Google's new Simplify feature rewrites dense text without losing key details

Google wants you to actually understand those dense medical studies, legal disclaimers, and financial filings.

Google Research unveiled a new system powered by Gemini models focused on minimally-lossy text simplification. Unlike summarization, which deliberately drops details to save time, or explanation, which often adds new context, simplification rewrites complex language into plain English while keeping every nuance, technical term, and relationship intact.

The feature is already rolling out inside the Google app for iOS under the name Simplify.

824 rounds of AI coaching AI

Writing instructions that make a model simplify dense jargon without changing the meaning is notoriously hard. Google solved this by removing humans from the prompt-tuning loop entirely.

They built an automated feedback system using two different Gemini models. Gemini 1.5 Flash handled the actual simplification rewriting, while a more capable Gemini 1.5 Pro evaluated the results against automated scores for readability and fidelity.

Pro analyzed Flash's missteps, updated the system prompt, and pushed Flash to try again. This self-refinement cycle ran for 824 iterations before hitting a performance plateau.

A 15% leap in medical reading comprehension

To see if this actually helped humans, Google ran a massive randomized controlled study covering nearly 50,000 multiple-choice answers across finance, law, technical fields, and medicine.

Overall, participants reading the simplified text scored 4% higher in accuracy. The biggest jump came from medical research papers on PubMed, where accuracy surged by 15% overall, and by as much as 38% on a single difficult passage.

Comprehension rose 6% in finance, 4% in legal documents, and 4% in computer science and aeronautics. Crucially, the gains held even when people were quizzed without being able to look back at the text, and participants reported significantly lower mental effort.

In the iOS Google app, users can now highlight any block of text on a web page, tap the new Simplify button, and read an inline rewrite without jumping to another tab or losing their place.

Why it matters

We think this is one of the smartest, most practical applications of LLMs for everyday web browsing.

For years, reading online has forced a bad trade-off: skim a summary and miss the critical caveats, or struggle through dense legalese and medical jargon you barely understand. Minimally-lossy simplification bridges that gap without cutting corners. If you are an iPhone user trying to parse insurance policies, medical studies, or technical documentation, it is well worth using right now.

What Google did behind the scenes is just as notable. Running an 824-iteration prompt refinement loop where a larger model coaches a smaller model shows how automated prompt engineering is replacing tedious manual tuning. Gemini 1.5 Flash is cheap and fast, making this inline feature realistic to deploy at scale.

The obvious catch lies in edge cases. As Google's researchers themselves noted, LLMs can still hallucinate errors, and simpler text can easily create a false sense of mastery. When users are reading about health conditions or legal liability, feeling confident about a slightly hallucinated simplification is arguably more dangerous than knowing you did not understand the original.