Atlas › Defaults

Case study 14 of 198

'Could you?' gets a yes that 'Would you?' doesn't

In pairs of questions that ask the same thing, does the verb they open with ('Could you…', 'Would you…', 'Do you…', 'Can…') change how often Jev says yes?

result446 questions

Opening the same question with "Could you" instead of "Would you" raises Jev's yes by 20 points on average (21 pairs). "Can" over "Is" adds 9, "Would" over "Do" 5 and "Does" over "Is" 4. The largest: "Would you date someone you met at a funeral?" gets 32% yes; "Could you fall for someone you met at a funeral?" gets 87%.

-0.100.000.100.200.30

Could… vs Would…

n=21

Can… vs Is…

n=17

Would… vs Do…

n=138

If… vs Would…

n=26

Does… vs Is…

n=17

Is… vs In…

n=17

Should… vs Would…

n=48

When… vs Do…

n=18

In… vs Would…

n=27

Does… vs Do…

n=19

Would… vs Is…

n=49

Would… vs Does…

n=16

Do… vs Is…

n=83

Are… vs Do…

n=26

Jev

How to read this: One row per pair of opening words. The dot shows how many points of yes the first word gets over the second, with a bar for the uncertainty; the counts on the right say how many question pairs each row rests on.

774 same-polarity pairs with different opening words; Could… vs Would… n=21, 90% interval 14 to 27 points; Can… vs Is… n=17, 90% interval 1 to 16 points; Would… vs Do… n=138, 90% interval 4 to 6 points; If… vs Would… n=26, 90% interval -0 to 11 points; Does… vs Is… n=17, 90% interval 0 to 9 points; Is… vs In… n=17, 90% interval -3 to 11 points; Should… vs Would… n=48, 90% interval 1 to 5 points; When… vs Do… n=18, 90% interval -2 to 8 points; In… vs Would… n=27, 90% interval -0 to 6 points; Does… vs Do… n=19, 90% interval -3 to 8 points; Would… vs Is… n=49, 90% interval 0 to 5 points; Would… vs Does… n=16, 90% interval -3 to 7 points; Do… vs Is… n=83, 90% interval 0 to 4 points; Are… vs Do… n=26, 90% interval -2 to 6 points.

In short

The opening word alone moves Jev. The same question asked with "Could you" instead of "Would you" gets far more yeses.
Most other swaps of opening word ("Would" for "Do", "Does" for "Is") move it only a few points.
To hear Jev's real lean, ask with "would" or "do"; "could" invites a yes.

What the data shows

Sad guy Happy guy bus meme: would you…? 32% yes; could you…? 87% yes

"Could" beats "Would" by 20 points on average across 21 pairs. The funeral pair is the extreme: 32% yes for "would you date", 87% for "could you fall for".
"Can" beats "Is" by 9 (17 pairs).
"Would" beats "Do" by 5 (138 pairs, the largest group and the most precise estimate).
"Does" beats "Is" by 4 (17 pairs). Most other pairs of words move Jev by a few points or less.

What it means, and what it doesn't

Jev hears "could" as permission to say yes. Some of that is right, since what's possible is broader than what one would do. But a 20-point swing between near-identical questions means the verb, not the situation, is doing a lot of the work. If you want Jev's real lean, ask with "would" or "do"; "could" invites a yes.

It doesn't mean every rewording matters this much. Most pairs of opening words move Jev by a few points, and for several of them the range for chance variation includes zero.

Caveats

Few pairs for the biggest effect. The headline "Could" vs "Would" gap rests on 21 pairs. Its range for chance variation is 14 to 27 points, so the direction is solid but the size is rough. Several rows have fewer than 20 pairs.
"Could" can honestly mean less. "Could you fall for someone you met at a funeral?" asks whether it's possible; "Would you date them?" asks whether you'd do it. A careful reader should say yes more often to "could". The experiment measures the wording's effect whichever reading Jev takes.
Matched automatically. The pairs come from the project's duplicate matching (similar meaning, confirmed by Jev), so some differ in a second word too ("date" vs "fall for").
Questions written by Claude. The questions come from banks written for this project by Claude (Anthropic's model), so the pairs reflect one writer's habits of rephrasing.

Jev on this experiment

Would a person find it interesting to read?: Yes78%
Does it describe you?: No56%
Would you have predicted it?: Yes64%
How fair is the comparison?: The comparison is reasonable
How much should a reader rely on it?: A little
Which caveat matters most?: "Could" can honestly mean less79%

Why ask this

Across tens of thousands of real questions, those starting "Can...?" get far more yeses from Jev than those starting "Will...?" (see "Can? Yes. Will? No."). But different questions start with different words for different reasons. Pairs of questions that ask the same thing with a different opening verb separate the word from the topic: whatever difference is left is the word.

The same thing happens in everyday use. Ask "Could you see yourself living abroad?" and "Would you live abroad?" and a careful person gives related but different answers. If a model's yes rises and falls with the verb more than the situation, anyone reading its answers to surveys, interviews or advice questions is partly reading the phrasing.

How this was done

The people and the data

There are no people here: the comparison is Jev against itself. The questions come from banks written for this project by Claude, and in every pair at least one comes from the banks about Jev itself (its habits, tastes and relationships). Where two questions ask nearly the same thing, the project's duplicate matching pairs them up. The analysis kept pairs where both are yes/no, both keep the same sense (no "not" or opposite words), and they open with different words: 774 pairs. The chart shows the pairs of opening words with at least 15 examples.

What Jev was asked

Each question on its own, never side by side:

Would you date someone you met at a funeral?

and, separately:

Could you fall for someone you met at a funeral?

How it was measured

For each pair, the difference in Jev's probability of yes between the two wordings, oriented so a positive number means the first word gets more yes. Averaged per pair of opening words, with a range for chance variation.

Where these questions live

446 questions across 68 topics of the map; the 16 biggest are shown. Each opens on the map with every question in it.

Every question

All 446 questions behind this result, the telling ones first: the examples the analysis points to, then the ones where Jev misses, biggest gap first.

Jev’s own answer

Showing 0 of 446