If you went looking for the macro Suggestions tab in your Zendesk Admin Center recently and could not find it, you are not going mad. Zendesk retired it on 30 April 2026 and replaced it with something that lives somewhere else and costs extra. That is worth understanding before you plan any automation around macros, because most of the guidance still online describes a screen that no longer exists.
What changed in April 2026
On 30 April 2026 Zendesk added two new categories to Copilot Recommendations. One was trust and safety recommendations, covering things like two-factor authentication and inactive agent seats. The other was macro content suggestions, and the announcement is explicit that this feature replaces the legacy Suggestions tab on the Macros page in Admin Center.
The new version is a genuine improvement in one respect. It analyses how agents resolve tickets and proposes concrete edits to existing macros as well as suggesting new ones, and each suggestion links to the related tickets so you can check the evidence rather than take it on faith. The old tab only ever proposed new macros, which is how libraries end up with four macros that say nearly the same thing.
The catch is access. Macro content suggestions are for customers with the Copilot add-on. If you do not have that add-on, the legacy tab went away and nothing took its place. That is a real change to what a Zendesk plan gives you, and it is the reason this post exists: a lot of teams have quietly lost a feature they were mildly relying on, and the workaround is worth knowing.
What the old feature actually did
Worth recording, partly because it sets the bar for any replacement and partly because the constraints explain why macro libraries drift in the first place. The legacy suggestions were generated by a model that analysed agent replies, and the documented boundaries were fairly tight:
- Your account needed at least 150 tickets in the last three months before the model would suggest anything at all.
- It analysed replies from the email and webform channels only.
- It looked at public comments added by agents. Public comments from admins were excluded, and so were all internal notes.
- You got up to 10 suggestions a month, added on the first of the month from the previous month's data, with existing suggestions refreshed weekly.
- Supported languages were English, French, German, Japanese and Portuguese.
Read that list as a description of coverage and the gap is obvious. A busy support team writes replies across chat, messaging and social as well as email, and a lot of the institutional knowledge lives in internal notes. The suggestion engine, old or new, sees a slice.
The half nobody automated
Both versions of the feature answer one question: what new macro should exist? That is the easy half. The half that actually costs support leads their afternoons is everything after a macro is created.
| What macro suggestions handle | What still lands on a human |
|---|---|
| Spotting a repeated reply worth templating | Deciding whether it is worth templating, or whether the underlying issue should be fixed instead |
| Proposing text for a new macro | Checking it does not duplicate three macros already in the library |
| Proposing edits to an existing macro, with the new version | Noticing the macro references a policy that changed in March |
| Surfacing up to ten candidates a month | Retiring the forty macros nobody has applied in a year |
| Email and webform replies from agents | The knowledge sitting in internal notes and chat transcripts |
| Suggestions, one at a time | Whether the library as a whole still matches how the team actually answers |
The right-hand column is macro hygiene, and it is a recurring, rules-shaped, judgment-light-but-not-judgment-free job. Which is to say it is a good fit for an agent and a poor fit for a support lead's Friday.
What to hand an agent
"Clean up our macros" will produce a confident mess. Specify the inputs, the tests and what it is allowed to do, and it becomes a review queue you can trust.
| Element | What to specify |
|---|---|
| Input | The full macro list with usage counts and last-applied dates, plus a sample of recent tickets in the categories each macro serves |
| Duplicate test | Group macros by what they actually say, not by title, and report clusters that overlap rather than picking a winner |
| Staleness test | Flag macros not applied in ninety days, and separately flag any that reference a named policy, price, timeline or product name |
| Coverage test | Repeated replies in channels the native engine ignores, which in practice means chat, messaging and internal notes |
| Permissions | Read and propose only. It drafts, it does not create, edit or delete a macro |
| Escalation | Where it cannot tell whether a claim inside a macro is still true, say so rather than rewriting the sentence |
The permissions row is the one to hold firm on. A macro is text your team sends to customers under your name, and an agent with write access to the macro library is an agent that can silently change what your company says. Propose-only costs you a click per accepted change and removes the entire category of failure.
A working setup, end to end
Once a month the agent pulls the macro list with usage data, runs the four tests, and produces a single review document: duplicate clusters, staleness flags, policy-reference flags, and any repeated replies from channels Zendesk's own engine does not read. A support lead works through it in one sitting, accepting or dismissing each item, and makes the actual edits in Zendesk.
The reason this works as a monthly job rather than a continuous one is that macro drift is slow. Running it weekly produces a mostly empty report and trains people to ignore it. If you want the pattern for turning this into a job that fires on its own rather than one someone remembers, writing a prompt for a recurring agent covers the scheduling side, and AI agent security best practices covers scoping the Zendesk credential so read-only really is read-only.
This sits alongside the rest of the support work an agent can take on. Zendesk ticket triage handles the inbound side of the same queue, the best AI agents for customer support compares the platforms that do this end to end, and for teams weighing whether to connect an agent to Zendesk at all, connecting an agent to a private API covers the mechanics without over-scoping the token.
What goes wrong
We run a version of this against our own support macros, and three failures recur.
The first is title-matching. Two macros called "Refund request" and "Refund, standard" may be identical or may differ on the one line that matters. An agent grouping by title will merge things it should not. Group by the body text and report the cluster for a human to resolve, rather than asking the agent to decide which survives.
The second is over-eager staleness flagging. A macro used twice a year is not dead if those two uses are a renewal window. We tested a flat ninety-day rule and it flagged genuinely useful seasonal macros every single run, which is how a report becomes noise. Adding "flag, do not recommend deletion" and letting the human see the usage pattern fixed it without any cleverer logic.
The third is the important one: an agent reading a macro cannot verify that the policy inside it is still correct. It can only notice that the macro makes a checkable claim. That distinction is the difference between a report that says "this macro cites a fourteen-day return window, confirm that is current" and one that quietly rewrites it to thirty days because that seemed more standard. Insist on the first.
The honest limit: none of this replaces the support lead who knows why a macro is worded the way it is. It replaces the part where they scroll through two hundred macros trying to remember which ones are still true.
Frequently asked questions
Where did the Zendesk macro Suggestions tab go?
Zendesk replaced it on 30 April 2026. The legacy Suggestions tab on the Macros page in Admin Center was superseded by Macro content suggestions, which now appear inside Copilot Recommendations. The announcement states directly that the new feature replaces the legacy tab, so if you cannot find the old screen, it has been removed rather than moved.
Do I need the Copilot add-on for Zendesk macro suggestions?
Yes, for the current version. Macro content suggestions are described as launching for customers with the Copilot add-on. Accounts without it lost the legacy Suggestions tab without gaining the replacement, which is why some teams experienced the April 2026 change as a feature disappearing rather than improving.
How did Zendesk decide which macros to suggest?
The legacy engine required at least 150 tickets in the previous three months, then analysed public comments added by agents on the email and webform channels, excluding admin public comments and all internal notes. It offered up to ten suggestions a month, added on the first of each month from the prior month's data, with existing suggestions refreshed weekly, in English, French, German, Japanese and Portuguese.
Can an AI agent replace Zendesk's macro suggestions?
It can cover the parts Zendesk's feature never did, which is usually the better trade. Suggesting new macros is the narrow half. An agent can additionally find duplicate macros, flag ones that have not been applied in months, flag ones making claims that may have gone stale, and read channels the native engine ignores. Treat it as macro hygiene rather than as a like-for-like replacement.
Should the agent be allowed to edit macros directly?
No. Macros are text sent to customers under your company's name, so give the agent read access and let it propose changes into a review queue. A human accepts or dismisses each one and makes the edit. The cost is a click per change, and it removes any possibility of the agent silently altering what your team tells customers.
How much does it cost to run this?
Considerably less than the Copilot add-on if macro hygiene is the only reason you were considering it. On Gravity the free tier covers one agent at $0 a month, which is enough for a monthly review job on a single Zendesk instance, and paid plans start at $20 a month with $20 of usage included, with extra usage available beyond the plan. For how that compares across the category, see our roundup of the cheapest AI agent platforms.
Sources
- Zendesk. "Announcing macro content suggestions and trust & safety recommendations." support.zendesk.com, announced and rolled out 30 April 2026, checked 24 August 2026. Source for the replacement of the legacy Suggestions tab, the Copilot add-on requirement, the ability to propose edits to existing macros as well as new ones, and the linking of related tickets as evidence.
- Zendesk. "Creating macros from macro suggestions for admins" and "Why don't I see macro suggestions?" support.zendesk.com, checked 24 August 2026. Source for the 150-ticket threshold over three months, the email and webform channel scope, the exclusion of admin public comments and internal notes, the limit of ten suggestions per month with monthly addition and weekly refresh, and the supported languages.
- Gravity pricing: free tier with one agent at $0 per month; paid plans from $20 per month including $20 of usage, with extra usage available beyond the plan. Checked 24 August 2026.
- Related reading: Zendesk ticket triage, best AI agents for customer support, prompts for recurring agents, what an AI agent is.