Most teams expect HubSpot AI agents to clean up their CRM. In practice an agent reads the CRM the way it is, and every empty close date, every unlogged call and every duplicate contact becomes part of its answer. The agent cannot tell which records are wrong.
So the useful question is not which agent to buy. It is what the agent will inherit from your portal and what to fix first. That holds for HubSpot’s own agents and for outside ones, including ours.
What HubSpot AI agents actually read in your portal
HubSpot’s own guide to setting up an account for AI features lists what generative AI can access. That is CRM data such as contacts, deals and tickets, conversation data such as call transcripts and emails, and files. HubSpot’s agents, collected in Agent Hub under Agents > Agent Hub, also use the business context you add: company messaging, ideal customer profiles and knowledge vaults.
Put simply, an agent works from three things: fields, logged activity and the context you gave it. If a deal has no next step, no logged email and an owner who left in spring, the agent sees exactly that.
Who this matters for
- Heads of sales who want an agent to flag risky deals. The flag is only as good as the activity reps log.
- RevOps and CRM owners who will be asked why the agent recommended something odd. Usually the answer is an empty field.
- Founders and budget owners. Plan requirements differ by agent: HubSpot’s customer agent and knowledge base agent need a Professional or Enterprise subscription, while the prospecting agent and custom agents can also run on Starter, and most agent work draws HubSpot Credits. An agent on a messy portal costs twice: the subscription, then the hours spent checking its mistakes.
How to prepare HubSpot for AI agents: what to check in your portal
These are HubSpot’s own recommendations, in the order we would do them.
- Decide who may switch agents on. Configuring AI settings needs Super Admin or Partner admin permissions. Agents without built-in permissions use access settings, for example owners and super admins can edit and run. Write down who owns each agent and who reviews its output before anyone turns one on.
- Look at your data quality first. Open Data Management > Data Quality. The data quality tools show duplicates, formatting issues and property insights in one place. Start with duplicate contacts and companies, then with deal properties that are rarely filled in.
- Merge duplicates. Use the duplicate management tool and merge records so there is one record per person and per company.
- Make the important fields required, and archive the dead ones. HubSpot suggests required properties on record creation, property validation rules for consistent formatting, and archiving properties with low fill rates. Every field an agent reads is a field it may misread, so fewer, cleaner fields mean fewer wrong conclusions.
- Log the activity. Send or log emails in HubSpot, record calls, and give activities clear, descriptive titles. This is the evidence an agent cites, and what is not logged does not exist for it.
- Add your business context. Enter your ideal customer profiles, products and services, buyer profiles and selling profiles in the Breeze context settings. An agent that scores leads against an ICP nobody wrote down scores them against nothing.
Where an agent helps once the basics are in place
Once the data is honest, agents are good at the work people skip: checking every open deal for a buyer who went silent, finding deals that stopped moving, flagging empty fields, sorting new tickets. We built the Alignify agents for that work. The part we care about most is control. Every agent starts in review mode, each finding shows the emails, meetings and fields it is based on, and proposed updates wait for an admin to approve or reject them. Only after a reviewed run and an approved update can a team schedule an agent and choose which fields it may change. The platform is in private beta.
Our take
Most HubSpot problems are process problems, and AI does not change that. An agent mirrors your portal the same way a report does, only faster and with more confidence. If your sales stages have no exit criteria and reps do not log calls, an agent will produce fluent recommendations built on gaps.
The honest cost is in hours, not licenses. Someone has to own the cleanup, the required fields and the agent’s output every week. If the answer to “who owns this” is nobody, fix the data and the owner first, then switch the agent on. A cleaner portal pays off even without an agent.
Private beta
Let an agent run this check in your HubSpot
Alignify agents read the deals, emails and meetings behind every record, show the evidence and wait for your approval before they change anything.
FAQ
How do I prepare HubSpot for AI agents?
Start with the data the agent will read, not with the agent. Run the six checks above, then turn on one agent, on one object, for a few weeks. Review what it proposes before you let it change anything, and add the next agent only when the first one stops surprising you.
What data do HubSpot AI agents use?
According to HubSpot, generative AI can be given access to CRM records, conversation data such as call transcripts and emails, and files. Agents also use the business context and knowledge vaults you add. What you have not logged, the agent cannot use.
Do I need a clean CRM before using AI agents in HubSpot?
Not perfect, but honest. An agent can work around some gaps and even point them out, but duplicates, missing activity and empty fields lead to confident answers that are wrong. Fix the fields your agent will rely on first.
This week, open Data Management > Data Quality and look at duplicates and property insights for deals and contacts. If the list is long, read our note on what to clean up first before you turn on any agent.
Alignify
On this page
Get the audit checklist
All 25 checks as a spreadsheet with auto-scoring.
Talk to Alignify
45 minutes, your portal on screen, no slides. We show what we would fix first.