Designing a Food Delivery App Feature That Catches AI-Written Reviews
Company: Swiggy
Role: Product Manager
Category: Product Design & Strategy
Difficulty: medium
Interview Round: Other
You are a product manager at Swiggy, the Indian food delivery platform, where customers can review the restaurants they order from and the dishes they receive. Design a product that identifies reviews of restaurants or dishes that were generated by AI, and explain how it would work for the people who write reviews, the people who read them, and the restaurants being reviewed.
```hint Ask what harm the review causes
Think about what actually goes wrong for a customer deciding what to order when a review was produced by AI, and whether that is true of every AI-assisted review, before you choose any detection signals.
```
### Clarifying Questions
- Can only customers with a completed order leave a review, and do reviews include written text and photos or mainly ratings?
- Is the target fake reviews meant to manipulate ratings, such as paid praise or attacks by competitors, or any review written with AI help, including genuine customers who use AI to write or translate?
- What should happen to a flagged review: removal, exclusion from the rating, a visible label, or human review?
- Are AI-generated food photos attached to reviews in scope as well?
### What a Strong Answer Covers
- A problem definition that separates genuine AI-assisted reviews from inauthentic ones, and a clear choice of which the product acts on
- The stakeholders: customers reading reviews, honest reviewers, restaurant partners and the review-operations team
- A detection approach that combines order, account, behavior and network signals with text and image analysis, with a realistic view of how unreliable AI-text detectors are on their own
- Product responses scaled to confidence, including effects on the displayed rating, appeals, and consequences for restaurants that buy reviews
- Metrics for precision, reduced manipulation and customer trust, and a plan for adversaries who adapt
### Follow-up Questions
- Your system flags reviews written in Hinglish, or by people writing in their second language, far more often than others. How would you detect and fix that?
- A restaurant's rating jumps after a burst of glowing reviews from real accounts with real orders. Should your product catch this, and how?
- Should Swiggy show readers a label saying a review may be AI-written? Argue both sides and decide.
Overview: A product design question about building a system that detects AI-generated reviews of restaurants and dishes on a food delivery app. It tests problem framing, stakeholder needs, detection signals, actions scaled to confidence, handling of false positives, and metrics for trust in ratings.