Instant AI phone call to new contact-form leads

Instant AI phone call to new contact-form leads

An n8n workflow. When someone sends your contact form, it reads their website, looks for their LinkedIn profile, logs
the lead in Google Sheets and has a Vapi voice assistant phone them, briefed on who they are.

What is inside

• workflow.json: the workflow as n8n exports it, 11 nodes, one of them a sticky note. It holds no credential and none
  of its author's spreadsheet. The two API keys, the Vapi assistant and the Vapi phone number are placeholders.
• README.md: these notes.

Before you start

• An n8n instance, on n8n Cloud or your own server.
• An OpenAI API key.
• A Tavily API key.
• A Google account and a spreadsheet for the leads (below).
• A Vapi account with an API key, an assistant and a phone number to call from.

The spreadsheet

Write these seven column names in row 1 of the spreadsheet's first tab:

Name, Phone number, Website URL, Status, Notes, Linkedin URL, Lead summary

Append row in sheet adds one row per lead. Status is always New Lead, and Notes holds the enquiry as the lead
typed it.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. Copy the spreadsheet's ID: the long part of its address between /d/ and /edit.
3. Add a Google Sheets OAuth2 credential, select it in Append row in sheet and replace YOUR_SHEET_ID with the ID.
   The node writes to the first tab.
4. Add an OpenAI credential and select it in OpenAI Chat Model, Simplify enquiry and Format phone number.
5. In Find Linkedin URL, replace <YOUR_TAVILY_API_KEY> in the Authorization header with your Tavily key. The
   brackets go too, so the value reads Bearer, a space and the key.
6. In HTTP Request, do the same with your Vapi private API key. Then put your assistant's ID in place of
   YOUR_ASSISTANT_ID and your Vapi phone number's ID in place of YOUR_PHONE_NUMBER_ID, both in the JSON body.
7. Set up the assistant in Vapi so it uses the three values the call sends (below).
8. Save, send the form once from its test address with your own phone number, then press Publish and link your site's
   contact button to the form's production address.

The Vapi assistant

The file does not include the assistant: its voice, model and script live in your Vapi account. The call hands it three
values in assistantOverrides.variableValues:

| Variable | What it holds |
|---|---|
| name | The Name field of the form |
| howWeCanHelp | The enquiry, shortened by Simplify enquiry into a phrase that speaks to the lead as "you" |
| BusinessInfo | The summary from Lead summary Agent |

Vapi's API reference, read on 2 November 2026, says these values fill the assistant's text with LiquidJS syntax: {{name}}
in a first message becomes the lead's name. A first message like this one fits the phrase Simplify enquiry writes,
which is built to finish a sentence:

Hi {{name}}, this is Sam from Acme. You recently submitted an enquiry about {{howWeCanHelp}}. Is now a good time?

Put {{BusinessInfo}} in the assistant's system prompt, so it knows the company before it speaks.

The same reference lists the endpoint as POST https://api.vapi.ai/call with this body, while the file posts to
https://api.vapi.ai/call/phone. If Vapi answers the old address with an error, change the URL in HTTP Request.

What runs, in order

| Node | What it does |
|---|---|
| On form submission | Takes Name, Phone number, How can we help you and Website URL |
| Scrape website URL | Fetches the page at the address the lead gave |
| Markdown | Turns that page's HTML into Markdown |
| Find Linkedin URL | Asks Tavily for the name's LinkedIn profile and keeps the first result |
| Lead summary Agent | gpt-4.1-mini writes up to 100 words on the lead and the company, from the page and the search result |
| Append row in sheet | Adds the lead, the profile's address and the summary to the sheet |
| Simplify enquiry | Turns the enquiry into a short phrase for the call |
| Format phone number | Adds a country code when the number has none |
| HTTP Request | Asks Vapi to place the call |

Lead summary Agent reads only the page at the address given, usually the home page, not the whole site.

Change the result

• The summary: the system message of Lead summary Agent, which caps it at 100 words.
• The search: the query in the body of Find Linkedin URL. It searches the name alone, and adding the company's
  domain narrows it.
• The country code: the prompt of Format phone number. Name your home country there, so a local number gets its code.
• What the assistant says: its first message and prompt in Vapi, not in n8n.
• The form: the four fields of On form submission. Other nodes read them by label, so rename them there too.

Good to know

• Website URL is optional on the form, but Scrape website URL needs it. A lead who leaves it empty stops the run at
  that node, with no row and no call. Make the field required in the form node.
• A common name can bring back someone else's profile, since the search uses the name alone and keeps the first hit.
  The summary then describes the wrong person.
• The country code is the model's guess, and the prompt's only example is +44. Vapi expects numbers in E.164 form, such
  as +14155551234, unless you switch that check off.
• HTTP Request carries on after an error, so a refused call still ends the run as a success. Look at its output under
  Executions. The three values go into the JSON body as they are, and a double quote or a line break in the summary can
  break it.
• Open form: anyone with its address can type any number, and your Vapi account will call it.
• Calling rules: many countries ask for consent before an automated or AI voice call. Say on the form that you will
  phone back, and check the rules where your leads live.
• Costs: three short gpt-4.1-mini calls, one basic Tavily search and the call's minutes on Vapi for each lead.
• Where the data goes: the website's page and the lead's details go to OpenAI, the name goes to Tavily, and the number,
  name, enquiry and summary go to Vapi.