We get a version of this question every week. Someone has just heard that a coworker's father died, or a college friend's mother, and they sit down to write the card. They open the box of stationery they bought three years ago. They write 'Dear—' and then nothing. After fifteen minutes of nothing, they wonder if it would be wrong to use AI.
It's a fair question. It deserves a fair answer.
The short version
Yes — if you use the AI to start, edit it heavily, and never let it ship without your name and your specific memory in it. No — if you use it to finish, paste the draft, and send it as is.
The line is whether the card is yours when it arrives.
Why people freeze on sympathy cards
Sympathy is the hardest occasion. There are two reasons.
The first is that you are afraid of saying the wrong thing. Grief is sensitive territory. There is a long list of things you are not supposed to write — 'they're in a better place,' 'everything happens for a reason,' 'I know how you feel' — and most people have absorbed enough warnings to be paralyzed about what's left.
The second is that the stakes feel high. The person reading it is already breaking. You don't want to be the bad sentence. So you sit with the pen. You don't write.
Where AI helps
AI helps with the freeze. It is good at giving you a structure to react to. You read a draft and your mind says not that, but this — and now you are writing.
That's the only thing it's for. The draft is a ladder. You climb up, then you kick the ladder away.
Where AI gets in the way
A model can produce a fluent sympathy paragraph. It can produce a hundred of them. None of them mention the person who died. None of them mention what you remember about that person. None of them mention you.
If you send the unedited model output, you are sending a generic condolence with your signature on it. The recipient will know. They may not say it. But the card will not land in the way you wanted it to.
The rule
Here is the rule we use, and it works for sympathy and for every other hard occasion:
For sympathy cards, that one specific thing is almost always a sentence about the person who died. Even if you didn't know them well — 'Your mother sounded like a wonderful person; I always loved the way you talked about her' works. The grieving person will read that line forty times. The model didn't give it to you. You did.
How we built Thoughtful around this
We built Thoughtful's AI to draft, not to finish. When you write a sympathy card in the app, the draft is a starting point, and the editor nudges you to add a memory of the person and to sign off in your voice. We won't let you send a card that is only model output — there is always a moment where you have to add the thing only you would say.
It is not perfect. It is honest. We think AI helps people send more cards, more often, to more of the people they love. We also think the moment of personal voice is non-negotiable. So we designed for both.
If you'd like to try it, Thoughtful is on the App Store. The first cards are free, and sympathy is one of the occasions we most want to make easier.