Update: Thursday, March 6, 2025 we graduated this feature to be available to all users asking a question on Stack Overflow with the Ask Wizard.

Question Assistant is a feature in which machine learning models are utilized to evaluate askers’ question drafts, and an LLM provides real-time suggestions based on the ML model’s outputs before they’re submitted for human review in Staging Ground or published to Stack Overflow. Before this feature was graduated, we ran two experiments with the initial goal to see whether AI can help askers get past superficial problems and streamline the question asking process.

Ultimately the outcome of our initial hypothesis was inconclusive, as we found that neither approval rates nor average review times improved significantly for the variant/experiment group compared to the control/baseline group in either experiment. However, we considered both a success, because of some of the other metrics we tracked. Specifically, we observed a meaningful increase in success rates, which validates the value to askers and positive impact on question quality.

Why use Question Assistant?

• Askers get instant suggestions and actionable feedback on how to improve their draft.
• Leverage Question Assistant to learn community norms and build confidence in their question writing skills for contributing knowledge in the future.
• Increases quality of content posted on Stack Overflow.

How does Question Assistant work?

Staging Ground:

1. New askers begin their question draft via the ‘Ask Question’ button with the Ask Wizard by writing a title and describing their problem.
2. Machine learning predicts the presence or absence of four feedback indicators in the draft (Problem Definition, Attempt Details, Error Details, Missing Minimum Viable Reproducible Example).
3. Information is passed to Google Gemini to display suggestions on how the asker can improve their draft within the right rail.
4. New askers use LLM-provided suggestions to revise their question draft. After incorporating the initial improvements, the asker can press “refresh” to get updated guidance before submitting their question to a reviewer.

Stack Overflow Ask Wizard:

1. Askers begin their question draft via the ‘Ask Question’ button by writing a title and describing their problem.
2. Machine learning predicts the presence or absence of four feedback indicators in the draft (Problem Definition, Attempt Details, Error Details, Missing Minimum Viable Reproducible Example).
3. Information is passed to Google Gemini to display suggestions on how the asker can improve their draft within the right rail.
4. Askers use LLM-provided suggestions to revise their question draft. After incorporating the initial improvements, the asker can press “refresh” to get updated guidance before publishing their question on Stack Overflow.

*AI does not edit users' content and the suggestions are completely optional. Questions are still routed to Staging Ground reviewers before becoming public.