AI agents in HubSpot work with the data in your CRM. To get reliable results, first decide which agent needs which data, then clean up exactly that data: properties, duplicates, required fields, and knowledge sources.
If you have used a HubSpot portal for years, you know the picture. The same company sits in the CRM three times, each with a different owner. Half of your contacts have no industry. In daily work, your team fills these gaps with what it knows. An agent works only with what is in the record.
Each agent draws on different data. According to HubSpot (as of October 2026):
In practice, this means an outdated owner, an empty industry field, or a knowledge article with old prices flows straight into the agent's work. The quality of its results follows the quality of these fields. Our page Breeze & Agents shows how to set up Breeze agents.
Five signals quickly show where a portal stands:
Each signal hits a specific agent. Duplicates and empty industry fields weaken audience selection in prospecting. Outdated knowledge articles lead to wrong answers from the Customer Agent.
You make your data AI-ready in five steps: define the first use case, review your data model and properties, merge duplicates, update your knowledge sources, and set up ongoing maintenance.
First decide which agent you will use first. That tells you which data matters. For the Prospecting Agent, it is contacts, companies, industry, owner, and interactions. For the Customer Agent, it is your knowledge content. Our article on AI workflows in HubSpot describes which AI processes pay off in sales and service. This way, you clean up with a clear focus and work through your portal step by step.
For each piece of information, decide which property holds it. If several fields cover the same thing, pick one, move the values over, and archive the rest. Replace free-text fields for industry or role with dropdowns that have fixed values. This makes segments and rules for agents unambiguous.
In Property insights, HubSpot shows how properties are used across your CRM. There you hide or archive fields nobody uses.
HubSpot automatically deduplicates contacts by email address and companies by primary domain. Companies created through the API are excluded from this domain check. You review all other duplicates in the duplicate management tool and merge them there or reject the suggestion. Existing customers need a Professional or Enterprise edition for this, and Data Hub Professional or Enterprise to handle many duplicates at once.
You fix inconsistent spellings under Formatting issues. There you accept suggestions or set rules for automatic correction.
Before the Customer Agent answers, review its sources: knowledge base, website pages, files, and question-and-answer pairs. Update or remove outdated articles, and exclude individual URLs from syncing. Keep personal, confidential, and customer data out, because content from the sources can appear in answers.
Clean data stays clean when someone owns it. Name one person responsible for the data model, set required fields in forms and pipeline stages, and plan a fixed review rhythm. HubSpot's data quality overview can send you a weekly summary by email for this.
You find further requirements such as governance, training, and success metrics in our article on the 6 requirements for AI with HubSpot Breeze.
HubSpot groups its data quality tools under Data Management > Data Quality (as of October 2026). The area includes an overview with recommended actions, duplicate management, formatting issues, data enrichment, and property insights.
Which features you can use depends on your pricing model:
Breeze helps with data maintenance too. The Data Agent (beta) fills properties with AI-generated answers, transforms data, and works as an AI step in workflows. It uses HubSpot credits. Our article HubSpot AI: From Useful Tool to Real Growth Driver covers more of the AI features in daily work.
The tools show where things get stuck. You decide which fields your processes need, who maintains them, and which values apply.
An audit pays off when your portal has grown over years, several teams and systems feed data into it, or a larger AI rollout is coming up. With a young, manageable portal, the HubSpot tools usually take you far on your own.
In the HubSpot Data Quality Audit from Mark Lotse, we review your portal in two to three weeks across five areas: data structure and properties, duplicates and consistency, segmentation, processes and reporting, and data sources and data flows. You get a prioritized list of actions with effort estimates, split into quick improvements and structural fixes. The audit costs €3,500 net for one portal. If you commission an optimization project within eight weeks, we credit the audit fee toward it. This gives you a clear picture of your portal and the order of next steps before your first agent goes live.