Apple introduced a new feature in iOS 18.4 that primarily affects the App Store. Ratings of apps and games are now automatically aggregated to give users a quicker overview. This feature is powered by a multi-stage AI system, which Apple explains in a post on its own Machine Learning Research Blog. If you want to know how Apple solves this technically and what it focuses on, you'll find the most important information at a glance here.
The App Store often contains hundreds or thousands of reviews per app. The new feature in iOS 18.4 is designed to help users understand these reviews more quickly. Apple uses large language models (LLMs) to transform the many reviews into short, concise summaries. Apple says these summaries should be fair, helpful, truthful, and safe. To achieve this, Apple has developed a multi-step process.
How the system works
The system works in several steps. First, reviews that violate guidelines – such as spam, obscene language, or fraudulent content – are removed. The LLM ignores these completely. The remaining reviews then run through several modules specially trained by Apple. Each module has a specific task: It extracts the most important statements, recognizes recurring themes, weighs positive and negative points, and creates a short summary. This summary is usually between 100 and 300 characters long. The goal is to create an overview that reflects the user's overall impression as accurately as possible – regardless of whether it concerns design, performance, or new features.
Dynamic adaptation to new ratings
An important point to note is that reviews in the App Store are constantly changing—for example, following updates, new features, or bug fixes. Therefore, the summaries must be flexible and updated regularly. The system is designed to automatically detect these changes and adjust the summaries accordingly. According to Apple, the LLM can also effectively process reviews that are very long or very short. The system also filters out off-topic comments or "noise"—i.e., irrelevant content.
Trained and checked
To ensure the summaries are truly useful, they were evaluated by human reviewers during the development phase. They assessed criteria such as usefulness, understandability, and security. Apple used this feedback to further improve the models. Each step in the process is based on a specially trained LLM. Apple emphasizes that the summaries are not simply generated automatically; each individual module has been specifically trained to accurately and balancedly reflect user opinions (via Machine Learning Research blog).
Apple relies on AI for more overview in the App Store
With iOS 18.4, Apple is bringing a new feature to the App Store powered by artificial intelligence that automatically summarizes reviews. This saves time and quickly provides you with an overview of other people's opinions. Behind this feature is a complex, multi-level system of different language models that filters, analyzes, and dynamically adapts content. Apple places particular emphasis on security, fairness, and accuracy. Currently, these summaries are only available in English and for a limited selection of apps. Over the course of the year, however, the feature will be expanded to include additional languages and available for all apps with a sufficiently high number of reviews. (Image: Shutterstock / miss.cabul)
- Apple plans to build US iPhones only in India by the end of 2026
- Reboot for Siri: Apple cleans up its voice assistant
