White Petal
A tool that checks what AI assistants say about a brand, compares it with competitors, and drafts the content changes worth trying.
- Status
- Live. Running at whitepetal.perfstaq.com, now presented as Perfstaq's AI visibility platform
- My role
- Co-builder. Most of the application code, with one other engineer.
- When
- Oct 2026
- Context
- Perfstaq product and client implementation
- Capabilities
- LLM apps, Web, Integrations

The problem
Buyers now ask an AI assistant which product to pick. A company has no easy way to see what those assistants answer, which competitors they recommend instead, or which sources the answer was built from.
What I built
A deliberately small application: a plain Node server, Postgres, and a front end with no framework.
- A runner that asks the same buyer questions across several AI assistants and records the answers.
- Competitor tracking: where another brand is recommended instead, and which sources were cited.
- Drafted fixes for review, such as robots.txt and llms.txt changes, schema markup and article outlines.
- Organisations with four roles, scheduled re-checks, metered usage with a monthly budget cut-off, and paid plans.
Decisions
Model answers vary from run to run, so each question is asked up to three times and the results are combined. A single answer would be noise.
Client organisations use their own encrypted API keys, with no fallback to a shared platform key. A client's usage can never land on someone else's bill.
Drafts are suggestions for a person to review. The tool does not publish anything on its own.