MyHeritage Scribe AI Alternative for Historical Documents

MyHeritage Scribe AI names no transcription model, and it does not let you correct the transcription it produces. If a surname is misread, the mistake stays in the transcription and carries into the translation and summary built on it. Leo is the alternative for anyone who needs a faithful transcription of historical documents: its own model, ATR-1, fine-tuned specifically for historical material, an editable transcription beside every image, and a workspace built for serious research. Leo also lets you choose from dozens of models for each part of the job and compare their readings side by side.

This guide sets out what Scribe AI is, which model reads your documents, the limits reviewers have reported, and why historians and genealogists choose Leo instead.

What MyHeritage Scribe AI is

Scribe AI is a feature of MyHeritage, the family history platform, launched at RootsTech on 5 March 2026. According to MyHeritage's launch post, you upload an image of a document, photo, gravestone or coat of arms, and it returns a report with key findings, a transcription, a translation into your interface language, extracted names, dates and places, historical context and suggested next steps. It also runs on record images in MyHeritage's database and, since April 2026, in the MyHeritage mobile app.

The report is the product. The transcription is one section of it.

Which model reads your documents?

The first question to ask any transcription tool is which model reads your documents, and whether it was built for historical material.

For Scribe AI, MyHeritage's launch post says it is "powered by artificial intelligence models trained to read handwriting and printed text in many languages". MyHeritage's privacy policy (last updated 1 June 2026) says that materials uploaded to Scribe AI "will be processed by Google". Neither names the model.

With Leo, the answer is public. Leo's transcriptions come from its own model, ATR-1, fine-tuned specifically for historical documents. You know what is reading your page, and what it was built to do.

You cannot correct a Scribe AI transcription

For anyone who works with historical documents, this is the limit that matters most in daily use. Scribe AI does not let users correct the transcription in the tool.

Nicole Elder Dyer of FamilyLocket pointed out the consequence in March 2026: errors in the transcription carry into the translation and summary that Scribe AI builds from it. A misread name becomes a mistranslated name and then a wrong fact in the key findings. MyHeritage's own advice, quoted by Family Tree Magazine, is to copy the result into a text editor and make the corrections there. Once you do, the corrected text is outside MyHeritage, separated from the image, and the translation and summary still reflect the original mistakes.

Errors reviewers found

Every machine transcription needs checking, and Scribe AI's reviewers found plenty to check. Because the text cannot be corrected in place, each of these has to be fixed somewhere else.

Surnames and dates. Testing parish registers, Wayne Shepheard found that almost all the errors were surnames, with dates the other main problem, and that Roman numerals gave Scribe AI trouble.
Occupations. Diane Henriks found an ancestor's surname wrong and his father's occupation given as master mason instead of master cartwright.
Crossings-out that are not there. Family Tree Magazine's Katharine Andrew found that Scribe AI reported details of a German death record as crossed out when nothing on the record was, and carried those notes into its translation. Lisa Louise Cooke of Genealogy Gems saw a line misread as a cross-out.

Why historians and genealogists choose Leo

Leo is built around the transcription itself: getting it right, keeping it beside the image, and giving you the tools to work with it.

One model, or the best model for each job

Leo is not one model. It is a workspace for managing transcription, where you choose the right model for each part of the job and compare the results.

One menu, dozens of models. The model library in the transcription window holds Leo's own ATR-1, the recommended default for historical documents, alongside models from OpenAI, Google and Anthropic and the cloud OCR engines Google Cloud Vision, Azure AI Read and AWS Textract.
The right model for each part. Use ATR-1 for historical handwriting and print. For clean, legible modern print, such as a typescript or a modern book, a small model or a cloud OCR engine costs a fraction of a credit an image, from 0.02 credits. Prices are per image, by model.
Compare readings. Each model's reading goes into its own tab on the same image, beside the original, and nothing is overwritten. The Interpolate Transformation reconciles two readings into one.

A highly accurate model, fine-tuned for historical documents

ATR-1 is a highly accurate model fine-tuned specifically for historical documents. It is zero-shot, so there is nothing to train: upload an image and transcribe. It reads the image at high resolution, and it is trained to transcribe what is on the page rather than normalise it, so it never silently modernises. An ancestor's surname spelled three ways, or a place name written as the clerk heard it, is evidence, and ATR-1 keeps it. Our guide to historical language normalization explains why the faithful text should come first.

Built-in failure detection hides output that looks wrong, retries it, and refunds the credit if it cannot succeed. Every correction you make feeds back into training, so later readings improve.

You can correct everything

In Leo, the transcription sits in an editor beside the image. You check a name against the writing and fix it in place, and the corrected transcription is the one you search, translate and export. That side-by-side editor is why Nicole Elder Dyer, in the same FamilyLocket roundup, named Kindex and Leo as her favourite tools: "I love the ability to check the original image and correct the text side-by-side."

Full flexibility for advanced workflows

Keep every version. An image can hold several transcription tabs, and nothing is overwritten unless you choose.
Transform without losing the original. Correct, Modernize, Interpolate (which reconciles two readings), Translate and Custom each write a new transcription tab. Summarize, Classify, Generate glossary, Extract named entities and Custom write annotation tabs. Prompts are editable, and the base transcription is never touched. Transcription and translation are two jobs, and Leo keeps them apart.
Search everything. Search runs across all your documents, transcriptions and annotations at once, with fuzzy matching for variant spellings of a name.
Organise. Nested folders and metadata on every document keep a family archive in order.
Work in batches. Transcribe or transform every image in a document in one go.
Export and share. Export to PDF, Word, HTML or TEI (XML), or share a read-only link that needs no account.
Automate. The Leo API and MCP server, now in beta, let your own scripts and AI agents work on your documents.

Leo has a free plan with 10 credits each month, and 1 credit transcribes 1 image.

MyHeritage Scribe AI vs Leo

Checked in October 2026.

The model: MyHeritage names no model for Scribe AI, and its privacy policy says uploads are "processed by Google". Leo runs ATR-1, its own model, fine-tuned specifically for historical documents.
Choice of model: Scribe AI gives you its own analysis. Leo offers ATR-1 plus dozens of OpenAI, Google and Anthropic models and cloud OCR engines, each run into its own tab to compare.
Correcting the transcription: Scribe AI does not let you correct it in the tool. Leo puts an editable transcription beside every image.
Errors downstream: in Scribe AI, transcription errors carry into the translation and summary. In Leo, translations and summaries run from the transcription you have corrected, each in its own tab.
Errors reviewers reported: reviewers found errors in Scribe AI on surnames, dates, Roman numerals and occupations, and crossings-out that were not there. In Leo, you fix any error in place, beside the image.
Getting text out: Scribe AI offers copy buttons and share links. Leo exports PDF, Word, HTML and TEI and offers read-only share links.
Free use: Scribe AI is free on a limited number of images, then needs a Complete or Omni plan (see MyHeritage's pricing page). Leo's free plan gives 10 credits every month, and 1 credit transcribes 1 image.

For the wider field, see our comparison of handwriting transcription software.

Whatever tool you use, check names, dates, places and relationships against the image. Our guides to verifying transcription accuracy and genealogy record transcription set out the method. In Leo, the fixes you make stay with the image.

Frequently Asked Questions

What AI model does MyHeritage Scribe AI use?

MyHeritage does not name one. Its launch post says Scribe AI is powered by AI models trained to read handwriting and printed text in many languages, and its privacy policy says uploads to Scribe AI are processed by Google. Leo's transcriptions come from its own model, ATR-1, fine-tuned specifically for historical documents.

Can I correct Scribe AI's transcription?

No. Scribe AI does not let you correct the transcription in the tool, and MyHeritage's own advice is to copy the text into a text editor and fix it there. Errors also carry into the translation and summary built on it. In Leo, the transcription sits in an editor beside the image, so you correct it in place.

What mistakes do reviewers report in Scribe AI?

Reviewers in 2026 reported errors on surnames, dates, Roman numerals and occupations, and text described as crossed out when it was not. Because Scribe AI does not let you correct the transcription, each fix has to be made outside the tool. In Leo, you correct the transcription beside the image, and ATR-1 is trained to transcribe what is on the page rather than normalise it.

Can I compare different AI models in Leo?

Yes. Leo's model library puts its own ATR-1, the default for historical documents, in one menu with dozens of models from OpenAI, Google and Anthropic and cloud OCR engines. Each reading goes into its own tab on the same image, so you can compare them against the original, and the Interpolate Transformation reconciles two readings. For clean, legible modern print, a small model or OCR engine costs from 0.02 credits an image.

Is MyHeritage Scribe AI free?

Only for a limited number of images. After that it needs a Complete or Omni subscription. Leo has a free plan with 10 credits every month, and 1 credit transcribes 1 image.

What is the best alternative to MyHeritage Scribe AI for historical documents?

Leo. Its own model, ATR-1, is highly accurate and fine-tuned specifically for historical documents, needs no training, and transcribes what is on the page without silently modernising it. The transcription sits beside the image where you can correct it, and Leo adds a model library of dozens of models to compare in tabs, Transformations into separate tabs, search across everything, batch work, export and an API and MCP server in beta.