SUMM AI — A Munich Start-up's Generative AI Brings Leichte Sprache to German Schools and Vocational Colleges

Up to 20 million people in Germany struggle to understand standard-register text because of learning disabilities, intellectual disabilities, or limited German proficiency. Leichte Sprache is a tightly controlled, simplified register — short sentences, one idea per sentence, no unexplained jargon — designed to make information genuinely accessible to this group. Munich-based start-up SUMM AI built a generative-AI tool that translates any text into Leichte Sprache or the slightly less restrictive Einfache Sprache at the push of a button.

More than 100 companies, public authorities and welfare organisations now use it, including a 2023 case where Bavaria's State Ministry of the Interior used it to publish state-election information in Leichte Sprache. In education specifically, schools, vocational schools and further-education providers use it to adapt exam information, school regulations and teaching materials — a resource the developer's own materials describe as especially relevant in special-needs schools, in vocational training for people with limited German, and in general foreign-language education.

A comparative review by the accessibility-focused outlet Multisprech.org found that SUMM AI's output consistently reaches the B1 reading level under the Common European Framework of Reference, but — like every other AI tool it tested — falls short of the stricter A2 level some readers need. A university bachelor's thesis evaluating the tool found it cuts translation time but that quality varies strongly by text type and length, with 'information pages' and texts over roughly 3,600 characters scoring worst on correctness and rule-conformity. The Leichte-Sprache research unit at the University of Hildesheim co-ran a year-long, publicly funded project (KI-GesKom, backed by Lower Saxony's Ministry for Science and Culture) evaluating SUMM AI for health communication with the publisher Apotheken Umschau, and published a practitioner handbook of lessons learned via Springer.

No education-specific study measuring learning outcomes (rather than text readability) was found, and both independent reviews found real quality limits for longer or more complex documents — the tool visibly still needs a human editor in the loop, which is itself useful, honest evidence about where automation currently helps and where it does not.