Zendesk's Auto Assist Now Reads Your Internal Notes. Here's What to Check.

On 2 November 2026, Zendesk rolled out a change to Copilot's auto assist: it now reads the internal notes on a ticket when it writes suggestions. Internal notes are where support agents write the version of events they would not send to a customer. That makes this a change worth five minutes of a support lead's time, whether or not you use Zendesk.

What changed

Zendesk's announcement, published on 30 October, opens: "Zendesk is glad to announce that auto assist now has access to the content of internal notes on a ticket, and might use that content when generating suggestions."

Under "What is changing?" it says: "Auto assist is now always aware of internal notes in tickets. It uses their information when generating suggestions, making itself more accurate to all processes that rely on internal notes."

Under "What do I need to do?" it says: "You don't have to do anything for auto assist to be aware of the internal notes. The content of internal notes is automatically included in auto assist's context on any ticket with internal notes."

The announcement also carries a caveat, and it belongs in any fair account of the change: "Note: Auto assist might decide to ignore some internal notes if they contain unhelpful or sensitive information."

What stands between a note and a customer

Auto assist does not reply to customers by itself. Zendesk's About auto assist article says suggestions "replace the composer" in the ticket, and that "Agents review and approve suggestions before they are sent, so agents remain in control and customers aren't aware of auto assist." The same article says auto assist "takes into account the sources listed above and all public ticket comments and internal notes in the conversation." It also says some actions, such as checking an order status, "may be performed automatically by auto assist if the action was pre-approved by you or another admin in a procedure."

So there are two safeguards between an internal note and a customer:

• The model's own judgement. The caveat says auto assist "might decide" to ignore notes that are unhelpful or sensitive. That decision is made by the model, case by case.
• The agent who approves the suggestion. Every suggested reply is reviewed by a person before it is sent.

The announcement doesn't describe a setting for keeping internal notes out of auto assist, and neither does the About auto assist article. We read both on 3 November 2026. If Zendesk adds or documents such a setting, it would change this picture, so check your own Admin Center before relying on this article.

Why internal notes are different

Public comments are written for the customer. Internal notes are written for colleagues, and they tend to hold things like:

• a refund or goodwill decision and the reason behind it (approve once, not again);
• an opinion about the customer (third chargeback threat this year);
• escalation context, including names of people and teams;
• details pasted in from other systems: order notes, account flags, sometimes personal data.

None of this was written on the assumption that it would shape a reply. Under the new behaviour it can. A note's content would not be sent to the customer word for word. But a suggestion built partly from that content can lean on it, and the agent who approves the suggestion is the last check.

What to check this week

• Read a sample of your own internal notes. Pull twenty recent tickets with notes and ask: would we be comfortable if a reply drew on this? If not, the issue is the notes, not just the AI.
• Agree what does not go in a note. Passwords, card details and opinions about a customer have never belonged in ticket notes. This change gives teams a concrete reason to write that rule down.
• Brief agents on approving suggestions. The approval step is the human safeguard. Agents should know that a suggestion can now reflect what a colleague wrote privately, and should read it with that in mind rather than approving on autopilot.
• Review pre-approved actions. Actions approved in a procedure can run without a per-ticket click. Check which ones you have enabled, now that the context behind a suggestion is wider.
• Ask Zendesk directly. If you need internal notes kept out of AI context, ask your account team whether that can be configured, and get the answer in writing.

How DeskLeap handles internal notes

In DeskLeap, internal notes are removed from a ticket's conversation before it reaches the AI Agent. Knowledge base articles marked internal are excluded too. This is how the product is built, not a setting you switch on. It also has a cost: our AI Agent can't use helpful context that an agent left in a note, so a team that wants that context used has to put it somewhere public, such as a knowledge base article.

Neither approach is free. Reading notes can make suggestions more accurate, which is Zendesk's stated reason for the change. Leaving them out keeps the place agents write candidly away from AI-written replies. What matters is that a support lead knows which approach their tool takes. As of 2 November, Zendesk's answer is that auto assist reads them.

On the same day, Zendesk also rolled out one-time passcodes for authenticating end users in messaging conversations with AI agents. The announcement says: "Authenticating users is essential for allowing AI agents to retrieve restricted knowledge and safely execute sensitive procedures." It is a separate feature for a separate product. It is worth knowing about because both changes are about what an AI may use, and when.

For a related change from the week before, see Zendesk Now Masks Personal Data in Ticket Comments. Here's What Agents Give Up.

DeskLeap is a ticketing, live chat and knowledge base platform. The platform is 100% free with unlimited human agents; an optional AI Agent add-on is $29/mo each.

Sources

Zendesk, Announcing internal notes access for Copilot auto assist (announced and rolled out 2 November 2026; published 30 October 2026; read 3 November 2026). Zendesk, About auto assist (updated 2 November 2026; read 3 November 2026). Zendesk, Announcing one-time passcodes to authenticate end users during messaging conversations with AI agents (rolled out 2 November 2026; read 3 November 2026). Quotations are verbatim from those pages. DeskLeap's handling of internal notes is described from our own code as of 3 November 2026.