question of numbers. Truth checking happens on figures because figures are easy to verify. A site's most expensive lies are elsewhere: in announced mechanisms that do not work, addresses that never answer, certifications that have been under renewal for three years.

What it costs, and what it returns

Refusing costs an uncomfortable conversation and sometimes a delayed publication. I had both.

What it returns is hard to measure and easy to describe: the day an investor, a regulator or a journalist asks where a homepage figure comes from, somebody in the company can answer. On the files where that answer does not exist, the question only gets asked once.