Model Overview
Processing Data
Only publicly available data, like app store reviews, and manually tagged data is processed by our AI models. These tags allow us to categorize and analyze feedback more effectively. Importantly, we use only the title and content of reviews for AI purposes.
Safeguards and Data Usage
All processing and analysis are conducted within the customer's AppFollow account, without external data sources. Manually tagged reviews are used as input to ensure that our models perform accurately and stay within their intended function. Automatic tagging is applied based on these inputs to provide a consistent categorization of user feedback.
Accuracy and Quality Assurance
The AI models are trained on manually tagged reviews to maintain high accuracy and quality. We use a feedback loop where incorrect tags can be reported, allowing for continuous improvement. Our QA processes include a trial period to validate tagging accuracy and coverage, precision, and recall metrics to monitor performance.
Handling Personal Data
We process two types of personal data:
- Authorized Users of AppFollow Platform: This includes contact, usage, location, technical and transactional data, processed per user rights such as access, deletion, and correction.
- Public App Store Reviewers: For users who post reviews, we process name/nickname, review content, location, language, and device details, based on platform-specific data-sharing capabilities.
For a detailed overview of our data handling practices and privacy commitments, visit our privacy policy: AppFollow Privacy Policy.
Transparency and the EU AI Act
AppFollow has text watermarking turned on for every supported OpenAI model that generates AI replies. The setting applies to the whole organization, so it covers every AppFollow workspace and plan. You don’t need to do anything.
What the watermark is
The watermark is an invisible statistical signal built into the words the model chooses. It adds no hidden characters, doesn’t change the reply’s wording or formatting, and needs no action from your team.
Why we use it
Article 50(2) of the EU AI Act requires providers of generative AI systems to mark AI-generated text in a machine-readable way, so it can be detected as artificially generated. For systems already on the market before 2 August 2026, this applies from 2 December 2026. Watermarking is part of how AppFollow meets this requirement.
Limitations
- Short replies may carry no detectable watermark. EU guidance exempts very short text, and many review replies are that short. Google Play replies are limited to 350 characters.
- If you edit or translate an AI reply before publishing, the signal gets weaker or disappears.
- Replies your team writes without AI are not marked.
- OpenAI’s watermark detector isn’t public yet, so neither AppFollow nor its customers can check individual replies.
Your obligations
Watermarking covers the marking requirement for AI providers. Other AI Act obligations may apply to how you publish replies, depending on your use case. Your legal team can confirm this.