Summary: Firecrawl’s ChatGPT plugin, written up by Richard Oliver Bray on the Firecrawl blog on October 11, 2026, connects ChatGPT to live business data — Google Maps listings, website scrapes, and people-data providers like Apollo and FullEnrich. In the demo, three prompts built a list of 40 local firms, checked 10 websites for chatbots, and surfaced decision makers’ work emails in about 21 minutes total. If you sell to local businesses, this changes lead gen from prompt-crafting to tool orchestration.
Why ChatGPT alone can’t build the list
ChatGPT is fine at scoring leads you already have or drafting the outreach email. What it can’t do reliably is the front of the funnel: find the actual firms near you, check their websites, and produce a real person’s work email. That requires current directory data and a contact database, which it doesn’t have built in.
The fix in the post is a plugin. Once Firecrawl is installed in ChatGPT, you type @Firecrawl before a prompt and the model can call scraping tools plus Firecrawl Alexandria — a catalogue of data providers including Firecrawl Maps search, Apollo, FullEnrich, and Data Legion. You don’t even need to name a provider; ask for “the relevant Alexandria providers” and ChatGPT picks.
The three prompts, and where the time went
The demo scenario is concrete: a business selling AI chatbots to accountants and financial advisers in Brighton, UK.
- Find the firms. A Maps search returned 40 listings — 20 accountants, 20 IFAs — in 43 seconds. One quirk: Firecrawl’s Maps tool has no radius parameter, so ChatGPT centred on Brighton and checked each listing’s coordinates itself. Asking for “within 20 miles” in one query won’t widen the search; you’d query nearby towns separately.
- Qualify by website. Scraping and screenshotting 10 firm sites took 5m 15s and found no chatbot on any of them. ChatGPT flagged its own blind spot here: late-loading chat widgets can hide from a single page load, and Firecrawl’s Interact tool — a cloud browser that can click and wait — is the stronger check, at a higher credit cost.
- Get the contact. The final prompt pulled each firm’s most senior decision maker and work email through Alexandria in 15m 15s. The results table was redacted because it holds real people’s data.
About 21 minutes from empty chat to qualified list with contacts.
The cost logic is the real lesson
Every Alexandria tool lists its credit price before it runs, and the workflow’s ordering matters more than the raw prices. Apollo’s people search costs 0 credits, so you see who exists at a company for free. A FullEnrich work email costs 20 credits — and you pay that only for firms that already passed the Maps and website checks. A Maps search is 5 credits per call for up to 20 places; Scrape is 1 credit per page.
This is the same principle behind ordering agent calls from cheap to expensive: exhaust the free signals before spending on the expensive one. I made a similar case in Cut Your Agent’s Token Bill in the Right Order — the sequence of calls often determines the bill more than any single call’s price.
Setup and limits, briefly
Setup takes a Firecrawl account (the free plan includes 1,000 monthly credits, no card), accepting terms for each people-data provider in the Firecrawl dashboard, and installing the plugin in ChatGPT. No API key gets pasted into ChatGPT — you sign in and approve, and calls bill to your account’s credits. The keyless tier doesn’t cover this workflow because people-data providers require a signed-in team that accepted terms.
Two caveats worth respecting before you run this for real. First, provider terms carry usage rules — FullEnrich’s terms bar marketing to personal email addresses, which is why the walkthrough only requests work emails. Second, expect some email lookups to come back empty, and verify names against each firm’s own site before relying on them. Start with the free plan on one town, confirm the output quality, and scale to paid credits only if the list converts.