Exit Interview via LLM: What Users Say When Closing the Paywall
OpenAI
An English-language AI news article describes how integrating a short LLM-powered interview after a paywall dismissal yielded a 34% response rate and uncovered key user objections, including the discovery that many users arrived via search expecting a free product. The article provides the architecture, prompts, and code framework used, along with insights from two weeks of data.
A developer built an LLM-based exit interview bot for a freemium online board game. When a user closes the paywall, a separate chat appears asking 5 questions max. Over 2 weeks, 59 sessions occurred with 20 responses (34%). The main finding was that a significant portion of users came from search with the word 'free' in the title. The system uses a separate database table, a transaction-based reply cap, a sweep for stale sessions, a second LLM call to generate a structured note, and a webhook to Telegram. The article shares the database schema, code for the capped transaction, and both the system prompt and the note prompt. Results showed 7 out of 20 respondents expected the product to be free, leading to a quick fix of the site title. The tool is recommended for any product with an exit moment like a paywall, abandoned cart, or cancellation, as long as the user remains on the page.
- Abbreviations
- LLM = Large Language Model — большая языковая модель
- CustDev = Customer Development — развитие клиентов
- SEO = Search Engine Optimization — поисковая оптимизация
- CTR = Click-Through Rate — рейтинг кликов
- API = Application Programming Interface — интерфейс программирования приложений
- JSON = JavaScript Object Notation — нотация объектов JavaScript
Source: Habr — хаб ИИ —
original