Can ChatGPT Write Your Thank-You Cards? (And Mail Them)

You have sixty gifts, sixty names, and a chat window open. The question is not whether ChatGPT can write a thank-you note — it obviously can. The question is what to ask it for.

Ask for one perfect note and you have made your problem worse. You now have one lovely paragraph and fifty-nine cards to adapt it into by hand. The prompt worth having produces a template — one message that fits every person on the list, with the gift and the name filled in per card.

Below: that prompt, why a template beats sixty generated notes, and the half every guide on this leaves out — actually getting them printed and into mailboxes.

First, where these notes are going

Thirty seconds of context, because it changes what you should ask the model for.

ShipNote turns a gift list into mailed postcards. You paste the list your registry already gives you — Amazon's Thank You List, Babylist's Gift Tracker export, a spreadsheet you kept yourself — and an AI parser pulls out three things per person: their name, the gift they gave, and their address.

That list is the whole trick. It means the two facts that make a thank-you note land — who this is, and what they actually gave you — are sitting in structured form before you write a single word.

Then you pick a card, write one message, add your return address, and pay. Each card is printed, stamped, and mailed for you. Real postcards, not email. They go out the next business day and arrive in 4-7 business days.

So "write my thank-you cards" is really two jobs, and AI handles them very differently:

• Reading the list — who gave what, and where they live. Already automated; there's nothing for you to prompt.
• Writing the words — the part you opened ChatGPT for.

The second job is what this page is about, and it has an unusual shape. You are not writing sixty notes. You are not writing one, either. You are writing one message that becomes sixty.

What prompt actually works for a batch?

The instinct is to feed it one giver at a time. Don't. You will be there all night, and the notes will drift in tone as the model re-improvises the same sentiment sixty different ways.

Ask for the reusable thing instead:

Write a warm, specific thank-you note template for wedding gifts, about 40 words, that will be printed on a postcard. Use [RECIPIENT] where the person's name goes and [GIFT_NAME] where the gift goes. [GIFT_NAME] will print directly after the word "the", so it has to read naturally as a plain noun phrase. Give me three versions: one warm, one short and modern, one a little more formal.

Three versions matters more than it sounds. Your grandmother and your college roommate should not get the same register, and the cheapest way to handle that is two or three templates, not sixty rewrites.

Note the constraint about the word "the". The gift name comes out of your registry list as a bare noun — "cast iron Dutch oven", "stand mixer" — so the sentence has to be built to receive it. Templates that say "thank you for your gift of [GIFT_NAME]" read fine; ones that say "thank you for [GIFT_NAME]" produce "thank you for stand mixer".

Why does the template have those brackets in it?

Because the batch already knows things your message doesn't.

It knows Aunt Ruth gave the Dutch oven and that Priya gave the stand mixer — that came out of your registry list. What it can't know is what you want to say about it. So you write the message once and mark the two spots where every card should differ:

Dear [RECIPIENT],

Thank you so much for the [GIFT_NAME] you got us for our wedding. We truly appreciate your generosity and are grateful to have you in our lives.

Best regards, Alex & Sam

[RECIPIENT] becomes that card's name. [GIFT_NAME] becomes that card's gift. Sixty cards render, each one naming a specific person and the specific object they chose — the two things etiquette actually asks for — and you typed the message a single time.

That is the whole reason to want a template out of ChatGPT rather than sixty finished notes. Sixty notes have to be pasted in sixty times, and at that point the model saved you nothing. One template gets pasted once and picks up the list's own data on the way through.

The two chips above the editor insert the tokens for you, and the counter underneath tells you whether the message still fits on the back of a postcard — the one constraint a chat window has no idea about.

It isn't a mail merge in the grim sense, either — the preview fills the tokens live as you type, so what you're reading is a real card with a real name on it rather than a form with holes in it.

How do you make individual cards your own?

This is the part that makes the whole approach worth defending, and it is where a generated note in a chat window can't follow you.

On the review step you page through the batch and see every card as it will print. Type directly on the back of any card and that one changes — nothing else in the batch moves. Your aunt's card gets the line about the Dutch oven and the short ribs you made in it. Your grandmother's gets three sentences you wrote yourself, from scratch, ignoring the template entirely. The coworker who sent a gift card gets the template as-is, because that is genuinely the right amount of effort for that relationship.

Card 1 of 3 here, but it works the same at card 1 of 80: the template has already filled in Ruth's name and the Dutch oven, and "Tap to edit" is where you take that card somewhere the template could never have gone.

Change your mind on one? There's a reset that puts that single card back to the filled template. So you can rewrite a card, decide the original was better, and undo it without touching the other fifty-nine.

Why this is the right shape for the job:

• You are never staring at a blank card. The failure mode of thank-you notes has never been "the wording wasn't good enough." It's the stack that sits on the counter for four months. Starting from a filled template means the worst possible outcome is a perfectly good card that names the giver and the gift.
• Your effort goes where it changes something. Writing sixty notes by hand means every one gets the same tired 20% of your attention by card forty. Editing fifteen means those fifteen get all of it, and the other forty-five are still specific and still on time.
• The specificity is automatic. The gift name is pulled from your list, not from your memory. That is the detail that makes a note land, and it is exactly the detail people get wrong when they're writing from a half-remembered pile of wrapping paper.
• Consistency without sameness. One voice across the whole batch, because it's one message — but a different gift and a different name on every card, and real handwritten-by-you paragraphs on the ones that warrant it.
• The edits are reversible. You can experiment on a card at 11pm and put it back.

What does AI still get wrong here?

Three things worth watching, because the model will do all of them confidently.

The one that actually stings is the first. A template should carry only what is true of everyone — gratitude, the gift's name, a warm close. Anything specific belongs in the cards you edit by hand, where you know it's true.

Step by step

1. Get your list. Paste the registry export; names, gifts, and addresses come out of it.
2. Ask for templates, not notes — the prompt above, three tone variants, ~40 words each.
3. Pick one and paste it in, keeping [RECIPIENT] and [GIFT_NAME] where they belong.
4. Read the preview. It fills the tokens live, so you're reading a real card.
5. Go through the batch and edit the ones that deserve more. Fifteen of sixty is a normal answer.
6. Send. Cards print and go out the next business day, and land in 4-7 business days.

Related reading

• How to print and mail an AI-generated card — the front of the card, if you're generating the artwork too.
• Do thank-you cards have to be handwritten? — the etiquette question underneath all of this, answered honestly.
• What to write in a wedding thank-you note — wording that works, if you'd rather start from examples than from a prompt.