Scrape Google Maps by area & generate outreach messages with n8n

Input a Maps location URL, keyword, and country → Bright Data scrapes local businesses → AI generates cold-call scripts → Store in Postgres/Supabase

Who

Sales teams, BD reps, and n8n users who need bulk local-lead extraction with AI-generated outreach scripts

You get

Automatically scrape Google Maps business listings for a target area, generate a personalized cold-call opener and talking points per lead, and persist everything to a database

You need

n8n · bright-data · gemini · postgres · supabase

Steps

1. Submit the form and trigger the Bright Data scrape job

   Fill the n8n form: Location URL (copy from the browser address bar, NOT the Share feature), Keyword Search (e.g., Restaurant), Country (e.g., US), and My company's segment (e.g., Marketing Agency). The workflow extracts latitude and longitude via regex from the URL, then POSTs to https://api.brightdata.com/datasets/v3/trigger with dataset_id=gd_m8ebnr0q2qlklc02fz, zoom_level=14, keyword, lat, and long.

2. Poll until the scrape finishes and download the snapshot

   The workflow enters a retry loop: initialize a counter to 0, then call the Bright Data node monitorProgressSnapshot for the snapshot_id. If status is not 'running', proceed to download; if still running and the counter is below 10, Wait 60 s, increment the counter, and retry; if the limit of 10 is reached, stop. The download node uses downloadSnapshot to fetch results.

3. Organize data and limit batch size

   Map raw Bright Data fields into a structured object: place_id, company.google_url, company.name, company.category, company.address, company.description, company.reviews_count, company.rating, company.services_provided, company.open_website, company.phone_number, company.top_reviews. The Limit node caps processing at 15 items (adjustable). SplitInBatches feeds items one-by-one so that an error on one item does not block previously processed items.

4. Check for a website, scrape it, and summarize

   For each item, check whether company.open_website exists. If it does, the AI Agent node calls the MCP tool scrape_as_markdown to fetch the homepage, then uses gpt-5-mini with the system prompt to produce a 300–500-word company summary. If no website exists, skip scraping and go straight to Merge.

   Execution prompt · Source text。AI Agent node 'Scrape & Summarize' text parameter (user message / tool-call instruction)

   Copy prompt

   =Use the tool 'scrape_as_markdown' with the url:
   {{ $json.company.open_website }}

   Execution prompt · Source text。AI Agent node 'Scrape & Summarize' options.systemMessage field

   Copy prompt

   =You are a web-scraping AI agent equipped with the tool 'scrape_as_markdown'.
   
   # TASK
   Scrape the homepage of the provided website and generate a structured summary of the company based on the main content.
   
   ## OBJECTIVES
   1. Scrape.
   2. Extract only the core content of the page.
   3. Identify key information:
      - Company mission or vision
      - Products or services offered
      - Target audience or market
      - Key messaging, branding, or positioning
      - Any highlighted achievements, updates, or announcements
   
   ## CONSTRAINTS
   - Prioritize meaningful textual content over visuals or layout elements.
   - Ensure clarity and neutrality in the summary.
   
   # OUTPUT REQUIREMENTS
   Return a summary of 300-500 words in the following format:
   [Company Name]:
   
   [Summary of company in 300-500 words, clearly structured and written in neutral, informative language.]

5. Generate the cold-call script and talking points with AI

   The Merge node combines the structured company data with the page summary, then passes it to the Message Generator (Basic LLM Chain). This node uses the Google Gemini Chat Model (models/gemini-2.5-pro) and the form-submitted 'My company's segment' to generate a personalized cold-call opener (2–4 sentences) and 3–5 talking points.

   Execution prompt · Source text。Basic LLM Chain node 'Message Generator' messages.messageValues[0].message field

   Copy prompt

   =You are a business development assistant helping an entrepreneur prepare for a cold call to a business lead.
   
   Take into account that my company segment is: {{ $('On form submission').first().json['My company\'s segment'] }}
   
   # TASK
   You will receive structured data about a business, including basic info, services, reviews, and contact details.
   
   Your job is to generate:
   1. A short cold call opening message the entrepreneur can use to start the conversation.
   2. A list of supportive talking points or insights the entrepreneur can use during the call to sound informed and build rapport.
   
   # OUTPUT STRUCTURE
   1. Opening Message (Spoken Script):
   - 2–4 sentences.
   - Friendly, confident, respectful tone.
   - Show that the caller is familiar with the business.
   - Mention something specific (e.g. high rating, food quality, atmosphere, community praise).
   - Introduce the caller’s intention (e.g. connect, collaborate, learn more).
   
   2. Call Support Notes (Bullet List):
   - Provide 3–5 relevant talking points about the business based on reviews, services, and offerings.
   - Suggest areas the entrepreneur can mention or ask about (e.g. “Your place seems popular with tourists and locals alike — how do you approach that balance?”)
   - Avoid quoting reviews verbatim — instead, summarize sentiment or praise.
   
   # INSTRUCTIONS
   - Generate a personalized cold call opening message
   - Generate a list of insightful, natural talking points to help the entrepreneur carry the conversation confidently
   
   # GUIDELINES
   - Output only the results, without any intro or outro text.

6. Merge results and upsert into the database

   Pass the Message Generator output text (sales_helper) together with the merged company data into the Postgres node, which upserts into table business_scraping_result on matching column place_id. Fields written: name, rating, address, website, category, place_id, google_url, description, top_reviews, page_summary, phone_number, sales_helper, reviews_count, services_provided. Connect to Supabase using the Transaction pooler parameters.

You are an n8n workflow setup assistant. The user wants to replicate the 'Scrape Google Maps by area & generate outreach messages' template. First confirm the user has credentials for: Bright Data API, OpenAI API (or Google AI API), and Postgres/Supabase connection details. Then guide them to import workflow.json, configure the form node fields, the Bright Data node's zoom_level and dataset_id, the AI Agent's MCP token, the LLM Chain model credentials, and the Postgres upsert table connection. Advise running the Create Table node manually before submitting the form for a test.