Sometimes, the question is wrong

Ellie Clifford / 2026-04-10

A while ago I had a conversation with a psychiatrist which went something like:

Psychiatrist: Sometimes, people with ASD¹ (like you) will interpret a question differently to how it was intended, usually more literally, and then when someone points it out, say “I wasn’t wrong, the question was wrong.” But how can the question be wrong, when 99/100 people understood what it meant?

Let me give you an example. Often people come to see me for a diagnostic assessment, and I say, “What is your home like?”. When they respond, “Well, it has four walls and two floors, and it’s Victorian”, they have answered literally correctly, but still haven’t answered the real question².

Me: Okay, sure, in that situation, the question wasn’t “wrong”, but in that case it’s not about literal interpretation, it’s about understanding the context, and modelling why you are asking the question. And often the question really is wrong. I’m sure I could come up with examples pretty easily. Besides, focusing on the 99/100 misses the point — if the 1/100 suggests that you could have phrased it in a better way to get 100/100 on the same page, shouldn’t you listen to them?

Psychiatrist: Interesting food for thought, but let’s move on.

I didn’t make my best arguments at the time, I didn’t actually give any examples, and we never came back to that. So clearly the only reasonable thing to do is not-so-vague-post about that argument on my blog over a year later³. Here we goooo…

Sometimes, the question requires unnecessary modelling

Questions that require you to have an excessively detailed model of the asker in your head in order to understand how to respond are hostile to neurodivergent people, and generally lead to worse outcomes among everyone.

Now, it’s true that having a good model of the asker sometimes really is necessary to answer a reasonable question well⁴. If you’re getting ready to go out and your girlfriend asks you “Does this dress make my shoulders look too wide?”, there might be many true things you could say, but you need to know her well enough to know whether she’d prefer you said “You look great!”, “Like, maybe a little, but in a hot butch lesbian way not in a masculine way”, or “Kinda, but it would work if you styled it with this jacket?”. Each response would be appreciated by some and annoy others⁵, and there isn’t always a way to avoid this without finding yourself 3 hours into a conversation that continues to recurse further and further into the meta⁶.

On the other hand, a question asked by someone you’ve just met, or in a context where precision is quite important (like talking to a medical professional) is not at all the same, and almost zero modelling should be required. A question on a form like “How often do you feel anxious (very often/often/sometimes/rarely)” is a bad question. What does “sometimes” mean to the asker? It doesn’t seem unlikely that someone might have been living with anxiety all their life, and then think “oh, these days I only have one panic attack a week, I’ll tick ‘sometimes’”, while the asker expects “sometimes” to mean a little manageable anxiety in particularly difficult scenarios. In this example the question should be concretely time-based (“about once a week”, “about once a month”), and probably also define how strong the anxiety should be to register.

Once upon a time, I had to fill in a similarly vague question on a medical form, and I asked for clarification from the person who had given me the form, and they essentially told me that they couldn’t answer my question, because that would bias the results.

????⁷

Sometimes, only a few recognise what’s wrong

Autistic people object to questions more often. But is this only because autistic people are more likely to struggle with doing all the modelling required to answer them, or is it also because autistic people are more likely to recognise problems in questions that affect everyone, and then write long ranty blog posts about them? Some of us seem to have the ability to (or inability not to) kick up a fuss when something is wrong, rather than just going “eh, whatever” — or, worse, typical mind fallacying into thinking that only autistic people could possibly have any issue landing on the same meaning of “often” as you.

I think this post itself is probably proof enough of my point.

Sometimes, the question is malicious

Malicious questions are everywhere. You’ve probably mostly seen them when politicians try to catch each other out, or in courtroom dramas, but they really are everywhere. Sometimes this takes the form of a question presented with a forced binary: classics include the relatively benign example of monogamous people asking their partners “Do you think my friend is cute?” (“Yes” means you have eyes for someone else (bad), and “No” means you think the friend is ugly (bad)), and the less benign “Do you agree to mass surveillance, or do you love child abuse?” (see Ross Anderson’s “Chat Control or Child Protection?”).

But sometimes, the malice is more subtle. A common question on forms is “What is your biological sex? (Male/Female/Other)”. It’s malicious, but not because of a forced binary (when the “Other” option exists), rather it’s malicious because answering in the way intended by the asker (i.e. with sex assigned at birth) would tacitly accept a malicious worldview⁸. On the other hand, for many people, answering truthfully to the question (insofar as it is even answerable) wouldn’t challenge the malicious worldview, rather it would result in the asker interpreting correct information incorrectly (which might be fine, but in very limited medical contexts it might be important that this doesn’t happen).

Of course, there are many other ways for a question to be malicious, and many more examples I have seen. But the latter example is one where while 99/100 could answer easily, 1/100 recognise it as not just wrong, but an attack on our recognisability as people.