How to Get AI to Read Handwritten Letters and Family Recipes

Everyday life Tutorial7 min read·Updated October 8, 2026
The short answer

Yes, an AI chatbot can read many handwritten pages. Photograph the page straight on in good light, upload it to ChatGPT or Gemini, and ask it to transcribe the text and mark any doubtful word with [?]. Treat the result as a draft: check names, dates and recipe amounts against the original yourself, because the AI can sound sure and still be wrong.

A box of recipe cards in your grandmother's writing. A bundle of letters in cursive you can barely read. An AI chatbot on your phone can look at a photo of the page and type out what it says. It won't always be right, though, so this guide shows both how to do it and how to catch the mistakes.

You can use ChatGPT or Gemini. Both are free to start with, and both let you attach a photo to a chat. If you haven't used one yet, see how to use ChatGPT first.

Photograph the page properly

Most bad readings start with a bad photo. Before you press the button:

  • Use even light, near a window or under a lamp, and avoid shadows from your hand or phone.
  • Hold the phone straight above the page, not at an angle, so the lines stay straight.
  • Keep it steady until the writing is sharp, then take the photo.
  • For a long letter or a page with writing in the margins, take one photo per section. Photograph hard-to-read passages separately.
  • Crop out what you don't need, such as a table or other papers, using your phone's photo editor.

Upload it and use a clear prompt

Open ChatGPT or Gemini, tap the button for adding a photo or file (look for a + or a paperclip), and choose your picture. Then type a request. Tell the AI what the page is, and ask it to be honest about what it can't read. You can copy this:

This is a photo of a handwritten family recipe. Please transcribe it exactly as written. If you can't read a word or a number, don't guess: write ? and give me your best alternative readings. Don't add anything that isn't on the page.

For a letter, change the first sentence: "This is a handwritten letter, probably from the 1950s, written in English." Giving the type of page, the rough period and the language helps the AI read unusual letter shapes and old phrases.

Check the result against the original

Read the AI's text next to the photo, line by line. Pay special attention to:

  • Names and places. In one test, ChatGPT read a messy handwritten name as a completely different person's name.
  • Dates and numbers. A 1 and a 7, or a 3 and an 8, are easy to confuse.
  • Recipe amounts and units. Teaspoon or tablespoon? Half or one and a half? Oven temperature? Check each one against the card.
  • Words the AI "fixed". Chatbots like to make a sentence sound natural, which may mean smoothing over what was really written.

Anything marked ? is where the AI is telling you it's unsure. Look at those words yourself, or ask a relative who knew the writer.

Ask again, or get a second opinion

If the first reading has gaps, you can improve it:

  • Tell the AI what you've worked out. For example: "This writer makes the letter 'u' with a long tail, and 'ss' looks like 'fs'." Then ask it to read the page again.
  • Start small. Crop one tricky line, paste it in, and ask for just that line.
  • Give the same photo to a second AI assistant and compare. Where the two readings differ, the page is probably hard to read and deserves a closer look from you.

What AI does well, and where it struggles

Clear, fairly modern handwriting, such as a recipe card, a postcard or a shopping list, usually goes well. Take the photo the right way up, though: OpenAI warns that its image feature may misread rotated or upside-down text.

Old or faded writing, unusual spelling, old abbreviations and local place names are much harder. People who use these tools for family history report that the AI can guess, skip words or produce something that sounds convincing and is wrong. OpenAI also says its image feature may be less accurate when an image is unclear. So treat the AI's text as a starting point, never as the final word. For an old document that matters, such as a will or a record you want to cite, the original page and a person who knows the period come first.

Privacy: a letter is not just text

Old letters often contain full names, addresses, health details and family secrets, some about people who can't agree to you sharing them. Before you upload:

  • Crop out addresses and anything you don't need read.
  • Check whether your chatbot is set to use your chats to improve the AI. In ChatGPT this is the setting called "Improve the model for everyone" under Settings, then Data controls. In Gemini, look at your Gemini activity settings.
  • Don't upload documents with ID numbers, bank details or passwords. See what not to tell an AI chatbot.

What to try next

If your handwritten pages are in another language, translating a photo or document with your phone is the next step. For tracing a family tree from old records, read AI for family history. And if you're wondering how far to trust any of this, see can you trust ChatGPT.

Sources

Published October 8, 2026 · Updated October 8, 2026How we test →

Frequently asked questions

Can ChatGPT or Gemini read cursive handwriting?
Often, yes, especially neat, modern handwriting such as a recipe card. Old, faded or very messy writing is much less reliable. Reports from people who use these tools for old documents say the AI can guess, skip words, or produce something that looks convincing but is wrong.
How do I know if the AI is making things up?
Ask it to mark any word it is unsure of with [?] and to give alternative readings. Then compare the result with the original page yourself, line by line. You can also give the same photo to a second AI assistant and look at where the two readings differ.
What is the best way to photograph a handwritten page?
Use even light without shadows, hold the phone straight above the page rather than at an angle, and keep it steady so the writing is sharp. For a long or difficult page, photograph it in sections and take difficult passages separately.
Is it safe to upload a private family letter?
Treat it like anything else you share with a chatbot. A letter may contain names, addresses and health details. Crop out what you don't need, and check your privacy settings about whether your chats can be used to train the AI.
Can I trust the amounts in a transcribed recipe?
Check them against the original every time. A misread number or unit, such as a teaspoon turning into a tablespoon, can ruin a dish, and the AI will not warn you about it unless you ask it to flag doubtful parts.
Radim S.
Founder & editor

Radim is a software developer who spends his days building with AI and his evenings explaining it to family members who don’t care how it works — only what it can do for them. The safety guides are checked claim by claim against primary sources before they go out.