What Actually Happens When Fathom's Meeting Notes Land in HubSpot

Fathom, the AI video note taker, plugs into HubSpot. It sits in your Zoom call, transcribes it, writes a summary with the action items, and pushes both into the CRM against the contact or deal.

That part works, and it is worth having. The interesting part is what it does not do, which is the bit you need to know before your team starts trusting the record.

What the integration actually puts into HubSpot

Install the AI Notetaker by Fathom app from the HubSpot marketplace and you get three things:

• The transcript. After the call, the full text of what was said goes to the contact or deal record.
• The summary. Fathom's AI writes a shorter version with the key moments and action items, so nobody has to read forty minutes of transcript to find the one thing that was agreed.
• Smart notes. Notes taken during the meeting save straight onto the CRM record instead of into somebody's notebook.

For a salesperson picking up a deal somebody else opened, that is a genuinely different experience from reading three lines typed at 6pm.

A note on the timeline is not a field in a report

Here is the thing that catches teams out, and it has nothing to do with Fathom being good or bad at its job.

The summary arrives as an activity on the record. Your pipeline reports, forecasts and dashboards do not read activities. They read deal properties. So you can have a perfect AI summary of the call where the client named their budget and their timeline, and your forecast still knows nothing about either, because no property changed.

Run one real meeting, then open the deal and ask a plain question: which fields are different from before the call? On most portals the answer is none of them. If you want the summary to reach your reporting, there is a walkthrough on copying Fathom's summary straight into your deal fields, and a related one on automating deal creation the moment a meeting ends so the record exists before anyone has to remember to make it.

Two more checks worth doing on that first test meeting:

• Where did it attach? Confirm whether the summary landed on the contact, the company and the deal, or only one of the three. Notes that live on a contact and not the deal are invisible to whoever is working the pipeline.
• Who did it match? Check the summary went against the right person. A client who dials in from a personal address, or a new person nobody has added to the CRM yet, is not the contact you think they are.

And it only helps if the meeting happens at all, which is a separate discipline. There is a piece on keeping meetings from falling through the cracks in HubSpot.

Before you turn it on: what your privacy notice needs to say

A transcript of a client call is personal information, and an AI note taker is collecting it. That puts this squarely inside the Australian Privacy Principles, and two of them matter here.

APP 3 covers collection. You collect personal information only where you need it for your work, and only by lawful and fair means. APP 5 covers notification: at or before you collect it, or as soon as practicable afterwards, you have to tell people what you are collecting, why, and who else will see it. The OAIC issued guidance in October 2024 on using commercially available AI products, and this is precisely the case it describes.

There is a second, sharper version of the problem. If health, financial or similar detail comes up on the call, and in a lot of client conversations it does, you may be recording sensitive information, which carries a higher bar than ordinary personal information.

So four questions, and they take one meeting to answer:

• Does your collection notice or privacy policy mention that meetings are recorded and transcribed by an AI tool? If it was written before you installed this, it does not.
• Who inside the business can read the transcripts once they are on the CRM record? In HubSpot, that is usually everyone with access to the deal.
• How long do they stay there, and who deletes one if somebody asks?
• Does the person running the call say it out loud at the start, every time?

None of that stops you using it. It stops the version of this where a client asks what happened to the recording of their call and nobody in the business knows the answer.

What still needs a person in the loop

The integration also works with HubSpot workflows, so something the AI picked up in a meeting can trigger an action: the classic example is Fathom reading that a client is ready for a proposal and HubSpot sending the proposal email or booking the follow up call.

Do the second half of that and be careful with the first. A workflow that creates a task for the account manager off an AI reading of one sentence is useful, and if it is wrong somebody notices in ten seconds. A workflow that emails a document with a price on it off the same reading is a bad idea, because the AI is interpreting a sentence like "we would probably be ready in the new year" and it does not know which of the four people on the call can sign anything.

The rule we use on this kind of build: anything that touches money, a contract or a customer's identity goes through a person. Everything else can fire on its own. That division is most of what our AI Engineering work ends up being, and it is a lot less exciting than the demo.

Where to start

Install it, run it on three internal meetings before it ever joins a client call, and fix the privacy notice in that window. Then take the two properties your forecast actually depends on and decide who is responsible for filling them in, because the AI writing a beautiful summary of the call does not change your numbers by itself.

Which raises the question worth sitting with. Open the last deal you won. If the summary of the conversation that won it is sitting in the timeline and nothing in your reporting knows it exists, what was the note for?