Get your CRM ready for HubSpot's Prospecting Agent
This one is for the RevOps leads, the HubSpot admins, and anyone who switched on a Prospecting Agent play and watched it suggest a contact your team already knows, at a company that churned last year.
What: Using Breeze Assistant to build a Prospecting Agent Readiness Plan. It checks the segments, personas and records the agent now uses to choose who to contact, sorts your contacts into readiness tiers, and tells you who fixes each gap before the agent spends credits on the wrong people.
Prompt of the week:
The Prospecting Agent has changed a lot this year. Early in 2026 it mostly drafted outreach for contacts you already had. Since HubSpot's Spring Spotlight in April and the August update, it runs inside plays. Each play targets company segments, watches for intent signals, finds up to three contacts per company that match your personas, and can pull new contacts from a connected data partner. It is now on every paid hub, it uses HubSpot Credits for each lead it writes to, and at UNBOUND in September HubSpot rebuilt Breeze Assistant to carry out tasks rather than just answer questions.
That shifts where bad data hurts. Messy records used to mean a bland email. Now they can mean the agent picks the wrong account entirely. A loose segment sends it after companies you would never sell to. An out-of-date ICP score ranks the wrong accounts first. A missing suppression flag lets it approach a competitor's employee, a customer halfway through a renewal, or someone who told a rep no last spring. And because the agent only weighs roughly the last year of engagement, older warnings in notes and closed-lost reasons may not count unless a field or list keeps them in force.
There is a new duplicate risk too. When the agent sources contacts itself, your duplicate rules have to catch what it brings in, not just what arrives through imports. And the agent sends from the contact owner's connected inbox, so every contact whose owner has no inbox falls back to a single sender.
HubSpot has added tools that help. Context Home scores how complete your business context is for AI, and the Data Model Health check flags messy properties, pipelines and objects. Neither tells you which records and segments are safe for the agent to work, what to do about the rest, or who should do it.
This prompt produces it.
Prompt structure
Paste this into Breeze Assistant with CRM data access on. Fill in the brackets, and paste your Context Home and Data Model Health findings where shown.
Before you paste it, check that CRM data, customer conversation data and files data are switched on in your AI settings, since the agent needs all three. It is also worth reviewing Context Home and setting feature-level credit limits for the agent first.
ROLE: You are a HubSpot RevOps architect preparing a portal for the
Prospecting Agent as it works in late 2026: plays that target company
segments, intent-signal monitoring, contact sourcing from a data
partner, and credit-based outreach.
GOAL:
Produce a Prospecting Agent Readiness Plan. It must tell me which
records and segments the agent can safely work now, what blocks the
rest, and who fixes each gap. I want decisions and owners, not only
percentages.
MY PORTAL:
- Company and what we sell: [ ]
- Hubs and tiers: [e.g. Sales Hub Pro, Marketing Hub Enterprise]
- Teams that will run plays: [Sales / Marketing / CS]
- Plays live or planned (segment, signals, personas, Sequence or
Adaptive, Review before sending or Send automatically): [ ]
- External data source connected to the agent: [Apollo / Surfe /
ZoomInfo / Seamless / none]
- Enrichment available in Smart CRM: [yes / no / unsure]
- Monthly credit budget for the agent: [ ]
- Markets and languages we sell into: [ ]
- Approx. contacts / companies: [ ] / [ ]
- Context Home findings: [paste]
- Data Model Health findings: [paste]
- Suppression rules outside HubSpot (spreadsheets, other tools): [ ]
WORK THROUGH THESE AREAS:
1. Targeting inputs
For each play, check the company segment definition, the personas
and the ICP-fit score or equivalent. Say whether each reflects the
ICP above, is outdated, or is too broad to control spend. Estimate
how many companies each segment contains.
2. Fields the agent depends on
Map the contact and company properties that affect targeting,
personalisation and sending. Include at least: job title,
seniority, persona, company name and domain, industry, employee
count, revenue, country, preferred language, time zone, contact
owner, lifecycle stage, lead status, last engagement date,
associated deals, and any custom ICP fields. For each, give:
- Role: TARGETING, PERSONALISATION, SENDING or SAFETY
- What goes wrong in agent output if it is blank or messy
- Fill rate and consistency, or SIGNAL MISSING if you cannot see it
- Who should maintain it: enrichment, workflow, rep, or admin
3. Record readiness tiers
Put contacts (with their companies) into:
- GREEN: targeting, sending and safety fields sound. Enrol now.
- AMBER: one fixable gap. Give the fix and whether a workflow or a
person must do it.
- RED: several gaps. Enrich, research or leave out of plays.
- STOP: must not be enrolled until a human reviews. Includes
opted out, unsubscribed, customers under an active contract,
competitors, open deals owned by a rep, closed-lost without a
re-engagement reason, and likely duplicates.
Give counts and percentages, the top five gaps behind AMBER and
RED, and a filter definition for each tier that I can save as a
list.
4. Duplicates and sourced contacts
The agent can bring in new contacts from our data partner. Check
duplicate rules on email, domain and company name. Flag where a
sourced contact could duplicate an existing record or land at a
company that is already a customer or in an open deal. Recommend
the rule or workflow that prevents this.
5. History and suppression
The agent considers roughly the last year of engagement. Identify
contacts whose relevant history is older than that, such as a
closed-lost reason or a do-not-contact note, and recommend a
property or list that keeps them out of plays.
6. Sending and localisation
Count contacts whose owner has no connected inbox, and suggest a
fallback sender per team. Check preferred language and time zone
coverage for each market above.
7. Enrichment and spend
For RED records, split by ICP fit. Recommend which to enrich,
which to research by hand, and which to exclude. Estimate credit
use per play from segment size, the three-contacts-per-company
limit and the daily suggestion limit. Say what enrichment trigger
should run on new records.
RULES:
- Never recommend deleting records. Default to suppress or archive,
and note that deletion needs a retention review.
- Do not recommend enriching STOP records.
- Flag any bulk change to emails, phone numbers or consent fields as
needing a data protection review.
- Separate admin fixes (workflows, rules, properties) from rep
judgement on individual records.
- For every consistency fix, show one before and after example of
how the value would change the agent's email.
- If you cannot see the data for a number, write
SIGNAL MISSING: [what to check] instead of estimating.
OUTPUT:
1. Summary: three sentences. Current readiness, the biggest blocker,
and the first action.
2. Targeting inputs: table of play, segment, persona, issue, fix.
3. Field map: table of property, role, impact if bad, fill rate,
owner.
4. Tiers: table of tier, count, %, main reason, saved-list filter.
5. Duplicates and suppression: table of pattern, volume, action,
owner, urgency.
6. Sending and localisation: gaps and fixes.
7. Enrichment and spend: table of segment, ICP fit, records, action,
estimated credits.
8. 30 / 60 / 90 day plan with owners: RevOps, Sales Ops, reps,
Marketing, CS, Data protection, Sales leadership.
Why this prompt works - and how to adapt it
Most readiness checks still ask whether the agent can write a good email. With plays and contact sourcing, the first question is whether it should be contacting this person at all. This prompt audits targeting and safety before it looks at copy.
A few things to note about how it is constructed:
Targeting comes before copy. With plays and contact sourcing, a loose segment or outdated ICP score costs credits and reputation before a single email is weak. Area 1 makes the Assistant check the inputs that drive spend.
Fields are grouped by what they do. Tagging each property as targeting, personalisation, sending or safety shows why a gap matters. A missing industry gives a bland email. A missing do-not-contact flag gives a complaint.
STOP is the hard gate. The agent will not know about a verbal no from last spring, a competitor's employee, or a customer mid-renewal unless a field or list tells it. Build that list before anything else.
Sourced contacts create a new duplicate risk. Before August, duplicates mostly came from imports. Now the agent can add records itself, so duplicate rules need to catch what it brings in.
The one-year window matters. Old closed-lost reasons and notes may fall outside what the agent weighs. A property or list keeps that history in force.
SIGNAL MISSING stops guessing. The Assistant can read your CRM but not the SDR team's spreadsheet or a tool that isn't connected. Where it can't see something, it should say so.
Adapting it for your portal:
Running plays on Send automatically? If nobody reviews drafts, every safeguard has to be a rule. Add: "We send without review. Turn every STOP condition into automatic list or workflow criteria, and tell me which gaps make Send automatically unsafe for each play." The output becomes a set of exclusion rules rather than a review queue.
Selling to a named account list? Readiness matters per account, not per contact. Add: "We target [N] named accounts. Assess readiness per account, flag any with fewer than [N] contacts across our personas, and check whether contact sourcing is filling those gaps." The tiers shift to account level.
Letting Marketing or CS run plays? Now the agent is on every paid hub, two teams can end up emailing the same person. Add: "Marketing uses plays for inbound follow-up and CS for expansion. Check that ownership, lifecycle stage and suppression rules stop teams contacting the same person in parallel."
Selling in several countries? Language, time zone and consent all vary by market. Add: "We sell in [countries]. Audit preferred language, time zone and country fields per market, and flag consent differences for each region."
No data partner connected? The agent can only source contacts through one external provider at a time, and you may not have one. Add: "No data partner is connected. For ICP-fit RED records, list the fields a rep can fill in under three minutes from public sources, and fit that work into daily prospecting." The plan becomes something reps can do without extra tools.
Re-running each quarter? Save the output and come back to it. Add: "Compare with the audit from [date]. Show tier movement, which issues were fixed, which came back, and credit spend per play since then." That turns a one-off clean-up into an ongoing health measure.
Beyond the prompt:
The readiness plan shows you where things stand. The order you act in matters as much as the findings.
Start with the STOP list. Save every STOP condition as an active list and use it to exclude or unenrol contacts across all plays, through workflows or play settings. It takes an afternoon, and it prevents the kind of email that ends up being discussed in a leadership meeting.
Next, fix one play's targeting before touching the rest. Pick your strongest segment and persona, tighten the segment definition, update the ICP score and keep the daily suggestion limit low. A narrow play that works tells you more than five broad ones that burn credits.
Run that play on Review before sending for the first few weeks, on GREEN records only. Let reps see and edit the drafts. Move it to Send automatically only once drafts rarely need changes, and only for segments where the STOP rules are all automated.
Then work through AMBER systematically. Workflows can standardise job titles, countries and industries. Leave reps a short weekly queue for the fixes that need a person's judgement.
Leave the RED decision until last. By then you will know what a good play produces, so you can judge which accounts are worth enriching or sourcing contacts for and which should simply stay out of plays.
Re-run the prompt each quarter. Track GREEN as a share of contacts in your target segments, alongside credits spent per meeting booked. The agent can safely take on more once the records and rules around it are sound.