Substack · AI product · Mid · 45 to 60 minutes

How should Substack handle AI transparency?

Substack is a platform where writers publish newsletters and posts directly to readers, often through subscriptions. Answer this product management case and get it graded line by line against a weighted rubric.

product strategy
trust and safety
AI product judgment
stakeholder empathy
measurement

The situation

You are the PM responsible for writer and reader trust at Substack. Generative AI tools are becoming common in writing workflows, from brainstorming to drafting to editing. Readers are starting to ask whether posts were written by a person, shaped with AI, or mostly generated, and some say unclear use could weaken trust in writers and in the platform. At the same time, many writers do not want Substack to police how they work, and some are worried about false accusations, privacy, or their content being used to train models. Leadership asks you for a recommendation on whether Substack should build product features around AI transparency, and if so, what the first version should include for readers, writers, or both. The decision is urgent because norms are forming now, and any move could trigger strong reactions from different parts of the community. Assume a meaningful share of active writers already use AI in some part of their process, but there is no perfect way to know exactly how much. Your challenge is to improve trust without turning Substack into an AI ban or creating a tool that overclaims certainty.

What to produce

Write a 1 to 2 page product strategy memo.

Your answer must

State the core product problem and the user trust issue you are solving, with separate needs for readers and writers.
Recommend a product direction for the first release, including what surfaces you would build for readers, writers, or both.
Explain how your approach handles uncertainty in AI detection, creator control, and data use concerns.
Define success metrics and leading indicators, plus 2 to 3 key risks and how you would mitigate them.
Propose a rollout and communication plan, including how you would gather feedback after launch and decide what to change next.

Constraints

You cannot assume any AI detector is fully accurate or can prove authorship with certainty.
The solution must work for a wide range of creators, including those who use no AI, some AI assistance, or heavy AI assistance.
You should not require manual review of every post by a human team.
You must address concerns about whether creator content is used to train models.

Format

Use short sections with headings: problem, users, recommendation, trade-offs, metrics, risks, rollout.

How Substack actually played it

After releasing a way for readers to estimate how much text was written by hand or with AI, plus publisher tools such as pre-publish checks and a 'how I make this' statement, Substack summarises early reactions. It explains the stance that AI use is not the problem but lack of transparency is. Useful as a case on shipping a contested AI feature and listening to the community afterwards.

Choose a clear stance (transparency over bans) and design to it.
Give creators controls, such as opting out of scanning on a post.
State how data is used and not used for training.
Follow up publicly on reactions to build trust.

Built from How writers are reacting to Substack's AI transparency tools by Substack Team.