The HubSpot Prospecting Agent: what it does, and what it costs

What HubSpot released

Two changes this year made HubSpot’s prospecting agent worth a second look.

On 14 April 2026, HubSpot moved the agent to outcome-based pricing. It had charged a recurring monthly fee per enrolled contact. It now charges $1 each time it recommends a lead for outreach.

On 16 September 2026, at its Fall 2026 Spotlight, HubSpot expanded what the agent watches and finds. It now monitors more than 40 buying signals, assembles buying groups at the accounts showing them, and drafts outreach for each contact.

How it works

The agent runs from plays. A play is a set of instructions for one persona, product or segment: which signals to watch, who to target, what to say, and what not to say. HubSpot’s documentation, updated 16 September 2026, describes this sequence:

• It watches for signals. Company news such as funding rounds, leadership changes and product launches, plus research activity and visits to your site.
• It reads your history with the account. Up to a year of engagement in HubSpot: form submissions, page views, calls, meetings, notes and email opens.
• It finds contacts. From your existing CRM records, or from one connected data provider at a time: Apollo, ZoomInfo, Surfe or Seamless.
• It recommends leads, up to a limit you set. Each play has a daily new contact suggestions limit. It may suggest fewer, because suggestions depend on which companies are showing signals that day.
• It drafts the outreach. Multi-step AI sequences and adaptive emails, written to the play’s instructions.

Two details matter when you set it up. Changes to a play apply only to contacts enrolled after the change, so a weak first play keeps running on everyone already in it. And the agent cannot be tested in a sandbox account, so the first real run is in production.

What it costs

The agent runs on HubSpot Credits, which cost $0.01 each. A recommended lead uses 100 credits, so $1. Credits come in packs of 1,000 for $10, or on pay-as-you-go. Seat-based subscriptions include a monthly allowance, shown in each account.

The daily limit is the budget control. A play capped at 20 suggestions a day, running every working day, can recommend about 440 leads a month, which is $440 if every suggestion is made. Three plays at that cap is $1,320.

The billable event is the recommendation, not a reply or a meeting. HubSpot’s Chief Customer Officer described the change as “you pay when it works.” What it measures is that the agent found a lead that cleared its own threshold. Whether that lead meets your definition of qualified is a separate question, and the one that decides whether the $1 was well spent.

Review mode or send automatically

Each play can run in one of two modes. In review mode, the rep approves each email before it goes, with in-app notifications or a daily email digest. In send automatically mode, the agent sends without review.

Start every play in review mode. The edit rate is the most useful number the trial produces: if reps rewrite most drafts, the play’s instructions are wrong, and sending automatically would put those drafts in front of buyers at scale.

Automatic sending also carries the deliverability risk to the rep’s own mailbox. Google’s sender guidelines ask senders to keep spam complaints below 0.3%, ideally below 0.1%, and warn that mail missing the requirements may not be delivered or may be marked as spam. A play that generates complaints costs more than the replies it gets. The setup that protects a domain is in the cold email deliverability checklist.

HubSpot’s performance claims

HubSpot’s product page claims 76% more qualified leads, 80% more meetings booked and a 26% higher deal win rate. It does not state the baseline, the time period or how the customers were selected. HubSpot also reported that agent activation grew 57% quarter on quarter before the pricing change.

Treat the claims as a reason to test, not as a forecast. The figures that hold up for AI in outbound are narrower: time saved on research, drafting and admin. We collected the ones that trace to a named source in the 2026 AI outbound benchmarks.

Does it replace an SDR?

No. It takes over the parts of the SDR job that are research and assembly: noticing that an account raised money, finding the three people who will care, and writing a first draft that mentions it.

It does not make calls, run discovery, handle objections or judge whether a conversation is real pipeline. Those are where SDR teams win or lose. At Locus, MQL to SQL conversion went from 10% to 25% without more leads or new tools. The change was a written definition of qualified, enforced at stage exit. An agent that recommends more leads does nothing for that number. A clear qualification standard does.

The agent is the AI layer applied to the targeting and outreach activities, the same argument as AI is a layer, not another tool. The three activities pipeline depends on are mapped in pipeline comes from three activities working together.

What a good play looks like in practice

A play is an ICP, a persona and a qualification standard written for a machine. The agent can only be as specific as those three are, so here is what specific looks like on real engagements.

The ICP in one number. For Easyship, the accounts worth a conversation are eCommerce businesses shipping a thousand parcels a month or more. A play built on a line like that gives the agent something to match against. “Mid-market eCommerce brands” does not. The method is in the ICP framework.

Qualification as named criteria. For WorkStore, every inbound lead was qualified on four criteria: budget, genuine need, timeline and the number of units required. Unit count was the one that most often separated a real opportunity from an enquiry. Write a play’s guardrails the same way, as criteria a rep could check, not adjectives. How to choose them is in the sales qualification framework.

Foundations before volume. At Landmark Group’s Logistiq division, the ICPs, personas, CRM workflows and sequences were the foundation the founding inside sales team worked from, and the division generated $500K ARR in its first quarter. An agent needs the same foundation: the play comes first, the daily limit second. The full record is on Shobhit Gupta’s profile.

How to run the 14-day trial

HubSpot gives first-time users 14 days of free access on Sales Hub Professional or Enterprise. It starts when you publish your first play, uses no credits, and switches the agent off when it ends. Fourteen days is short, so decide what you are measuring before you publish.

• One play, one persona. Pick the segment where your ICP is clearest and your current reply rate is known.
• Write the exclusions. Existing customers, open opportunities, competitors and any region you do not sell into.
• Review mode only. Every draft goes through a rep.
• A modest daily limit. Enough for reps to review every draft properly.
• Four numbers. The share of recommendations reps accept, the share of drafts they rewrite, the reply rate against your current baseline, and meetings booked.

If reps accept most recommendations and send most drafts with light edits, the play is worth paying for. If they reject the leads, the play’s targeting is wrong. If they rewrite the emails, its instructions are. Both are fixable, and both are cheaper to find in the trial than after it.

Who should turn it on

Turn it on if you already run HubSpot, your ICP is written down, and your SDRs spend more time researching than talking. The trial costs nothing and the per-lead price makes the cost easy to cap. Getting the CRM foundations right first is HubSpot and Salesforce setup.

Fix the basics first if your CRM has no agreed stage definitions or your reps disagree on who a good account is. The agent will follow the play you give it, at volume. A pipeline audit shows whether the foundations will hold.

It is not a reason to change CRM. Salesforce teams have a comparable route in Salesforce in Claude. The wider choice is in HubSpot versus Salesforce, and the dedicated AI SDR products are compared in the AI BDR platform map. Apollo’s equivalent is covered in Apollo’s AI agents, and HubSpot’s Claude connector in Claude for sales teams.

References

• Fall 2026 Spotlight. HubSpot, 16 September 2026. Prospecting Agent monitoring 40+ buying signals, assembling buying groups and drafting outreach.
• HubSpot’s Customer Agent and Prospecting Agent: now you pay when the task is complete. HubSpot, 14 April 2026. $1 per recommended lead, 57% activation growth.
• Use the prospecting agent. HubSpot Knowledge Base, updated 16 September 2026. Plays, signals, data providers, daily limits, sending modes.
• Opt in to 14 days of free access to the prospecting agent. HubSpot Knowledge Base. Eligibility and trial terms.
• HubSpot Credits. HubSpot. Price per credit, packs and pay-as-you-go.
• AI Prospecting Agent for Sales Teams. HubSpot. Product page and performance claims.
• Email sender guidelines. Google Workspace Admin Help. Spam rate thresholds and delivery consequences. All accessed 2 October 2026.