The free HubSpot AI Playbook

HubSpot holds your data, but you are still the memory. You remember what a client said last Tuesday, what you promised, and which deals are quietly going cold. An AI assistant connected to HubSpot, with a place to keep notes and a handful of saved skills, takes that load off you. This playbook sets that up, then shows how the skills that stick become agents for your whole team.

Where to start

Six skills are coming up. You don't need all six on day one. Answer six quick questions and we'll tell you which one to set up first and in what order.

Set up your assistant

Four steps, about an afternoon. Do them in order: the rules come before the connection on purpose.

1. Pick an assistant your company approves

   Most major AI assistants can now handle what this playbook needs. Check for four things:

   HubSpot's official connectorconnects with your own HubSpot login and permissions

   Projects or memorykeeps context from one conversation to the next

   Saved instructions or skillsrun a long prompt with one line

   Scheduled tasksnice to have, for the morning brief

   Before you connect anything, make sure your company has approved the tool for customer data. If there's an IT or security team, this is a five-minute email that saves a hard conversation later.

2. Give it operating rules

   Teach it how to behave before you teach it what to do. Paste these into your assistant's project instructions or custom instructions. They head off the common AI-in-the-CRM mistakes.

   Operating rulesCopy

   You are my AI assistant at [your company]. I'm the [your role]. We sell [what you sell] to [who you sell to], and our deals run through the [pipeline name] pipeline in HubSpot.
   Rules:
   1. Read HubSpot freely. Never create, update, delete or bulk-edit records unless I approve that exact change.
   2. Never send an email or message in my name. Draft only, in a [your tone] tone.
   3. Before any task with more than a few steps, show me a short plan and wait for my go-ahead.
   4. Save what matters. Decisions, commitments and account details go in my notes so you remember them next time.
   5. If a request is vague or risky, ask me first.
   6. If something can't be undone, stop and confirm.

3. Connect HubSpot

   HubSpot now has an official connector for AI assistants. Add it from your assistant's connectors or integrations settings, or from HubSpot's own AI tools page, and sign in with your HubSpot login. The assistant sees what your HubSpot user can see, and nothing more.

   If your connector has a setting that asks for approval before any change to HubSpot, turn it on. Then test with one question: "How many open deals do I have in the [pipeline name] pipeline, and what's the total value?"

   Start read-heavy. Let it earn write access after two weeks of output you trust.

4. Give it a memory

   This is the step most people skip, and it's the one that makes the rest work. Your assistant's project memory holds the basics. For anything you want to keep, use a set of notes it can read and write: account context, call summaries, past reports. Next session it reads them and picks up where you left off.

   Here's a structure that works. Copy it, then adjust once you see what you actually use.

   /ai-workspace
      /accounts        one folder per account
         /example-co
            context      who they are, goals, history, preferences
            /calls       every call summary
      /briefs          morning briefs
      /audits          data quality audits, month by month
      /reports         monthly report narratives
      /playbooks       how your team does things, in plain English
      /templates       email and proposal language you reuse

   Rule of thumb: if you'd want to remember it next month, it goes in a note.

Six skills

A skill is a saved prompt with the steps, rules and format built in. Instead of retyping a long prompt, you type one line: "Run my morning brief." Save each one below in your assistant's skills or saved instructions, or as a note it can follow.

Build my morning pipeline brief for [your company]. From the [pipeline name] pipeline in HubSpot, pull:
1. Deals that changed stage since yesterday
2. Deals closing this month, largest first
3. Open deals with no activity in 14 days, flagged as at risk
4. My tasks due today and anything overdue
5. Today's meetings, with the contact, company and open deal
Then check my account notes for anyone I'm meeting today.
Format it as: Moved / At risk / Today's meetings (with context) / My top 3 priorities. Keep it to one screen.
Save it to /briefs/[today's date].

Prep me for my meeting with [contact or company].
From HubSpot: the contact and company records, open deals (stage, amount, close date), the last five activities and emails, and any open tickets.
From my notes: everything in /accounts/[company], including past call notes, decisions and open commitments.
Give me one page:
1. Who they are and why they're looking at [what you sell]
2. Where the deal stands in the [pipeline name] pipeline
3. Open commitments, mine and theirs
4. A suggested agenda
5. One risk to watch
Save it to /accounts/[company]/prep-[date].

Here are my notes or the transcript from today's call with [company]: [paste]
1. Summarize the decisions, action items (mine and theirs), open questions and risks.
2. Draft a follow-up email in a [your tone] tone, written as the [your role] at [your company]. Draft only.
3. Write a short call note I can log on the HubSpot record.
4. List any HubSpot fields that should change (deal stage, amount, close date, next step) and wait for my approval before changing them.
5. Save the summary to /accounts/[company]/calls/[date] and update /accounts/[company]/context with anything that changed.

A new inbound lead just came in: [paste the form submission or the HubSpot record link]
1. Research the person and their company from HubSpot and their public website: what they do, their size, and whether they look like a fit for [what you sell] sold to [who you sell to].
2. Check HubSpot for existing contacts, companies or deals with them.
3. Tell me in two lines: fit or not, and why.
4. Draft a first reply in a [your tone] tone that answers what they asked and suggests a next step. Draft only.
5. Suggest an owner and lifecycle stage, and wait for my approval before setting either.

Run a data quality audit on the [your company] HubSpot portal. Report only, change nothing. Count:
- Contacts missing an owner, lifecycle stage or email
- Companies with no associated contacts
- Open deals in the [pipeline name] pipeline missing an amount or close date
- Open deals with a close date in the past
- Likely duplicate contacts and companies
For each one: the count, the ten worst examples and the source if it's visible.
End with a cleanup plan, quick wins first. Save it to /audits/[date] so next month's audit can show the trend.

Write the monthly sales and marketing report for [your company] for [month].
From HubSpot: new contacts, deals created, won and lost with amounts, [pipeline name] pipeline value now versus the start of the month, and the top lead sources.
Compare it with last month's report in /reports.
Write one page in plain English for leadership:
1. What happened, with the numbers
2. Why it happened
3. Three specific things to do next month
Save it to /reports/[year-month].

From skills to agents

A skill is you asking. An agent is the same job running for the whole team: started by a trigger, working across everyone's deals, with a named person checking the output.

Every skill above has a team version. Here's what changes when a skill grows up.

Signs a skill is ready to become an agent

• Three or more people run it every week.
• You've trusted the output for two weeks straight.
• There's a clear owner for the result.
• The job needs data that lives outside HubSpot.
• It stops happening when the person who runs it is out.

That's the work we do. We map how your team works today, build the first agent on the workflow with the biggest payoff, test it on your real work, and run it with your team. A person approves anything that sends, changes or deletes data. See how we build agents.

Guardrails: read freely, write carefully

• The assistant reads HubSpot freely. It never writes without your approval of that exact change.
• Emails and messages are always drafts. You hit send.
• No bulk updates, imports or merges without an export first.
• Anything hard to undo means stop and confirm.
• Customer data only goes into tools your company has approved.
• Start read-only for two weeks. Expand once you trust the output.

These are the same rules we set on every agent we build for clients.

Your first two weeks

The skills you used more than twice are your agent candidates.