Landwehr turns signup questions into a live prompt corpus
A second question on the signup form captures the exact prompts that brought users in, then feeds them to a visibility tracker.
Malte Landwehr described his setup in a LinkedIn post on 16 September 2026. When someone signs up and picks "LLM / Chatbot" as their referral source, a follow-up question asks which prompt they used.
Those prompts are stored. Once a week, Claude Cowork reads them from PostHog via MCP, strips gibberish and duplicates, and pushes the clean set into Peec AI via MCP. Peec AI then returns source and visibility data for each prompt.
Landwehr described the same approach on the podcast, framing it as a second question layered onto standard self-reported attribution (listen).
We think the interesting move here is the sourcing, not the automation. Most prompt lists used for AI Search visibility work are guessed, scraped, or bought. This one comes from people who actually converted, which means the visibility scores map to prompts that already produced a user.
Before copying the pattern, check what your signup flow already collects and whether a free-text prompt field is legal under your consent basis. Then compare the prompts users report against whatever list you currently track. If the overlap is thin, your visibility numbers are answering a question no customer asked.