HubSpot Breeze Prospecting Agent: A Practical Guide

Prospecting is the part of sales most teams know they should do well and rarely have time to do consistently. Researching accounts, watching for the right moment to reach out, finding the correct contact and writing something that does not sound like a template all take hours that reps would rather spend in conversations. HubSpot's Breeze Prospecting Agent is built to take on that groundwork. HubSpot describes it as a tool that "watches for buying signals, sources contacts, and personalises outreach, automatically", which is a fair summary of what it does day to day.

If you run sales on HubSpot, this is a genuinely useful addition, and it is worth understanding properly before you switch it on. This guide covers what the Prospecting Agent is, where it performs well, how to get real value from it, and the honest line where a custom Claude agent does work it was never designed for. For the wider picture of how the Breeze family fits together, our overview of AI for HubSpot sets the context. Here we go deep on prospecting specifically.

What the Breeze Prospecting Agent is

The Prospecting Agent is one of HubSpot's autonomous Breeze Agents. It sits alongside the Breeze Assistant, the in-app conversational AI (Copilot) that answers your questions grounded in your CRM, and Breeze Intelligence, which handles data enrichment and buyer intent. The distinction matters. The Assistant responds when you ask it something. The Prospecting Agent is autonomous: it monitors your data continuously and takes multi-step actions within guardrails you set, without waiting for a prompt each time.

Work is organised into what HubSpot calls "plays". When you create a play, you choose which buying signals the agent monitors and refine its research and outreach strategy for a specific audience segment and persona. The agent then continuously evaluates companies in your chosen segments against those signals. According to HubSpot's Knowledge Base, the signals it watches include job postings, funding rounds, technology adoption, leadership changes and engagement or visitor spikes, plus configurable research signals (by keyword or company domain) and visitor intent signals that show companies are researching topics you care about.

When a company trips a signal, the agent gets to work. For list building and enrichment it draws on connected third-party data providers (HubSpot names ZoomInfo, Apollo and Surfe) to find and enrich contacts with phone numbers, emails and job titles. It then drafts personalised outreach emails grounded in the triggering signal, account research and a "selling profile" you configure, which holds your company information, tone preferences and calls to action such as meeting links, documents or URLs. It can also create reminders for non-email touches such as calls or LinkedIn messages, and multi-language drafting is in beta.

You choose how much autonomy to grant. In "review before sending" mode a rep approves each draft before it goes out. In "send automatically" mode the agent sends on its own, which is the setting teams tend to move to once they trust the output quality. HubSpot's product page cites customer results of 14% more qualified leads, 14% more meetings booked and 5% higher deal win rates. Those are vendor-reported figures, so treat them as HubSpot's own numbers rather than an independent benchmark, but they point in a sensible direction.

Where it shines

The Prospecting Agent is strongest exactly where it should be: turning data you already hold in HubSpot, plus signals it can see across the web, into timely, relevant outreach. For a sales team already living in HubSpot, that is a real advantage. The research, the contact sourcing and the first draft all happen in the system reps work in, so there is no copying between tools and no separate prospecting platform to learn.

It is also good at consistency. Account research considers engagements from up to the past year associated with contact records, including form submissions, page views, calls, meetings, notes and email opens. That means the agent can spot that an account has been quietly active and surface it before a rep would have noticed manually. Pairing a fresh external signal, such as a funding round or a leadership change, with that internal engagement history is the kind of join-up that good prospecting depends on and that busy reps routinely miss.

The play structure deserves credit too. Rather than a single blunt setting, you can tune separate plays for different segments and personas, each watching the signals that matter for that audience. That makes the agent useful for a team selling several products or into several verticals, not just a single motion. And because it runs natively inside Sales Hub, the outputs land where pipeline already lives, so reps can act without leaving their workflow.

HubSpot also reports that Prospecting Agent customer activations were up 57% quarter over quarter, which suggests adoption is moving in the right direction. For many mid-market sales teams, this is a sound way to get autonomous prospecting working without a heavy build.

Getting real value from it

Most teams who switch the Prospecting Agent on underuse it. The output quality depends almost entirely on the configuration you give it, and that is where the work sits.

Start with the selling profile. This is the agent's brief: your company information, the tone you want, and the calls to action it should offer. A thin or generic profile produces thin, generic emails. A specific one (clear positioning, a real sense of how you talk, the right meeting link or document for each play) produces drafts a rep is happy to send. Treat it as you would a brief to a new SDR.

Next, get the plays right. Choose signals that genuinely indicate buying intent for each segment rather than switching everything on. A keyword research signal that is too broad will flood you with weak leads, and because costs run on HubSpot Credits and accrue per lead recommended for outreach, a noisy high-signal segment can consume credits faster than it returns value. Tighter signal selection is both better prospecting and better economics.

Understand the guardrails, because they shape what the agent will and will not do. Per HubSpot's Knowledge Base, there is a 1,000-emails-per-day account send limit; no more than three agent-sent emails to a single contact over a rolling 90-day period without a strong engagement signal; the agent prioritises the first three sends then monitors weekly to adjust; and it researches a maximum of 10 contacts at once with the rest queued. These are sensible limits that protect deliverability and stop the agent over-contacting people, and they are worth explaining to your team so nobody expects unlimited volume.

Finally, decide on autonomy deliberately. Begin in "review before sending" so reps can see the quality and correct the selling profile and plays based on real drafts. Move to "send automatically" only once the output is consistently good. The Spring 2026 outcome-based announcement notes a free 28-day trial on the repriced agents, which gives you a sensible window to tune the configuration before committing.

Where a custom Claude agent goes further

The Prospecting Agent is deliberately scoped. It reasons over HubSpot-resident data, its connected providers and the web, and it works from HubSpot's signal catalogue and its own outreach workflow. Within that scope it is well designed. The limits are simply the edges of that scope, and they are worth being honest about.

It does not natively reason over data that lives outside HubSpot. If the strongest buying signal in your business is product usage, billing or finance history, support tickets or something sitting in your data warehouse, the Prospecting Agent cannot see it. Its signals come from HubSpot's predefined catalogue plus the research and visitor signals you configure, not from a bespoke scoring model trained on your own win and loss history. Enrichment depth also depends on the coverage of the connected providers, so niche or under-covered accounts can come back thin. And account research looks back roughly a year, which is fine for most cases but limiting if your buying cycles are long.

This is where a custom Claude agent earns its place, not as a replacement but as a complement. A bespoke build can score prospects against your actual win and loss data and your real ICP rather than a fixed signal set. It can pull cross-system context (product usage, billing, support history, warehouse data) into the qualification decision. And it can apply bespoke routing and qualification rules, your choice of model and stricter governance where you need it. HubSpot's Run Agent workflow action lets a custom agent run side by side with Breeze, so the two operate together rather than competing. Breeze handles the prospecting motion it is built for; the custom agent handles the cross-system judgement it is not. This same do-both pattern applies across the suite, including the Breeze Customer Agent on the service side.

How SpotDev helps

SpotDev is a HubSpot Diamond Partner, so we implement, configure and optimise the Breeze Assistant and the Breeze agents to get real value from HubSpot's own AI. We are also a Claude specialist with in-house engineers (nothing subcontracted) who build custom Claude agents for the work Breeze was never scoped to do, wired cleanly into HubSpot. We have delivered more than 300 technology projects across the two disciplines.

Because we do both, we can give you a straight answer that a Breeze-only or a Claude-only shop cannot. Often the Prospecting Agent, configured properly, is genuinely enough, and we will tell you so. Sometimes the value depends on data and logic that live outside HubSpot, and a custom build pays back. We help you tune the agent first, then identify where a bespoke Claude agent adds something the agent cannot, and build it to run alongside Breeze rather than against it. If that is the conversation you need, talk to a Claude-specialist engineer.

Frequently asked questions

How much does the Breeze Prospecting Agent cost?

HubSpot moved the agent to outcome-based pricing on 14 April 2026. According to HubSpot's company-news announcement, the billable outcome is a lead recommended for outreach, priced at 1.00 US dollar per lead via 100 HubSpot Credits, replacing the previous recurring monthly charge per enrolled contact. The agent requires a Sales Hub subscription (HubSpot lists Starter, Professional and Enterprise) and the repriced agents include a free 28-day trial. Credits are consumed when a lead is enrolled for prospecting.

Will it send emails to prospects without my approval?

Only if you choose that mode. The agent runs in either "review before sending", where a rep approves each draft, or "send automatically", where it sends on its own. Most teams start in review mode and move to automatic once they trust the quality. HubSpot also applies guardrails: a 1,000-emails-per-day account limit and no more than three agent-sent emails to one contact over a rolling 90-day period without a strong engagement signal.

Can it use data from systems outside HubSpot?

Not natively. The Prospecting Agent reasons over HubSpot-resident data, its connected providers such as ZoomInfo, Apollo and Surfe, and the web. It does not natively read product usage, billing, support or data-warehouse systems. If those hold your strongest buying signals, a custom Claude agent can bring that context into qualification and run alongside Breeze using HubSpot's Run Agent workflow action.

Is the Prospecting Agent the same as the Breeze Assistant?

No. The Breeze Assistant (Copilot) is conversational and responds when you ask it something, grounded in your CRM. The Prospecting Agent is autonomous: it monitors data continuously and takes multi-step actions within the guardrails you set. They are different parts of the same Breeze family and are designed to work together.

Work with a Claude specialist

SpotDev designs, builds and deploys custom Claude agents and enterprise Claude rollouts for UK businesses, with fixed packages from £8,000 to £45,000 and a first rollout live in two to three weeks. Explore our Claude implementation packages or talk to one of our engineers.