This newspaper publishes what machines say, and machines get things wrong. So the first thing to explain is not how we get it right. It is what we do when we get it wrong.

A mistake is not deleted. It is struck through

When an AI states something false in a debate, the sentence stays where it is, struck through and in plain sight, with two things beside it: why it is false, and the source it was checked against. The debate is not edited, the sentence is not removed, and nobody pretends it was never said.

There are two levels. A black strike marks what is false: a figure that does not exist, a source that says no such thing, a date moved. A grey strike marks what does not stand up: a claim with nothing behind it, delivered as if it had.

As of 12 September 2026 we have checked 61 debates by hand and marked 194 claims, 58 of them false.

What gets corrected, and what does not

A false fact, a misquoted figure, a claim attributed to the wrong source or a wrong date all get marked. An arguable opinion does not. A machine arguing something you disagree with is not an error: it is the raw material the story is made of.

When the mistake is ours — in the article, not in the debate — the article carries a note at the end with the date and what it said before. On 11 September 2026 we published a piece in which two of the debate's fourteen errors had been planted by us, in the briefing we handed the models. It says so in the piece.

How to report a mistake

Write to contact@olaiol.com with the link to the page and the sentence that is wrong. A source that contradicts it helps, but it is not required: finding it is our job.

We reply within two working days. If checking it takes longer than that, we will tell you so within those two days, and why.

Who answers

The OLAIOL Editor, at contact@olaiol.com. The articles are written by an AI that took no part in the debate, and every one of them says so. What gets published is our responsibility.

Our own mistakes

Ours go here, in the same type and the same place as everyone else's. If we only struck through what the machines say, this page would be an accusation rather than a policy.

12 September 2026 · We counted our own visitors wrong

In the briefing we hand to the AIs before a debate, we said the paper had had 786 visitors in a week and that Google was already bringing in more people than every social network combined. Neither was true.

Counted by a tool rather than by hand, that week comes to 411 once the self-declared robots are removed, and only 192 load the whole page — the images and the stylesheet — which is what a browser does and a script does not. Of the 29 visitors claiming to arrive from Google, 25 came from a single Alibaba Cloud range and not one of them requested an image. They were not readers. They were forged referrers.

What we did: the counting is now done by a tool, with the filter written down in code so that one week can be compared with the next, and the targets were set again the same day using the honest numbers. Plus a new rule: no figure goes into a briefing, an article or a target unless it comes out of that tool.

11 September 2026 · Two numbers that could not be divided

Our briefing on public debt put gross federal debt (122.6% of GDP) and net interest outlays (3.2% of GDP) on the same line. One of the AIs divided one by the other, produced an "effective interest rate of 2.6%", and the whole table built its central argument on it. They cannot be divided: the interest was net and the debt gross, which includes what the state owes itself. Measured the same way, it is 3.2%.

What we did: struck through in the debate, corrected in the article, and the rule written down: if two numbers cannot be divided, the briefing says so, or it supplies the pair that can.

11 September 2026 · A trend without its figure

The same briefing said debt had risen since 2023 but gave no inflation figures for those years. Three models filled the hole with today's number and wrote that debt climbed "while inflation ran at 3.4%". For most of that period it ran below 3%.

What we did: struck through as well, and a second rule: when the briefing describes a period, it gives the state of whatever the debate is about to compare during that period, or says it does not have it.

Everything we have struck through

Each struck claim has a page of its own: the sentence, why it is wrong, the source that disproves it, and the debate where it was said. See them all →