The short version
Somebody makes an argument. Rather than the argument being addressed, something about them is — their motives, their record, their employer, their character. The conversation moves there. It does not come back.
Where the name comes from
Latin: argumentum ad hominem, an argument directed “to the person.” The phrase is older than the meaning it now carries. For a long stretch it named something narrower and quite respectable — an argument built out of the opponent’s own stated commitments, showing them that their principles don’t get them where they think. The modern sense, play the man and not the ball, is the one that won.
What it looks like
The budget doesn’t balance. There’s a hundred grand unaccounted for on line nine.
Yeah, well, you’re an asshole.
The impact assessment is using traffic figures from 2019.
He’s been against this development from the day it was announced.
This architecture won’t survive the migration.
Easy to say when you weren’t here for the last one.
The study found no effect.
That lab takes industry money.
The first one is the version everybody can already name, and it’s also the least dangerous, because nobody is fooled by it — including the person saying it. It doesn’t need an explainer. You can hear it working.
The other three are the ones that do damage. They arrive in a reasonable tone, they’re frequently true, and they get treated as though something has been answered. Every reply on this page could be perfectly accurate about the person it describes. Not one of them is an answer.
Why it doesn’t work
A claim and the person carrying it are different objects, and arguments don’t inherit properties from their couriers. A liar can hand you a true sentence. A person of flawless character can hand you a false one, sincerely, at length.
What makes the move so effective is that it’s almost relevant. Motive genuinely is useful information — it tells you where to look, how hard to check, which numbers to pull first. It is a reason to audit the claim. It is never a substitute for auditing it, and the whole trick is the swap between those two.
It’s also enormously efficient, socially. It converts a question that would require work — go and find out whether the figures are from 2019 — into a question everybody in the room already has an opinion about. Nobody needs to open the appendix once the subject is whether he’s a NIMBY. The discussion gets louder, more confident, and completely detached from anything checkable.
And the original claim doesn’t go anywhere. It’s still there when everyone’s finished, unexamined, and now slightly radioactive: whoever raises it next inherits the accusation along with it.
This is not ad hominem
- Credibility, where credibility is the evidence. If someone’s testimony is the case — an eyewitness, an expert judgement you have no way to check — then their reliability is directly on point. That’s not attacking the argument, it’s weighing the evidence.
- A conflict of interest, raised as a reason to verify. “They’re industry-funded, so I’d want independent replication” is a proposal. “They’re industry-funded, so it’s false” is the fallacy. Same fact, completely different job.
- Rudeness. Insulting someone while answering them is bad manners, not bad logic. “That’s wrong, and here’s why, and you’re being a prick about it” has addressed the argument.
- Arguments where the person is the subject. In a hiring decision, a trial, or a conversation about whether to trust somebody, their character is the actual topic.
The test is a swap: would the argument still be wrong if somebody else had made it? If it would, say why. If you can’t, then you haven’t answered it, whatever else you’ve done.
How to call it out
- “Say you’re right about him. Are the figures from 2019 or aren’t they?”
- “That’s a reason to check it harder. It isn’t a reason to skip checking it.”
- “Granted on the funding. What’s the answer to the question?”
The strongest move is to concede the personal point immediately and completely. It costs you nothing — you weren’t defending him, you were defending a sentence — and it removes the only thing the attack had to feed on. What’s left is the original question, sitting in the open, with the conversation’s exit blocked.