Which tool to choose
For these short, tone-driven texts, ChatGPT, Claude and Gemini are all fine: the difference is small. If the message is delicate and you care about getting the words just right, Claude tends to handle the emotional register well without overdoing it. Pick one and work on it over several passes, don't settle for the first version.
The interfaces change often: to write a message all you need is the normal chat, no special feature to look for.
How to do it
The machine gives you the structure and unblocks you from the blank page. You add the sincerity, and that's the part that matters.
- Settle three things: who the person is to you, the occasion, and the tone.
- Find a concrete detail to slip into the message: it's what saves it from the generic. The working syntax:
Write me a birthday message for my friend Marco, who's turning 40. We've known each other for twenty years, I want an affectionate tone but with a joke. Detail to mention: our Thursday five-a-side football matches. Keep it short, four or five lines, and no clichés from a greeting card.
- Ask for two or three versions with different tones and take the best of each.
- Read the result out loud: if it's not something you'd actually say, rewrite it until it sounds like you.
A concrete example
Anna had to thank a colleague who had covered for her during a difficult week, and didn't know how to say it without seeming formal. She explained the situation to ChatGPT and added a detail: the colleague had stayed beyond hours to finish a job for her. The AI gave her three versions. Anna took the opening of one, the closing of another and changed one sentence with her own words, mentioning the evening they'd stayed late. The message was short, specific, true. The colleague replied, moved.
When it does NOT work (and how to fix it)
If the message sounds fake, like a greeting card
Clichés are the sign that the personal is missing. Fix: remove every sentence that would fit anyone and put in one that's true only for that person, a memory or a detail. One specific line beats ten elegant, empty ones.
If it's too long and rhetorical
The AI tends to inflate emotional texts. Fix: explicitly ask for four or five lines, simple words, no big words. The short measure almost always moves people more than pompous prose.
If it's condolences and they sound perfunctory
Here the AI must be used with caution: for grief it's scaffolding, not your voice. Fix: have it give you an outline, then write the heart of the message yourself, with a concrete memory of the person who's gone and honest words. If the one who's grieving is close to you, two sincere, imperfect lines are worth more than a perfect text you don't feel.
A tip from someone who really uses it
Always ask for several versions with different tones and then take them apart and reassemble: the opening of one, the joke of another, your own closing. The best message is never one whole draft, it's the edit you make yourself. And the ultimate test is reading it out loud: if you feel like you're reciting, it's not yours yet.
Frequently asked questions
Can it be done with the free version?
Yes. Writing greetings, dedications and thank-yous is a text task that the free versions handle without practical limits. For a single message you won't come near any usage cap.
Can people tell an AI wrote it?
If you leave the draft as is, often yes: the generic tone gives it away. If you put in one true detail and rewrite it in your own words, no, because at that point the content is yours and the AI only gave the structure. The personalization work is exactly what erases the fingerprint.
Is it dishonest to use AI for condolences?
It's not the medium that measures sincerity, it's what you put into it. Using it as an outline to find the words when grief blocks you is legitimate. It becomes empty if you delegate everything to it and send a text you don't feel: someone in grief senses the difference between a warm message and a correct one. For the people you truly love, let the AI unblock you, but let the final voice be yours.