Deciphering the Past:
Using AI to Read Old Cursive

You have been staring at it for forty-five minutes. A letter in your great-great-grandmother's hand, written in 1847, dense with the flowing loops and connecting strokes of nineteenth-century script. You can make out individual words here and there — a name, a date, what might be "Ohio" — but the sentences dissolve into a tangle of ink every time you think you have found the thread. Your eyes ache. Your magnifying glass is not helping. And somewhere in those two pages is the information you have been looking for for years.

If this sounds familiar, you are not alone — and you are no longer stuck. A new generation of AI tools has made the transcription of difficult historical handwriting genuinely accessible to anyone with a smartphone and an internet connection. Not perfectly, not without caveats, but well enough to transform an impenetrable document into readable text in the time it takes to make a cup of tea.

This tutorial will explain how these tools work, which ones are best suited for different types of documents, and how to use them step by step — even if you have never used any kind of AI tool before.

The old script is not impenetrable. It is simply unfamiliar. And in 2026, you no longer need to become an expert in nineteenth-century handwriting to read what your ancestors wrote.

✍️ Why Old Handwriting Is So Difficult — and Why AI Can Help

Modern handwriting — the kind most of us were taught in school — is designed for speed and personal expression. Historical handwriting was designed for a different purpose: to be legible to anyone who might need to read it, within a specific cultural and educational tradition. The problem is that those traditions varied enormously by country, region, era, and social class, and they are now foreign to most modern readers.

🧰 The 2026 Toolkit: Which Tool for Which Document

Not all AI transcription tools are created equal — and the right tool depends on the type of document you are working with. The landscape in 2026 offers three main options, each with different strengths.

Accuracy figures are approximate and vary significantly based on document quality, script style, and language. CER = Character Error Rate.

📸 Step One: Getting a Good Image — the Foundation of Everything

The single factor that most determines how well any AI tool performs on a document is the quality of the image you provide. An AI that can achieve 95% accuracy on a clear, high-contrast scan of a clean 1850 letter may produce 60% accuracy on a blurry smartphone photo of the same letter taken in poor light. Before you open any transcription tool, invest five minutes in getting the best possible image.

🔬 Step Two: Using Transkribus — The Complete Guide

Transkribus is the most powerful dedicated handwriting recognition tool available for genealogical documents. It can save you countless hours by automatically transcribing handwritten documents into searchable text, and its free tier — 50 credits per month — is sufficient for processing several pages of documents. Here is how to use it from start to finish.

📋 Seeing It in Action: A Before and After

To give you a concrete sense of what Transkribus produces, here is a typical result from an 1847 letter — the kind of family correspondence that fills genealogical collections and defeats casual readers. The left column shows what the handwriting looks like described as text; the right column shows the AI's transcript after initial processing (before human correction).

Notice two things about this example. First, the AI has correctly resolved the long-s character — the old-fashioned letterform that makes "Sister" look like "Sifter" to a modern reader — in most instances. This is precisely the kind of specialised historical knowledge that Transkribus's training data provides. Second, the AI has flagged two words in brackets as uncertain: "secured" and "Harrison's." These are your review targets — the places where the original image needs a second look.

💬 Step Three: Using Claude or ChatGPT as a Second-Pass Corrector

Once you have a Transkribus transcript, there is a powerful additional step available: uploading both the original document image and the Transkribus transcript to Claude or ChatGPT and asking the AI to compare them and suggest corrections. When AI language models are used to correct the output of HTR software like Transkribus, they can achieve near-human accuracy levels — character error rates as low as 1.8%.

📁 Step Four: What to Do With Your Finished Transcript

A verified transcript is an extraordinary research asset — and one that most genealogists never create for their documents. Here is how to make the most of it.

✦ The Words Are Already There

Your ancestor wrote that letter carefully. They pressed pen to paper and formed each word with deliberate attention, because letters in 1847 were not dashed off — they were composed, and every word was chosen. The meaning has always been there, waiting in the ink. What has changed is that in 2026, you no longer need years of palaeography training to access it.

The AI tools described in this tutorial are not perfect, and the verification step is not optional. But they transform the experience of working with difficult documents from something that requires specialist skills, excellent eyesight, and extraordinary patience into something that almost anyone can do — with a free account, a decent photograph, and twenty minutes.

The letter is waiting. Go read it.