Revealed preference, not stated preference

See what AI coding assistants actually recommend

We analyze real Claude Code, Codex CLI, and Cursor transcripts to measure how often your product gets mentioned, installed, and configured in developer sessions.

Analyze your product Sign in

Signal detection

transcript analysis

$ observatory scan --source claude-code

scanning ~/.claude/projects/**/*.jsonl

found 2,847 sessions across 412 projects

supabase████████████████░░░░ 847

prisma██████████████░░░░░░ 712

stripe████████████░░░░░░░░ 623

vercel██████████░░░░░░░░░░ 501

tailwind████████░░░░░░░░░░░░ 489

■ installed ■ configured ■ recommended ■ mentioned ■ compared

How it works

01 — Ingest transcripts

We scan JSONL conversation logs from Claude Code and Codex CLI. Every user message, assistant response, and tool invocation is parsed.

02 — Extract signals

Package installs, config file writes, import statements, and text mentions are matched against a 60-vendor taxonomy. Each signal is classified by action type: installed, configured, recommended, compared, or rejected.

03 — Measure presence

See how your product ranks across thousands of real developer sessions. Compare against competitors. Track changes over time.

What are AI assistants saying about your product?

Enter your domain. Get a report in under a minute.

Start free analysis