On a paid Claude plan (Pro, Max, Team or Enterprise), add HubSpot from Customize > Connectors, sign in to the right HubSpot account and set its write tools to Blocked for the first trial. Then ask tightly scoped questions about deals, owners and recent activity, and request a source-linked daily call list to check before anyone calls.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowThe paid-plan rule is HubSpot’s own: Claude’s pricing page lists connectors on every plan including Free, but HubSpot’s guide requires a paid Anthropic subscription. The connector can also create and update records, which a pipeline question does not need. Keep the initial output in chat, reconcile it with HubSpot, and allow specific writes only after the team has shown that its selections are dependable.
How do you connect Claude to HubSpot safely?
To connect Claude to HubSpot safely, pick the official HubSpot connector, grant only the permissions your pilot needs, and block write tools for the first trial.
HubSpot’s guide, updated 17 November 2026, lists the connector for all HubSpot products and plans and asks for an active HubSpot user account alongside the paid Claude subscription. One Claude account connects to one HubSpot account at a time, so anyone who manages several portals has to disconnect and reconnect to switch. You do not need to buy a separate HubSpot AI agent merely to ask CRM questions.
In Claude’s connector-management area, find HubSpot and select the official listing, which Anthropic marks as verified and made by HubSpot. Connect, choose the HubSpot account, and on the page headed “Connecting HubSpot Connector for Claude to HubSpot” select only the permissions the pilot needs before clicking Connect app. HubSpot’s instructions reach this through Settings and Connectors; Claude’s current guidance uses Customize > Connectors. Use the entry shown in your interface.
HubSpot says Super Admins and users with App Marketplace Access can connect without prior app approval; other users need a Super Admin’s approval. The administrator controls approved data permissions and who may install. On Claude Team or Enterprise, the relevant owner also needs to enable the connector. These are separate gates, so identify which administrator needs to resolve a blocked setup.
Block write tools during the initial trial. HubSpot recommends setting Write tools to Needs Approval once updates are permitted, and warns that Always allow lets edits happen without asking. Its guide does not mention a second switch: the permission setting in Claude’s message box. Anthropic is making Auto the standard setting for Pro, Max, Ultra and Team subscriptions; under it, a tool set to Needs Approval skips the prompt unless the automated check raises a flag, so pair Needs Approval with Manual. The write side is wide. HubSpot lists create and update rights on contacts, deals, tasks, notes and pipelines, and its marketing tools can create and even publish website content. The first call-list pilot should not change deal stages, create contacts or clean up fields merely because the model notices incomplete records.
Verify the connection using one known deal. Ask for its identifier, pipeline, stage, amount and owner, then compare them with the original record. Do not start with an account-wide forecast. A single-record check makes it easier to spot the wrong HubSpot account, missing permissions or an unexpected interpretation of your request.
Allow an illustrative half-hour for permissions, account verification and the first known-deal test. Record subscriptions separately from staff time, using your actual account charges rather than an assumed bundle price. If you need broader CRM structure first, use the small-team HubSpot setup tutorial before adding a conversational reporting layer.
How do you define which deals a pipeline question covers?
A pipeline question should cover a defined cohort: state the pipeline name, included stages, responsible team, date field and date range before asking Claude. A pipeline is the sequence of sales stages a deal moves through. A cohort is the specific set you are analysing. “This month’s sales” could mean deals created this month, deals expected to close this month or completed sales recorded this month. Those produce different answers.
For an illustrative landscaping company, the owner chooses open garden-maintenance deals in the main service pipeline, owned by two salespeople, with expected closing dates in the next 30 days. The owner excludes completed and lost deals. This is an internal reporting definition, not a claim that every business should use the same stages or time window.
Ask Claude to state the filters it applied and identify missing fields. Require source records or identifiers beside actionable conclusions. If a result depends on engagement history, such as calls or notes, check that those records are accessible. HubSpot says that where its Sensitive Data settings are on, the connector sees Sensitive Data properties and engagements only if a Super Admin granted that read-only access during installation or reauthentication, and an older connection may need disconnecting and reauthenticating before engagement history appears. Do not treat an access gap as proof that no contact occurred.
Keep totals separate from interpretations. A returned amount is a recorded deal value. A likelihood of closing is an estimate unless tied to a defined method. Avoid requesting a confident forecast before checking data completeness. The owner’s first useful question may be how many relevant deals lack a next step, rather than how much revenue the business will win.
Ask ten questions that lead to a sales decision
1. Which open deals have no recorded next action?
Ask: “Within the defined cohort, list deals without a recorded next action, including owner, stage, amount and source record. Distinguish a missing field from an explicitly cancelled action.” This reveals administrative gaps without deciding that every blank field deserves an immediate sales call.
An illustrative result shows four deals with no next action. One has a recent note saying the customer requested a pause; another has no supporting context. Put the paused deal in a later-review queue and ask its owner to clarify the second. Do not automatically create four chase tasks. The missing-field count is useful, but the next action still needs context.
2. Which proposals are stale by our own definition?
Define stale before asking. For the illustration, the manager chooses proposals with no recorded customer-facing activity for ten days. This threshold is a management rule, not a HubSpot default or a universal sales benchmark. Request the last relevant activity and its source, with unavailable activity marked as unknown.
An illustrative deal has a $2,400 proposal and no logged call for 12 days, but a note records an agreed follow-up next week. Keep that agreement visible and avoid chasing early. Another proposal has no activity or agreed pause and merits owner review. The question should expose candidates for attention rather than label every older proposal as neglected.
3. What is due to close soon without a clear next step?
Ask for deals whose expected close date falls within a defined period and which lack a documented next step. Include the expected date and last available customer interaction. Tell Claude not to change any dates or stages. The output is a review list for the salesperson, not an instruction to make the reporting numbers look better.
A removals firm’s illustrative $1,800 deal is expected to close in three days, yet the inventory review remains outstanding. A sensible next action is to arrange that review if contact is appropriate. It is not to mark the deal as won or move the close date automatically. Check the actual operational dependency before choosing the call’s purpose.
4. Where are amounts or owners missing?
Request separate lists for missing amount, missing owner and both missing. Treat zero as different from blank unless the business has explicitly defined them as equivalent. Ask Claude to return identifiers and the field values it saw. This makes the proposed data-quality work reviewable without granting permission to fill gaps from guesses.
An illustrative tutoring centre has a trial-session deal worth $0 and an annual tuition enquiry with no amount recorded. The first may be intentional; the second may need a quotation. Assigning the same estimated value to both would distort the pipeline. Have the responsible person establish the actual amount and approve any later update.
5. Is the team’s workload uneven?
Ask for open deals and due follow-up actions by owner within the same cohort. Include both count and recorded value, but do not equate either with workload. Some jobs need detailed assessments while others need a brief confirmation. Ask the manager to identify unusually demanding cases before recommending reassignment.
For an illustrative two-person team, one salesperson has eight deals and the other ten. That does not establish an unfair split. If the eight include six site assessments and the ten are routine renewals, the first person may have more work. Use the report to start a specific scheduling conversation, not to rank effort from deal counts alone.
6. Which deals need information from the customer?
Ask Claude to identify missing customer inputs evidenced by notes or tasks: inventory details, scope confirmation or a preferred appointment. Require supporting records. An absent property does not prove the customer has failed to respond. Mark the distinction between an explicit outstanding request and a field that staff have not maintained.
In the landscaping illustration, one note says the customer will supply planting measurements. The proposed call purpose can be “confirm whether measurements are ready”. Another record merely has an empty measurement field. Its owner should first check whether the information arrived elsewhere. A good call list avoids making the customer repeat information the business already received.
7. Which customer conversations should I read before calling?
Ask for a concise pre-call brief for a specified deal: latest available conversation, unresolved questions, promises recorded by the team and the source for each point. HubSpot’s connector supports engagement context including calls, meetings, notes and emails where permitted. Ask for the specific evidence instead of a broad personality profile of the contact.
An illustrative bookshop wholesale enquiry includes a request for a revised delivery schedule. Claude’s brief also calls the customer “price-sensitive”, but no source supports that description. Remove the label. Prepare a call about the delivery question and the accepted commercial details. Do not let inferred traits become permanent CRM notes or influence how the team treats the person.
8. Which records might represent duplicate opportunities?
Ask for possible duplicates as a review queue, using business identity, deal context and available identifiers. Do not merge, delete or overwrite anything. Similar names are leads for checking, not proof. Two departments of one customer may have legitimate separate opportunities, and repeat service requests may concern different periods.
An illustrative removals customer has a home move and a storage enquiry. A superficial duplicate check could collapse them into one deal. The corrected queue explains why they may be related but distinct. Have a salesperson inspect the records and decide whether any change is needed. Keep the connector read-only while establishing that judgement process.
9. How does the recorded pipeline value break down?
Request total recorded value by stage and owner, with the deal count, missing amounts and currency identified. Do not sum unlike currencies into one total without an approved conversion method. Ask Claude to show the included records or a checkable breakdown so the owner can reconcile the result with the same filtered view in HubSpot.
Suppose an illustrative subset contains $4,000, $2,500 and $1,500 deals. The recorded total is $8,000. If a fourth deal has no amount, say “$8,000 recorded value plus one unvalued deal”, not “the pipeline is worth $8,000” without qualification. An arithmetically correct subtotal can still mislead if it conceals missing information.
10. Who belongs on today’s short call list?
Ask for a manageable number of candidates after applying exclusions and the owner’s priorities. Request one reason to call, one clear purpose, the responsible salesperson and the source record for each. Separate calls from internal checks and other communication. A person awaiting a document may need a corrected document before any call is useful.
Set an illustrative capacity of five calls rather than requesting every possible prospect. At 12 minutes per call including preparation and notes, that is a one-hour block. The capacity limit should reflect available staff time. A ranked list of 40 contacts is not a daily plan for a salesperson with an hour free.
If you want help fitting this into how your sales team actually runs, I offer one-to-one AI consultations for small businesses.
How do you build a short daily call list from open deals?
To build a short daily call list, drop deals that are paused, awaiting internal checks, already scheduled or lacking contact context, then have Claude rank the rest, customer-requested callbacks first. Each call needs a deal ID or source link to check against the live record.
Return to the illustrative landscaping company’s 18 open deals. Two are on agreed pauses, three need internal quotation checks, four already have future follow-ups and four lack sufficient contact context. That leaves five candidates for the morning call block. These categories are deliberately non-overlapping in this exercise; verify real records rather than subtracting overlapping counts.
Of the five, two need missing scope information, two have unaddressed proposal questions and one has a customer-requested callback. Ask Claude to put the requested callback first, then rank the remaining calls using the owner’s chosen business priorities. Do not let recorded deal value override a specific customer promise or an outreach restriction.
Use HubSpot read-only. Apply the cohort and exclusions above.
Build at most five calls for today's one-hour block.
For each: deal ID, source link if available, owner,
evidenced reason, call purpose, latest relevant activity,
contact eligibility evidence and any missing information.
Put customer-requested callbacks before speculative follow-ups.
Exclude agreed pauses and contacts our team must not approach.
Keep internal data checks in a separate list.
Do not create tasks, change records or contact anyone.
Review the proposed list against the live records immediately before use. An earlier customer reply may have resolved the question. A teammate may have made the call after Claude retrieved the information. Stamp your working list with the retrieval time and have the caller check current activity so the team does not confuse a snapshot with a live queue.
The tutorial on building a daily call list covers the broader selection exercise. Here, the CRM evidence is the distinctive requirement: every reason for contact should be traceable to a record, and any missing eligibility information should stop the contact from entering the ready-to-call group.
How do you check who may be called before ranking deals?
Before ranking calls by commercial value, check that each contact may receive the planned outreach, using the permitted communication and exclusions the responsible person has defined. A contact in a CRM is not automatically appropriate for every kind of outreach, so that person should also flag any rules requiring specialist advice. Check applicable calling and marketing obligations with a qualified adviser where needed. This tutorial supplies an operational review, not a universal legal permission to call.
Use three groups: eligible for this contact, excluded from this contact and insufficient information. Keep the third group separate from the first. A missing restriction field is not proof of permission. Ask the CRM owner how the business records customer preferences and requests to stop contact, then use that accepted process as an exclusion before scoring opportunities.
For an illustrative failure test, include a high-value deal whose customer has asked for no further sales calls. The list should exclude it even if its expected close date is tomorrow. If Claude ranks it first because of value, revise the selection instructions and verify the source fields. Do not repair the list silently and call the trial successful.
Call preparation should contain enough context to help the salesperson, without copying unnecessary personal information into a shared list. A brief stating “customer requested revised scope” may be sufficient. Avoid exporting full conversation histories to a team-wide document when only the assigned salesperson needs the detail.
Record the outcome after each accepted call using the business’s normal CRM process. The call-logging and HubSpot tutorial addresses a connected route for call records. Do not assume a call happened just because it appeared on Claude’s list. A proposed action and a completed activity are different records.
When should Claude be allowed to write to HubSpot?
Claude should be allowed to write to HubSpot only after several accurate read-only runs, and then for one limited write that serves a real need, such as creating an approved follow-up task. Set write tools to Needs Approval, keep the chat in Manual, inspect the precise record and values, and verify the result in HubSpot. Keep deal stage, amount and contact-preference changes outside that first write test. Three HubSpot rules help here: every create and update made through the connector is recorded in the account’s Audit Log against both the user and the Claude connector; bulk creates or updates are limited to 10 records at a time; and the connector cannot delete records, so removing a mistaken entry is done in HubSpot itself.
Include failure recovery in the test. If a task-creation request times out, search for the intended task before repeating the operation. Otherwise a retry may create duplicate work. Keep an explicit record of what was approved and what HubSpot actually shows; a conversational confirmation alone does not establish the final source-system state.
Measure usefulness over an illustrative five-day trial: selected calls, exclusions caught, missing-context cases, incorrect selections and time spent reviewing. Suppose manual list preparation takes 20 minutes daily and the checked Claude route takes 12. Across five days that is 40 minutes, or $20 at a hypothetical $30 hourly internal rate, before setup and subscriptions.
Accept the workflow only if the team can reconcile the cohort, explain every selected call and keep excluded contacts out. A smaller list with clear reasons is more usable than a longer list assembled from guessed intent. Keep the ten questions as separate tools for management decisions rather than running them all every morning regardless of need.
If you log the five-day trial in a spreadsheet, The Complete Excel Course with AI Integration on Coding Liquids covers that side.
Further reads
- HubSpot vs Pipedrive for AI Sales Automation in a Small Business — Compare the underlying CRM choice.
- How Much Does HubSpot's Prospecting Agent Cost for 200 Leads? — The credit sum for 200 leads a month on HubSpot's Prospecting Agent: allowances by plan, top-up options and what each reply or meeting costs.
- Which Claude Connectors Should a Small Business Switch On? — Nine Claude connectors by job (email, calendar, files, accounts, CRM, payments, jobs, meetings, contracts), with what each sees and changes.
Browse all AI-for-business tutorials
Sources: HubSpot Knowledge Base, Set up and use the HubSpot connector for Claude; Claude Marketplace, HubSpot; Claude pricing page; Claude Help Center, Use connectors to extend Claude's capabilities and Get started with Claude Cowork. Verified 18 November 2026.
