Before you turn an agent loose
Twenty-three questions, five short steps
HubSpot keeps context on your business, your customers and how your team works, then runs agents on it. This checks whether that context is good enough to act on — before an agent writes to a record or emails a customer in your name.
- Five steps, four or five questions each. Business, customers, team and process, revenue if it applies, and governance.
- Every question has a five-minute check. A way to confirm the answer inside HubSpot rather than from memory.
- You get a score, a ranked fix list and an order to turn agents on in. Copy it, print it, or hand it to Claude to start fixing.
New to HubSpot? The words this uses
- Portal
- HubSpot's word for your account — the whole thing your company works in.
- Agent Hub → Context
- The part of HubSpot where you write down what its AI should know about your company, your customers and your team.
- Record
- One contact, company, deal or ticket.
- Property, or field
- One piece of information on a record, like Industry or Record owner. HubSpot calls them properties; most people say fields.
- Knowledge vault
- A set of documents you let HubSpot's AI read — product sheets, pricing, policies.
- Lifecycle stage
- The field that tracks how far along someone is, from lead to customer.
- Ideal customer profile
- The kind of company you sell to best, described specifically enough that you could filter a list by it.
- Agent
- HubSpot AI that does work on its own — researching, drafting, updating records, contacting people.
- Credits
- What HubSpot charges AI and agent work in. Different actions cost different amounts.
- Super Admin
- HubSpot's highest permission level, and the only one that can change AI settings for everybody.
Everything scores in your browser. Nothing is sent anywhere, and nothing is saved when you close the tab.
✓Your result
Where the score comes from
Fix these first
What to turn on, in order
| Agent | Verdict at your score | Credits per action |
|---|
Credits are what HubSpot charges agent work in, and these are its published figures for one action. Generating a piece of content costs roughly a hundred times what looking up a record does, which is why the cheapest agents to run are also the safest ones to start with.
Fix the context, not the score
Build your HubSpot AI Context with Claude
This prompt asks for your website, reads it, and then works through every section of Agent Hub → Context one at a time. It asks before it assumes, writes facts rather than marketing copy, flags anything confidential before it goes somewhere every AI tool can read, and will not move on until you approve the section in front of you. At the end it hands you the whole document, organised so you can paste each block into the matching HubSpot field. It opens in Claude with the prompt already in the message box, unsent, so you can read it before you start.
Read the prompt first
If copying doesn't work in your browser, select the text above and copy it by hand.
Mapped to the context HubSpot keeps in Agent Hub — Business knowledge, Customers, Team & process,
Personal, Knowledge vaults and its own Recommendations — plus the governance HubSpot doesn’t ask you about.
Section names under Business are taken from a real account; those under Customers and Team & process are our
best read and worth checking against yours. The band thresholds and fixes are Growth’s, drawn from
HubSpot’s published guidance and from portal work across mid-market implementations.
This tool does not read your HubSpot account; it asks you — and gives you a way to check each answer rather than guess it. A control that exists but has not been reviewed in over a year is counted as partial, because that is how it behaves in practice.
Growth Operations Firm · Elite HubSpot Solutions Partner · October 2026