Telegram Voice or Text Triggers Apollo Lead Scraping, Parsed by GPT-4, Stored in Supabase

Input: Telegram voice/text command → Process: Whisper transcription + GPT parameter extraction → Apify Apollo Scraper → verify and deduplicate → Output: Supabase new leads table + Telegram receipt

Who

Growth hackers, SDR teams, and founders who want to collect and organize prospect leads via Telegram voice or text commands

You get

Turn natural-language commands into verified, deduplicated structured leads written directly to a Supabase database, with a Telegram count confirmation

You need

n8n · telegram · apify · supabase · postgres

Steps

1. Set up Telegram trigger and receive voice or text messages

   After importing the workflow into n8n, confirm the Telegram Trigger node (User message) is configured with a Bot Token. Send a text message or voice message to the bot. For voice, the flow routes through the Download File node to fetch the audio; for text, it proceeds directly.

2. Transcribe voice with Whisper and extract structured search parameters via GPT

   Voice messages are transcribed by the OpenAI Whisper node (Transcribe1); text messages pass through directly. Then the AI Agent node (Scraper agent) uses the GPT-4.1-nano model, following its systemMessage, to parse natural-language instructions into a fixed-format JSON array. The JSON contains location, business, and job_title fields, with values in lowercase and spaces replaced by plus signs. The Structured Output Parser node enforces the output format; Simple Memory provides conversation context. The Generate query payload node wraps the JSON as {"query": {{ $json.output }}}.

   Execution prompt · Source text。Scraper agent node's systemMessage field (AI Agent node → Options → System Message)

   Copy prompt

   =# Role
   You are LeadChat. Your single job is to return a **JSON array of one object** that tells the sub-workflow what to scrape.
   
   # Required JSON schema
   ```json
   [
     {
       "location": ["city+country"],      // lowercase, spaces→+
       "business": ["industry"],          // lowercase, words→+
       "job_title": ["role"]              // lowercase, words→+
     }
   ]

3. Generate Apollo search URL and call Apify Scraper to fetch leads

   The Create URL node (Code node) reads the {"query": ...} structure and builds an Apollo.io people-search URL with parameters including personTitles[], personLocations[], qOrganizationKeywordTags[], etc. Then the Run an Actor node calls the Apify Apollo-io Scraper Actor (ID: jljBwyyQakqrL1wae), passing the generated URL with totalRecords=500, getPersonalEmails=true, getWorkEmails=true, timeout=10000, waitForFinish=60.

   Execution prompt · Source text。Create URL node's JS Code field (Code node → JavaScript)

   Copy prompt

   /************************************************************
    * Build Apollo People-search URL from parser output
    * Handles both shapes:
    *   { query: [ { … } ] }      ← array
    *   { query: { … } }          ← object
    ************************************************************/
   
   // 1. Obtener el JSON que llega del nodo anterior
   const inputData = $json;
   
   // 2. Normalizar para que paramsData siempre sea un objeto
   let paramsData;
   
   if (Array.isArray(inputData.query)) {
     // Caso antiguo: query es un array
     paramsData = inputData.query[0];
   } else if (inputData.query) {
     // Caso nuevo: query es un objeto
     paramsData = inputData.query;
   } else {
     throw new Error('Falta la propiedad "query" en el input');
   }
   
   // 3. URL base de Apollo (interfaz web)
   //    Para llamar a la API REST cambia a https://api.apollo.io/v1/mixed_people/search
   const baseURL = 'https://app.apollo.io/#/people';
   
   // 4. Construir los parámetros dinámicos
   const queryParts = [
     'sortByField=recommendations_score',
     'sortAscending=false',
     'page=1',
   ];
   
   // Helper para arrays → personTitles[]=ceo
   const addArrayParams = (paramName, values) => {
     values.forEach((val) => {
       const decoded = val.replace(/\+/g, ' ');          // “barcelona+spain” ⇒ “barcelona spain”
       queryParts.push(`${paramName}[]=${encodeURIComponent(decoded)}`);
     });
   };
   
   // job_title → personTitles[]
   if (Array.isArray(paramsData.job_title)) {
     addArrayParams('personTitles', paramsData.job_title);
   }
   
   // location → personLocations[]
   if (Array.isArray(paramsData.location)) {
     addArrayParams('personLocations', paramsData.location);
   }
   
   // business → qOrganizationKeywordTags[]
   if (Array.isArray(paramsData.business)) {
     addArrayParams('qOrganizationKeywordTags', paramsData.business);
   }
   
   // campos estáticos
   queryParts.push('includedOrganizationKeywordFields[]=tags');
   queryParts.push('includedOrganizationKeywordFields[]=name');
   
   // 5. Montar la URL final
   const finalURL = `${baseURL}?${queryParts.join('&')}`;
   
   // 6. Devolver resultado
   return [{ json: { finalURL } }];

4. Extract fields, filter verified emails only, and deduplicate against existing records

   Data returned by Apify is mapped to unified fields by the Extract Info node: firstName, lastName, emailAddress, linkedInURL, jobTitle, companyName, location, country, websiteURL, businessIndustry, seniority, Number, emailStatus. The Only Keep Verified Emails node filters for entries where emailStatus='verified'. The Select already scraped mails node reads existing emailAddress values from the Postgres Leads_n-mail table. The Keep only the new leads node (Compare Datasets) has mergeByFields set to an empty object {} in the template, so the deduplication matching logic may not work as intended and requires user correction.

5. Insert new leads into Supabase database

   The Create rows with new leads node (Supabase) writes deduplicated new leads into the Leads_n-mail table. Field mappings use variables from the previous step. Note: the businessIndustry field expression references a nonexistent node 'Apollo Scraper' (actual node name is 'Run an Actor'), so this expression will evaluate to null and requires user correction. Also, the template sticky-notes declare columns including seniority and number, but the insert node does not map these fields, which is another template defect. You must create the corresponding columns in Supabase beforehand.

6. Limit count and send confirmation via Telegram

   The Limit node controls the final receipt volume. The Set Telegram message node constructs the text. The Confirmation message node (Telegram) sends this to the specified chatId. Note: the template hardcodes chatId as 123456789, which must be replaced; the message text refers to 'Google Sheet' as legacy wording, while data is actually written to the Supabase table.

   Execution prompt · Source text。Confirmation message node's Text field (Telegram node → Message → Text) and Set Telegram message node's assignments[0].value

   Copy prompt

   ={{ $input.all().length }} new contacts have been added to the Google Sheet!

You are an n8n workflow setup assistant. The user wants to replicate a Telegram-driven AI lead-scraping flow. Assist in this order: 1) Confirm the user has an n8n instance and required credentials (Telegram Bot Token, OpenAI API Key, Apify API Token, Supabase/Postgres credentials); 2) Guide importing workflow.json; 3) Check credential configuration node by node, especially Apollo Scraper Actor parameters and Supabase table name; 4) Remind the user to fix template defects: complete the Keep only the new leads mergeByFields, correct the businessIndustry node reference, and add seniority/number field mappings; 5) Suggest testing with a text message first, then a voice message; 6) Remind the user to replace hardcoded chatId and Apify token with environment variables or credential references.