
Turning on HubSpot AI notetaker for a sales team comes down to four settings decisions before the first call is recorded: who gets access, how the bot joins, which deal properties it may update without review, and where a person has to approve. Get one of them wrong and the tool produces summaries no rep trusts. As one step in an ai automation for business plan, the feature puts a transcript, a follow-up task and a coaching record against every qualified meeting.
This guide is for the person who owns the sales process, not the person who builds the product: a VP of Sales setting rules for a team, or a RevOps admin who will run the configuration. It assumes the team books discovery and demo calls through HubSpot and holds them on Zoom, Microsoft Teams or Google Meet. A team that sells mostly by phone, over WhatsApp or face to face gets little from the feature, because the bot joins video meetings only.
The minimum is a paid Sales Hub or Service Hub plan at Professional or Enterprise level, with a Sales or Service seat assigned to every user who should record calls. Super Admins control the account-level switch, and a user with the Add and edit users permission can change who holds which permission, so decide who owns both administrative roles before anyone opens Settings. One Super Admin account is enough to start; a second admin prevents a support queue in the first week.
Three checks decide whether the first meeting records. Each rep connects their own calendar, and no admin can do it for them. Zoom hosts switch local recording on so the call saves to the host's computer. Google Meet hosts either admit the bot when it asks to join or set meeting access to Open, and Microsoft Teams needs no extra installation.
Ask for the seat list before the first toggle. Users without a Sales or Service seat cannot be granted the feature, and a licence bought after the rollout starts needs the same access step, so keep the ai automation seat list and the access list in one document that a manager can check.
Eligibility is narrower than most teams expect. A meeting has to be created or stored in HubSpot, carry at least one CRM contact whose email domain differs from your company domain, and have Google Meet, Microsoft Teams or Zoom set as its location. That domain rule quietly excludes internal standups, which is the point: the tool records conversations with customers and prospects, not team meetings.
Recurring meetings need an opt-in to the Sync Recurring Meetings to HubSpot beta. Without it the bot skips a series even when every occurrence looks eligible on the calendar.

Consent and eligibility decide whether a recording exists, so settle both before the account-wide switch. Behaviour differs by channel, and a rep should know which one applies before the call.
On Google Meet and Microsoft Teams the bot receives recording permission as soon as it joins, then plays an audio message and posts in the meeting chat a few seconds later. On Zoom the host sees a pop-up and must approve recording before anything is captured. The chat notice is posted once, so anyone who joins late never sees it unless the host pins the message.
The default comes with an audio message, a chat message and a pop-up, and an admin can choose between a spoken and chat announcement with a language and voice, or a chat message only. A custom bot name helps customers recognise who is in the room. Test the announcement on an internal call with the settings you plan to ship, because the wording is what a customer hears before they decide whether they mind.
Explicit consent is a setting, not a habit. When an admin turns it on, HubSpot emails each invitee before the meeting, and if any participant declines or does not reply, the meeting is not recorded. That is the cleanest option for teams in regulated sectors.
A waiting room gives the host a second gate. When the host joins, HubSpot shows a notification that the bot is waiting; if the host denies access, it never joins, and if it waits more than five minutes without being admitted, it leaves. Either route protects the customer, as long as the host knows the rule on the day.
Retention matters more than most teams assume. Recordings are kept for two years in HubSpot's file manager, which is longer than many internal policies allow for a customer call. Match the setting to the retention rules your business already applies to call notes, and tell participants what happens to the file.
Work through the steps in order. Each one ends with something you can see in the account, so a rollout that stops halfway leaves a visible gap rather than a silent one.

Most rollouts here stumble on governance while the settings work as designed. The team has not agreed what the settings mean.
Granting access to every eligible seat is fast and usually right, but it also gives the feature to contractors, former staff and shared accounts that still hold a seat. Audit the user table before the toggle, not after. In an SME with a small admin team, the person who buys seats often sits away from sales, so the list needs a sales owner to sign it off.
Automated suggestions feel efficient until a wrong call outcome lands in a forecast. Keep automation off for the properties that feed pipeline reporting, and let a rep approve those changes. The credit meter is the other half of the decision: a wide set of custom properties on automatic multiplies the run cost without improving the summary a manager reads.
A bot that joins without a clear announcement damages trust faster than a missed summary costs time. Set the announcement, decide on explicit consent, and tell reps what to say when a customer asks. If your team handles personal data under Singapore's PDPC guidelines, the notice, the purpose and the retention period should match the policy you already publish, because a recording is personal data like any other.
Teams that run first conversations on WhatsApp or a mobile line, then move a deal to HubSpot later, will find the notetaker absent from the calls that matter most. Check where discovery actually happens before promising a manager that every call will be captured.
Verification belongs in a weekly ai automation routine. Four checks tell you whether the feature is doing what the plan promised.
Open a recent meeting record and look for the recording, the transcript, the summary and the suggested follow-up in one place. If the summary sits on the meeting but no task or draft email follows, the automation stops at the wrong point. Check also that manual fields are still waiting for approval rather than drifting into the deal unread.
Compare the credit count against the number of transcripts and the custom properties you enabled. Each custom property costs 10 credits per transcript run, so a jump with no matching call volume points to a property someone added without telling the team.
Run a consent audit on a sample of calls. A meeting with no recording should show a decline or a host who kept the bot out, and a meeting with a recording should show a notice that a customer could hear or read.
Start a coaching sample. The Calls Index can be filtered by user, team, call outcome and duration, and a manager who reviews one or two calls per week with each rep gets more from the tool than one who skims every recording. Start with the summary and open the transcript only where it points to a moment worth coaching, such as a missed discovery question or a weak answer to an objection.

The notetaker answers a narrow question: what was said. An ai automation layer answers the question a sales leader actually cares about, which is what happens to the deal next, and that is the layer Topkee builds.
Topkee's Custom AI Agent Development service is scoped as a permissioned workflow rather than a general chatbot. The agent reads authorised material, produces an output, saves a record or submits an item for approval, and a person decides at defined checkpoints. Its published enquiry and follow-up workflow runs in five stages: receive the enquiry, compile a requirement summary with any missing budget, timeline or requirements flagged and draft a follow-up in brand voice, have the salesperson review and confirm the draft, write the result back to the CRM with the owner, the next step and a processing record, then sample real tasks to fix data gaps. The metrics it reports are first-response time, data completeness and the overdue-follow-up ratio, and the service states plainly that it measures process improvement rather than promising a conversion or revenue lift.
That covers what the notetaker leaves open. The summary exists, but nothing enforces a next step, and nothing checks whether the CRM reflects what the buyer said. Topkee's delivery principles put the guardrails in the same place: sources traceable, external sends and budget or order changes behind an approval, outputs accepted against pre-agreed samples, and a handover that includes the configuration and training. HubSpot appears in the knowledge base as one of the CRMs the agent can connect to, not as a product-level integration with the notetaker, so the sensible split is to let HubSpot capture the meeting and let a permissioned agent carry the record into the follow-up. Published reference pricing starts at a single-scenario build with one standard system connection and a monthly maintenance tier, which is a cost benchmark for the work beyond the subscription, not a quote for your account.
Write down what should happen within an hour of a qualified call: who owns the follow-up, which fields must be current, and what a manager will check. That one page tells you which parts of the record the notetaker can produce and which parts need a permissioned agent and a person.
If your team already automates outreach, the same discipline applies to sequences; LinkedIn automation in HubSpot sequences shows how the toggles behave before you add another channel, and how AI agent traffic changes an SEO plan covers the visibility question once your meeting notes feed public content.
Yes. It joins meetings on all three platforms without an extra app, and each channel has its own route to recording permission, from the host's approval pop-up on Zoom to automatic permission on Google Meet and Microsoft Teams.
Yes. A rep can clear auto-join in their own calendar settings, and the meeting scheduler and the activity timeline both offer a per-meeting invite toggle. A host can also remove the bot during the call from the video platform itself.
They sit on the meeting record that HubSpot ties to the related CRM records, together with the recording, the transcript and the suggested property updates. A manager who opens one record sees what was said next to the deal it belongs to.
The bot announces itself through an audio message, a chat message and a pop-up. An admin can add a pre-meeting consent email, and when that setting is on, a decline or a missing reply keeps the meeting unrecorded. A waiting room adds a second gate, and the bot leaves after waiting more than five minutes without being admitted.
It is available on Sales Hub and Service Hub Professional and Enterprise plans, and every user who records calls needs an assigned Sales or Service seat. A user without a seat cannot be granted access, even after the account-level switch is on.