How Ancestry Uses LLMs for Record Digitization and Customer Tools

TL;DR

Ancestry has collected over 71 billion records from 88 countries, building 148 million family trees over 42 years.
Digitizing these records was historically a slow and expensive manual process.
Investments in AI and machine learning, particularly since 2017 and accelerating after a 2020 acquisition, have sped up digitization.
AI advancements have enabled new user tools, including facial recognition and handwriting recognition.
Early AI projects focused on proprietary machine learning models and computer vision systems starting in 2014.
The company incorporated BERT-based models by 2019 for quicker data extraction, especially for obituaries.
The advent of ChatGPT and other large language models (LLMs) further accelerated digitization of unstructured data.
Ancestry now uses a blend of proprietary and open-source LLMs, processing nearly 200 languages with minimal training.
User-facing AI features include Face Match (facial recognition, July 2024) and a document transcription feature for handwritten notes (April 2025).
AI Stories, launched in December 2025, provides narrated audio stories of ancestors.
By the end of 2025, over 50% of Ancestry's historical records published were AI-generated.
AI has tripled the rate of content growth, with billions of new records added annually.
External AI use cases, like language translation for document transcription, were added in June 2026.

Continue reading the original article