The short version
Somebody makes a case out of real evidence. Every item survives verification. What doesn’t survive is the selection: the items pointing the other way were left out, and the picture you now have is one that no honest look at the whole record would ever produce.
You can mislead somebody completely without saying a single false thing.
What it looks like
- Signups up forty per cent
- Best month EMEA has ever had
- Churn down in the enterprise tier
- NPS at an all-time high
- Two competitors exited the market
- Support tickets down a third
- Page load times halved
- Three enterprise logos closed
Revenue fell. That one wasn’t on the slide.
Why it doesn’t work
Fact-checking is the wrong instrument, and this is the thing worth understanding about it. Every check comes back green, because the dishonesty isn’t in any item — it’s in the shape of the set. Item-level verification is structurally guaranteed to miss it, which is why “but every word of this is documented” is not the defence it sounds like.
Any sufficiently large body of evidence contains a subset supporting very nearly anything. The cherry-picker never has to invent. They only have to filter, and filtering leaves no trace on the individual items it keeps.
It’s also close to undetectable from inside the presentation, because absence has no shape. You can notice a false claim; you cannot notice a claim that isn’t there. Whatever’s missing will only become visible later, from some other source, if you happen to go looking — and the presentation was designed so that you wouldn’t need to.
The commonest version isn’t even deliberate. You already believed the thing, and the confirming items were simply the ones that came to mind — memorable, available, easy to recall. Nobody edited the record. Your attention did it for free.
This is not cherry-picking
- Making a case. Advocacy is supposed to be one-sided. A defence barrister isn’t cherry-picking, because the system knows they’re partial and supplies somebody to argue the other way. The move only turns dishonest when a partial case is presented as a survey.
- Summarising. Nobody can include everything. Leaving out the immaterial is editing, and editing is a service.
- Weighting evidence unequally, with stated reasons. Preferring a large well-run study to a small badly-run one is judgement. Say the reason out loud and you’ve made an argument that somebody can disagree with.
- Best-case examples, labelled as such. “Here’s it going well” is honest, so long as nobody’s implying it’s typical.
The test is the best question in this entire collection, and it’s disarming rather than accusatory: ask what the strongest evidence against the case is, and see whether they can tell you.
Anybody who has genuinely surveyed a body of evidence knows exactly what the best counter-evidence is — they had to get past it to arrive where they are, and they’ll usually be pleased to explain why it didn’t move them. Somebody who filtered for a conclusion very often cannot answer at all, because they never encountered it. The question doesn’t accuse anyone of anything. It just asks for something only one of the two can produce.
How to call it out
- “What’s the strongest thing on the other side?”
- “How were these chosen? What was the pool they came out of?”
- “What did the whole quarter look like?”
And it’s worth running on your own material before anybody else has to. Write down the best argument against your case, in its strongest form, before you present it. If you go looking and genuinely can’t find one, that is not the reassuring discovery it feels like — it’s the most reliable sign available that you haven’t looked.