NameDrop

Does ChatGPT name-drop you? NameDrop is a self-hosted tracker for how often your brand — and your competitors — appear in AI assistant answers to the questions your customers ask. A single Cloudflare Worker collects answers daily from ChatGPT, Perplexity and Gemini, detects Brand mentions, and serves a dashboard plus CSV exports. An alternative to hosted AEO trackers like Peec.ai, at a fraction of the price.

Domain vocabulary is defined in CONTEXT.md — Brand, Prompt, Topic, Surface, Run, Response, Mention, Position, Visibility, Share of Voice, Source, Domain, Domain Type, Fanout. Those terms are used in the code, schema and UI.

ChatGPT collection requires a paid cloro.dev plan (Lite or above). The ChatGPT Surface is a real chatgpt.com session driven through cloro's API — that is what makes the numbers match what real users see. The free tier runs a single concurrent job, which puts a daily sweep past the Cloudflare Cron Trigger's 15-minute cap; collection needs the Lite plan's concurrency. Note the supply risk: cloro scrapes chatgpt.com, which sits in ToS-gray territory. If it dies, an OpenAI API fallback path lives in src/lib/surfaces/chatgpt.ts — swapping the adapter back is a small, tested change, at the cost of measuring the API instead of the product.

Shape

One Cloudflare Worker does everything:

• fetch — the Astro dashboard and the CSV export endpoints.
• scheduled — the daily collection sweep. One sweep is every active Prompt against one Surface, on its own schedule (07:00, 07:20 and 07:40 UTC), because a Cron Trigger is killed at 15 minutes of wall clock and all three Surfaces in one firing does not fit.

Storage is D1. Auth is Cloudflare Access in front of the Worker; there is no auth code in the app.

The dashboard has five views: Overview (Visibility trends and the Brand leaderboard), Prompts (per-Prompt metrics for content targeting), Runs (browse what the assistants actually said), Sources (Domain leaderboard and Domain Type mix), and Health (per-Surface collection coverage). /export/*.csv serves Peec-shaped exports for downstream content tooling — see docs/exports.md.

Quickstart

Prerequisites: a Cloudflare account, Node 22+, and API keys for Perplexity, OpenAI (Domain classifier + calibration), Gemini, and cloro.dev (paid plan — see above).

1. Configure what you track. config/ ships as a worked example in the podcast-hosting niche — brands.json (mark exactly one Brand isSelf; the example uses Castos as "you" against real competitors), prompts.json, topics.json, and domain-types.json (a tiny seed map; unseeded Domains are auto-classified by LLM at first sight). Replace with your own market.

2. Create the D1 database and paste the returned database_id into wrangler.jsonc (replacing PLACEHOLDER_SET_BY_WRANGLER_D1_CREATE):

   npx wrangler d1 create namedrop

3. Apply migrations: npm run db:migrate:remote

4. Set the four provider secrets:

   npx wrangler secret put PERPLEXITY_API_KEY
   npx wrangler secret put OPENAI_API_KEY
   npx wrangler secret put GEMINI_API_KEY
   npx wrangler secret put CLORO_API_KEY

5. Deploy: npm run deploy

6. Put Cloudflare Access in front of the Worker. This is not optional — there is no auth code in the app; an unprotected hostname serves your whole dashboard and database to the internet. See docs/deploy.md for the exact policies, including the service-token setup for scripted access to the exports.

Collection then runs daily on the cron schedule. The Health page shows per-Surface coverage; docs/deploy.md covers verifying the first sweep and re-running a lost date.

Running it

• docs/deploy.md — one-time setup, Access, re-running a date.
• docs/costs.md — where the money goes and why the collection settings are what they are. ChatGPT is effectively the whole bill; Gemini and Perplexity are a rounding error. Read this before changing anything about how ChatGPT is called — several obvious cost ideas are measured dead ends.
• docs/known-issues.md — failure modes the original deployment actually hit, and the open ones. Read this before debugging anything that looks like a timeout, a 500, or missing data.
• docs/exports.md — the CSV exports.

What is tracked

config/ holds the versioned config: topics.json, prompts.json, brands.json (with alias lists), and domain-types.json (the Domain Type seed map). Editing tracking is a git commit — POST /api/sync-config reconciles the files into D1, upserting changes and deactivating anything removed. Nothing is ever deleted, so history stays queryable.

Development

npm install
cp .dev.vars.example .dev.vars   # provider API keys, never committed
npm run db:migrate:local
npm run dev

Tests

npm test        # builds, then runs the suite
npm run typecheck

Tests run the real Worker in workerd against a real local D1 via Cloudflare's Vitest workers pool, driving it through its two real entry points — the request handler and the cron handler. The only thing faked is the HTTP boundary to the AI providers, using recorded fixtures. Parsing, D1 writes, metric SQL, page rendering and CSV output are all exercised for real.

npm test builds first because the pool loads the built Worker from dist/server. worker-configuration.d.ts is generated by wrangler types (run automatically before npm run typecheck) and is not committed.

License

MIT