Send AI-written cold outreach and follow-up drafts with Gmail and Google Sheets using Firecrawl, Groq, and Gemini

AI-Powered Cold Outreach & Follow-Up Automation (with Do...While variant)

Automate the full cold email lifecycle — from intelligently crawling a prospect's website, to AI-writing a personalised email, verifying the address, creating a Gmail draft, and managing up to 5 context-aware follow-ups — all tracked in Google Sheets.

Good to know

• This template contains two complete, identical workflows side by side. The only difference is how the async scrape polling loop is built. Choose the version that suits your setup, then delete the other.
• Version A (left) — Standard. Uses built-in n8n nodes only. Works on n8n Cloud and any self-hosted instance with no extra setup.
• Version B (right) — Do...While. Uses the n8n-nodes-do-while community node to replace 5 nodes with 2. Requires installation under Settings → Community Nodes → n8n-nodes-do-while.
• Firecrawl batch scraping is billed per page. Test with a single domain first to understand your usage before running at scale.
• Gmail drafts are created — not sent automatically. Review and send manually. This is intentional so you stay in control of what goes out.
• The workflow stops following up after 5 attempts per lead (follow_ups < 5). Leads marked unsubscribed are permanently excluded.

How it works

Cold Email Branch (runs first, one lead at a time):

1. Reads all rows from a Google Sheets lead database and filters for leads with status = pending and follow_ups = 0.
2. Crawls the prospect's website sitemap using Firecrawl to discover all URLs.
3. Groq (ultra-fast LLM) filters the URL list down to the 10 most strategically useful pages — removing sitemaps, legal pages, RSS feeds, and other non-business content.
4. Firecrawl batch-scrapes all selected pages simultaneously and returns scraped markdown. Because Firecrawl scraping is asynchronous, the workflow polls for job completion rather than waiting blindly.

• Version A: polls using a SplitInBatches → Get Status → IF → Wait → Set loop (5 nodes, built-in only).
• Version B: polls using Get Status → Do...While (2 nodes, community node). The Do...While node has a built-in 5-second wait between iterations and a 30-iteration hard cap (≈ 2.5 min max) that exits gracefully instead of timing out.

1. All scraped page markdown is aggregated into one payload and passed to a Gemini AI agent, which reads the business, writes a personalised cold email, and extracts the best contact email it found in the content.
2. If the AI found a contact email, it is immediately verified with VerifiEmail. If no email was found in the scraped content, a Firecrawl AI Agent actively navigates the site (contact pages, footers, mailto links) to find one — then verifies that too.
3. If the email passes verification, a Gmail draft is created. If no valid email can be found at all, the lead is marked no_email in the sheet so it is never retried, preventing wasted API calls.
4. On success, the sheet row is updated: status = emailed, follow_ups = 0, email = <found address>.

Follow-Up Branch (runs automatically after cold emails complete):

1. Re-reads the lead database and filters for leads with status = emailed, a non-empty email, and follow_ups < 5.
2. Fetches the full Gmail conversation thread for each lead so the AI has full context.
3. A Gemini AI agent — enhanced with per-contact window memory (keyed to email address) — reads the entire conversation history and writes a context-aware follow-up reply. If the conversation suggests no follow-up is needed (e.g., a positive reply was received), the agent sets skip: true and the workflow moves on without creating a draft.
4. A Gmail draft is created in the same thread as the original email, so replies are properly threaded.
5. The sheet increments follow_ups by 1 after each successful draft.

How to use

1. Choose your version. Decide whether you want Version A (standard, built-in nodes) or Version B (Do...While, community node). If you choose Version B, install the community node first: Settings → Community Nodes → install n8n-nodes-do-while. Then delete all the nodes from the version you are not using.
2. Set up your Google Sheet. Create a sheet with these columns in order: domain, niche, region, email, status, follow_ups. Add your prospect domains, set status = pending and follow_ups = 0 for each new lead.
3. Connect all credentials. Add credentials for: Google Sheets (OAuth2), Gmail (OAuth2), Firecrawl API, Groq API, Google Gemini API, and VerifiEmail API.
4. Update the sheet ID. Open any Google Sheets node and update the documentId to point to your spreadsheet.
5. Customise the AI prompts. Open the ✍️ Write Cold Email agent node and update the system prompt with your name, company, offer, and tone. Do the same for ✍️ Write Follow-Up Email.
6. Test with one lead. Add a single row with status = pending and run the workflow manually. Verify the scraping, email generation, verification, and draft creation all work before adding more leads.

Requirements

• Firecrawl account for website crawling, scraping, and AI agent
• Groq account for fast URL filtering
• Google Gemini account for AI email generation (cold + follow-up)
• VerifiEmail account for email deliverability checking
• Google Sheets (OAuth2) for the lead database
• Gmail (OAuth2) for creating drafts and reading threads
• (Version B only) n8n-nodes-do-while community node installed

Customising this workflow

• Swap any LLM. Groq is used for URL filtering (fast, cheap) and Gemini for email writing (creative). Both can be replaced with any model your n8n instance supports — simply swap the model sub-node attached to each chain or agent node.
• Change the follow-up limit. The filter node checks follow_ups < 5. Update this number to control how many times the workflow will attempt to follow up before stopping.
• Add a scheduler trigger. Replace or supplement the Manual Trigger with a Schedule Trigger to run the workflow automatically — for example, every weekday morning to queue up the day's drafts.
• Add a webhook for new leads. Instead of manually adding rows, connect a form, CRM webhook, or Apollo/Hunter.io export to append rows directly to your Google Sheet, then let the workflow pick them up automatically on its next run.
• Adjust the scraping depth. The AI URL filter is set to keep a maximum of 10 pages. For large or complex sites where you need more context, increase this limit in the 🤖 AI: Filter Strategic URLs node's prompt.
• Understanding the Do...While node (Version B). The n8n-nodes-do-while community node is a true while loop for n8n — it repeatedly executes connected nodes until a condition you define evaluates to true. In this workflow it replaces the classic SplitInBatches + IF + Wait + Set polling pattern. The node is wired so that Get Crawl Status runs first (do), then the condition $json.data?.status !== 'scraping' is checked (while). If still scraping, it waits 5 seconds and loops; if done, it exits to the Aggregate node. The maxIterations: 30 cap ensures the loop never runs for more than 2.5 minutes regardless of scrape duration.