Most software companies that use AI write a one-paragraph notice on the privacy page and call it disclosure. We think this product deserves a longer answer. People send cards about important moments. The question of who actually wrote what is fair, and answering it well is part of the product.
Here is the long version.
What the AI is
The AI is a drafting partner. When you sit down to write a birthday card for your sister, Thoughtful uses what you've told it about her — the relationship, the inside jokes, the memories you've logged — and produces a starting draft. The draft tries to sound like you, based on cards you've written in the past, your preferred tone, and the way you write to people in similar relationships.
The draft is a ladder. You climb up, then you kick it away. The editor surfaces a small prompt every time you write a card: add at least one specific memory or detail that only you would mention. That sentence is what turns the draft into a card.
What the AI is not
The AI is not the author. It does not finish your cards for you. It does not send anything you haven't approved. The current generation of large language models can produce fluent prose at scale, and we deliberately do not let the product run that prose straight to the recipient. The point of a card is the person on both ends of it. We won't let the model be either.
How we use your data
We do not train any AI model on your private cards, drafts, or notes. The data you put into Thoughtful — the people you've added, the cards you've written, the relationship details — stays in your account and is used only to make the app work for you.
We use foundation models from Anthropic to generate drafts. The inputs to those models are the context for a single card, sent for a single generation, and not retained for training by either us or Anthropic. We pay for that distinction; it's part of what the subscription funds.
We do not sell your data to advertisers. We do not have an ad product. We do not have a social feed.
How the AI learns your voice
We learn the way you write from the cards you write inside the app. Word choice, sentence length, sign-off style, formality. The more cards you've written, the better the drafts get at sounding like you. None of this leaves your account.
You can also teach the AI explicitly: paste a few examples of messages you've written in the past, and we'll use them as a reference for tone. This is optional. Many people don't bother; the system catches up after a handful of cards.
The hard occasions
For sympathy and other grief-adjacent occasions, the app behaves a little differently. The drafts are shorter. The editor is more insistent that you add a memory of the person who died. We default to suggesting you write rather than generate, and the prompts in the app reflect that.
Grief is the highest-stakes occasion to send a card for, and the one most people freeze on. The job of the AI is to help you start. The job of finishing it belongs to you.
Why we wrote this page
Because the question of whether AI belongs in the act of writing something heartfelt is a real question, and 'we use AI' isn't an answer. We think the right way to use AI for this is as a draft assistant that gets out of the way the moment you have something to say. Everything we build is designed around that.
If you have questions or concerns about how we use your data — or want to suggest a way the app could be better — write to us at hello@thoughtful.app. We answer.