Google Play

Google Play recommends devs document their gen AI app testing

Abner Li | Jun 6 2024 - 9:00 am PT

Google Play today is “providing guidance to help developers enhance the quality and safety of AI-powered apps” as adoption of those technologies increase, with increased testing being top of mind.

Google starts by advising developers to review its AI-Generated Content Policy:

“Prohibit and prevent the generation of Restricted Content”
“Give users a way to report or flag offensive content”: This was first announced in October

Meanwhile, developers have to make sure these gen AI apps are prompted responsibly. Google will remove an application from Play if it’s advertised “for an inappropriate use case.”

…review your marketing material to ensure that your ads accurately represent your app’s capabilities, and that all ads and promotional content associated with your app, across all platforms, meet our App Promotion requirements.

Finally, Google is encouraging rigorous testing, including having safeguards in place for “prompts that could manipulate your generative AI feature to create harmful or offensive content.” Notably, Google recommends developers “start documenting this testing because we may ask to review it in the future to help us better understand how you keep your users protected.”

You are accountable for the experience in your apps, so it’s critical for you to understand the underlying AI tools and models used to create media and to ensure that these tools are reliable and that the outputs are aligned with Google Play’s policies and respect user safety and privacy.

In the future, Google is adding “new app onboarding capabilities” to make “submitting a generative AI app to Play even more transparent and streamlined.”

Google today also talked about how it’s using large language models “to address new complexities that come with apps with generative AI features.”

We’re using new technology like large language models (LLMs) to quickly analyze app submissions, including vast amounts of text to identify potential issues like sexual content or hate speech, and flag them for people on our global team to take a closer look. This combination of human expertise and increased AI efficiency helps us improve the app review experience for developers and create a safer app environment for everyone.