Cold email drafted from a lead website and your offer

Cold email drafted from a lead website and your offer

An n8n workflow that drafts one cold email for one lead. You fill in a short web form with the lead's first name, company, website address and your offer in one sentence. The workflow reads the text of the lead's website, and an AI agent writes a short email that cites a real detail from that text. When the site gives too little to use, you get a line for a person to look at instead of a made-up email.

What is inside

• workflow.json: the workflow as n8n exports it, 6 nodes. All six run, and there are no sticky notes. It holds no credential.
• README.md: these notes.

What we changed from the original copy

• Chat Model used gpt-4o-mini with a temperature of 0.4. OpenAI's deprecation page, read on 12 November 2026, shuts gpt-4o-mini down on 14 January 2027 and names gpt-6-luna. That model's page says Chat Completions supports function calling only with reasoning effort set to none, and n8n's chat model node offers only low, medium and high. The agent node calls its model as a tool-calling chat model, so the node uses gpt-5.6-terra, whose page lists function calling without that limit. It has no temperature setting, because that page lists none.
• The credential link is gone, and the file's name is ours.
• Nothing else. The form fields, the two fetch nodes, the agent's rules and the completion page are as the file had them.

Mindket read this copy against n8n's source. It was not run on a live n8n and was not tried against any real website.

Before you start

• An n8n instance, on n8n Cloud or your own server.
• An OpenAI API key.
• A lead: a first name, a company and a website address that loads without a login, plus a one-sentence description of what you offer.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. Add an OpenAI credential and select it in Chat Model.
3. Open Lead Form and use its test address to open the form in your browser.
4. Fill in the four fields. Start the website address with https://, because the fetch node takes the address as typed.
5. Read the draft on the page that follows. When you are happy with it, publish the workflow so the form has a lasting address.

What happens in a run

1. Lead Form shows a form titled "Cold Email Writer" with four required fields: Lead first name, Lead company, Lead website URL and Your offer (one sentence).
2. Fetch Website requests the address and keeps the answer as text. It gives up after 15 seconds, and a failed request does not stop the run.
3. Extract Site Text takes the text of the page body into a field called site_text. It also continues on error.
4. Write Email gives the model the lead's details, your offer and the site text with its spaces collapsed, cut at the first 6,000 characters. The system message holds the rules. The model may use only what the page text says, and it treats that text as data rather than instructions. When the text is empty, an error page or under about 40 useful words, it answers [NEEDS-HUMAN] with a one-line reason.

The email gets a SUBJECT: line under 60 characters and a plain-text body of 60 to 120 words. The body has one concrete observation, one sentence tying it to your offer and one closing question. The model addresses the lead by first name, avoids a short list of stock phrases and writes in the language of the lead's site.

1. Show Result puts the model's answer on a completion page titled "Your draft is ready".

What to change

• The tone and the rules: the system message in Write Email.
• The length: the body-length rule in that system message.
• How much of the site is read: the 6000 in the text of Write Email.
• The model: Chat Model.
• Where the draft goes: replace Show Result with a mail or sheet node.

Good to know

• The workflow only drafts. Nothing is sent, so read and edit the draft before you use it.
• The model is told to use the page text only, but a model can still misread a page. Check the detail it cites against the site.
• The form has no sign-in. Anyone who has its address can make your n8n fetch a web address they type and spend your OpenAI credit. Share the lasting address only with people you trust, or turn on the form's basic authentication.
• A page with little text, a login wall or a block on automated requests gives the model nothing to cite, so you get the [NEEDS-HUMAN] line.
• The page text comes from the whole body of the page and the model sees only the first 6,000 characters of it, so a long menu can crowd out the content.
• Rules on unsolicited email differ by country. Check the ones that apply to you before you send a draft.
• Data: the lead's name, the company, the offer and the site text all go to OpenAI.
• Cost: one model call per draft, billed by OpenAI.