Building an AI Marketing Agent with HubSpot + ChatGPT

Learn how to combine HubSpot’s CRM data with ChatGPT’s intelligence to build a marketing agent that automates campaigns, scores leads, and personalizes content at scale.

What You’ll Build

An AI-powered marketing agent that:

Reads HubSpot CRM data — contacts, deals, email performance
Analyzes with ChatGPT — lead scoring, content recommendations, campaign optimization
Takes action — updates HubSpot records, drafts content, triggers workflows via Zapier

Prerequisites

A HubSpot account with API access (any tier works)
An OpenAI API key (ChatGPT/ GPT-4)
A Zapier account (for no-code integration) or access to HubSpot’s API directly
(Optional) Hermes Agent for skill-based orchestration

Step 1: Connect HubSpot to ChatGPT via API

Option A: Direct API Connection (Developer-Friendly)

# Test HubSpot API connection
curl -s "https://api.hubapi.com/crm/v3/objects/contacts" \
  -H "Authorization: Bearer YOUR_HUBSPOT_ACCESS_TOKEN" \
  -H "Content-Type: application/json" | jq '.results | length'

Option B: Zapier Bridge (No-Code)

Create a new Zap: HubSpot + ChatGPT
Trigger: Contact Created or Deal Stage Changed in HubSpot
Action: ChatGPT — Send Prompt with HubSpot data
Response: Update HubSpot via Update Contact action

Step 2: Define Your Marketing Agent Personality

The prompt template for your AI marketing agent:

You are a Senior Marketing Operations Agent at {COMPANY_NAME}.
Your role is to:

1. SCORE leads based on: engagement level, company size, industry fit, email history
2. RECOMMEND next actions: send brochure, book demo, add to nurture sequence
3. DRAFT personalized follow-up emails using HubSpot contact data
4. ANALYZE email campaign performance and suggest improvements
5. FLAG high-value accounts for immediate sales attention

Available data: {CONTACT_FIELDS: name, company, title, email_opens, website_visits, deal_stage}
Channel: Email (HubSpot), CRM updates (HubSpot), Internal alerts (Slack)

Step 3: Build the Lead Scoring Workflow

Hermes Agent Skill Approach

Save as a Hermes Agent skill for reusable execution:

# Hermes Agent workflow: Score new HubSpot contacts
# 1. Fetch new contacts from HubSpot (API limit: 100)
# 2. Send each contact to ChatGPT for scoring
# 3. Update HubSpot with score + recommendation
# 4. Alert team for high-value leads via Slack

# Fetch contacts created in last 24h
CONTACTS=$(curl -s \
  "https://api.hubapi.com/crm/v3/objects/contacts/search" \
  -H "Authorization: Bearer $HUBSPOT_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "filterGroups": [{
      "filters": [{
        "propertyName": "createdate",
        "operator": "GTE",
        "value": "'$(date -d "24 hours ago" +%Y-%m-%d)'"
      }]
    }],
    "properties": ["email", "firstname", "lastname", "company", "hs_lead_status"]
  }')

Serverless Function Approach (Production)

Deploy as a Cloudflare Worker or Supabase Edge Function:

// HubSpot + ChatGPT Marketing Agent — Edge Function
const HUBSPOT_TOKEN = env.HUBSPOT_TOKEN;
const OPENAI_KEY = env.OPENAI_API_KEY;

async function scoreLead(contact) {
  const response = await fetch("https://api.openai.com/v1/chat/completions", {
    method: "POST",
    headers: {
      "Authorization": `Bearer ${OPENAI_KEY}`,
      "Content-Type": "application/json"
    },
    body: JSON.stringify({
      model: "gpt-4",
      messages: [{
        role: "system",
        content: "You are a lead scoring agent. Score leads 1-10 based on company size, title, and industry."
      }, {
        role: "user",
        content: JSON.stringify(contact)
      }]
    })
  });
  const data = await response.json();
  return data.choices[0].message.content;
}

// Run daily on new contacts
export async function scheduled(event) {
  const contacts = await fetchHubSpotContacts();
  for (const contact of contacts) {
    const score = await scoreLead(contact);
    await updateHubSpotContact(contact.id, { hs_lead_score: score });
  }
}

Step 4: Automate Content Personalization

Use ChatGPT to draft personalized email sequences for HubSpot campaigns:

Cold Outreach Personalization

# For each lead in a campaign, generate personalized intro
curl -X POST "https://api.openai.com/v1/chat/completions" \
  -H "Authorization: Bearer $OPENAI_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-4",
    "messages": [
      {"role": "system", "content": "Write a marketing email intro. 
        Tone: Professional but warm. 
        Personalize using: company name, industry, contact title.
        Keep it under 100 words."},
      {"role": "user", "content": "Contact: VP of Engineering at Acme Corp (SaaS, 200 employees)
        Product: AI-powered code review tool"}
    ]
  }' | jq '.choices[0].message.content'

Campaign Analysis Agent

Schedule weekly campaign performance analysis:

# Hermes Agent cron job: Weekly campaign analysis
# 1. Pull HubSpot email campaign stats
# 2. Analyze with ChatGPT
# 3. Generate recommendations
# 4. Post report to Slack

curl -s "https://api.hubapi.com/marketing/v3/emails" \
  -H "Authorization: Bearer $HUBSPOT_TOKEN" | \
  jq '.results[] | {name, stats: {opens, clicks, replies}}' > /tmp/campaigns.json

# Send to ChatGPT for analysis
curl -s -X POST "https://api.openai.com/v1/chat/completions" \
  -H "Authorization: Bearer $OPENAI_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-4",
    "messages": [
      {"role": "system", "content": "Analyze these email campaigns. 
        For each: What worked? What can improve? 
        Suggest A/B test ideas for underperformers."},
      {"role": "user", "content": "CAMPAIGN_DATA_HERE"}
    ]
  }' | jq '.choices[0].message.content'

Step 5: Deploy as a Recurring Automation

Option 1: Hermes Agent Cron Job

# Schedule daily lead scoring at 8 AM
hermes cron create \
  --schedule "0 8 * * *" \
  --prompt "Run the HubSpot lead scoring agent: fetch new contacts, score with ChatGPT, update HubSpot, alert Slack for 8+ scores."

Option 2: Cloudflare Workers Cron Trigger

// wrangler.toml
// [triggers]
// crons = ["0 9 * * *", "0 14 * * *"]

Option 3: Supabase Edge Functions

Deploy via MCP: the Supabase toolset can deploy edge functions that connect HubSpot Webhooks to ChatGPT.

What to Monitor

Metric	Target	Tool
Leads scored/day	All new contacts	HubSpot dashboard
Score accuracy	>80% agreement with manual scoring	Monthly audit
Email campaign CTR	+15% vs non-personalized	HubSpot email stats
Automation errors	<1% failure rate	Worker logs
API costs	<$50/month for ~5000 contacts	OpenAI usage dashboard

Reviews

Guides

OKF Bundles

FAQ

Q: Do I need a paid HubSpot plan to use the API? A: Yes — HubSpot API access for CRM objects requires at least the Starter plan ($50/month). Marketing Hub also requires a paid plan for email campaign APIs.

Q: Can this agent handle 10,000+ contacts? A: Yes — but you’ll hit API rate limits. Use batch processing (100 contacts per API call) and add rate limiting. The ChatGPT API handles rate limits gracefully with retry logic.

Q: What if ChatGPT’s lead scoring disagrees with my sales team? A: That’s actually valuable feedback. Log disagreements and periodically review them to tune your prompt. Over time, ChatGPT will learn your team’s preferences.

Q: Can this work without Zapier? A: Yes — use direct API calls (as shown in Step 3’s Hermes skill approach). The direct approach gives you more control but requires a small amount of code.

Q: How often should this agent run? A: Daily for lead scoring (covers new contacts), weekly for campaign analysis, and on-demand for content personalization. Running lead scoring more than once per day is usually overkill for most teams.