AI

Hi,For us, the main reason to explore AI Agents was to see how we could handle repetitive support requests more efficiently while still keeping the customer experience personal.Before experimenting, we focused on a few important things:1. Clear goalsWe first identified which type of requests would actually benefit from AI. The goal wasn't to automate everything, but to reduce repetitive work and give agents more time for complex conversations.2. Good documentation and knowledgeWe reviewed our Help Center content and made sure the information was accurate and easy to understand. This was one of the biggest lessons for me—AI needs good knowledge to give useful answers.3. Clean processes and workflowsWe looked at existing ticket fields, tags, triggers, and automations before introducing anything new. This helped avoid adding AI on top of an already complicated workflow.4. Team alignmentIt was important that everyone understood what AI would handle and when a conversation should be passed

Hey Community! Today we’re hosting a Digital Desk webinar on Admin Copilot, and I’d love to keep the conversation going here!For those using Admin Copilot, I’d love to hear what’s working for you:How are you using it in your instance? Any helpful insights, recommendations, or cleanup opportunities you’ve uncovered? Any tips you’d share with other admins?We’re also hosting the webinar again on July 23rd, so I’ll link the registration here for anyone who is interested!Looking forward to learning from this thread!- Emma CollinsSr. Content Manager, CX Programs

Hello All, I’m new in this community, and thanks for accepting me first.A quick question, one of my customer wants to use Zendesk AI to autoanswer per chat or per mail but doesn’t want to expose Zendesk Knowledge base making articles public.So is it possible or maybe can Zendesk develop a feature to for each articles define :Fully Public Internal but can be use by AI to answer publicly  Internal only (For Agents)Thanks

I am editing the Escalation Reply in Advanced AI Agents. The system-built one is just one box that escalates to a human. I have made the below flow instead so that the customer gets some feedback after the escalation is submitted. I have noticed that this flow does not update the requester field to reflect the email address entered by the user in this flow. I think that I can use Parameters and then Configured Actions to update the user’s data but I don’t have enough knowledge to be sure. Below is what I think I need to do but would appreciate anyone’s input on this!Add a new “If...” under Customer Message that says “Entity Recognized” > “Email” Select to Collect Parameter “email” Configure an action to Update User with the email fieldScreenshots are below. Does this seem right? Thanks! 

Hi folks!We have a website that sells our product (nothing groundbreaking here!) and we want to use AI Agents to engage people who are on the website, specifically the product pages; ask them questions and recommend a product based on the answers. We have an AI Agent that I’m building out, which mostly helps with post-purchase, but I’m wondering if there’s a way I can also have the agent behave a certain way with customers on those specific web pages. I’m thinking potentially we could have Proactive Messaging which adds a user tag and triggers a specific Use Case which is built around product recommendation? Has anyone tried that, or anything similar? Thanks!

On Friday - August 14th, our custom agent workspace stopped loading:  We had 6 custom agents, with 2 active. Any help would be appreciated. We are getting a 500 (Internal Server Error). 

We have 4 brands set up in Zendesk, each representing a different application we offer. Some customers use multiple applications/brands, but all support requests come into a single general support email address connected to Zendesk.We're trying to set up Advanced AI for Email to determine which brand a customer's email relates to, using this logic:First, check the customer's initial email message to see if they explicitly state which brand/application they're referring to. If so, treat that as the brand and follow the corresponding segmented condition. If the brand isn't stated in the email, fall back to a custom organization field called "Org Application Contracted for" and use that field's value to determine the brand.I tried building this with segments in the dialogue workflow, but it seems Zendesk isn't pulling in the organization field value before it evaluates the segment condition — it just falls through to the fallback path in the use case dialogue workflow rather than matching

Hi everyone,I’m curious how folks are organizing their Zendesk knowledge bases when balancing human agents and AI agents.Much of the advice around deploying AI agents suggests that the knowledge you expose them to directly dictates their scope and response quality. If AI knowledge needs to be tightly controlled to ensure accuracy, does it make more sense operationally to maintain a single, permissioned KB or split them up? For those who are already down the path of deploying AI agents, what setup has worked best for you?Unified KB: Sharing one KB across human and AI agents, restricting what the AI sees via permissions or labels? Dedicated KB: Creating a separate knowledge base tailored specifically for AI consumption? Would love to hear how you’re managing this!

Hi , I hope you are doing. One habit that has saved a lot of time is taking a few minutes to review my setup before creating a new automation. It's easy to jump straight into building something new, but a quick review often prevents duplicate work and keeps the instance much cleaner.Here are the five things I always check:1. Look for an existing automationBefore creating anything new, I search to see if a similar automation already exists. Sometimes a small update to an existing one is all that's needed.2. Check related triggersAutomations and triggers often work together. I make sure a trigger isn't already handling part of the workflow to avoid unexpected behavior.3. Review ticket fields and tagsI check whether the automation relies on the correct fields and tags. Reusing existing fields usually keeps reporting cleaner and avoids unnecessary complexity.4. Keep the logic simpleIf an automation has too many conditions or actions, I stop and ask whether it can be simplified. Simple work

Welcome to the next installment of Zendesk Voices, our series highlighting the expertise of Zendesk customers and partners.Have the needs of your front line team changed a lot in the age of AI? What feels different about equipping agents with the right tools now?We’d love to hear about:How your front line team’s needs have changed What tools matter most now What feels harder, easier, or more important in the age of AI How you help agents stay ready and supported Whether you see changes in results, feedback, or bothHelpful tips:Be specific and share real examples. Tell us what changed, why you made the change, and what happened next. Both numbers and stories are welcome. Keep it clear, honest, and useful for others. Please share your own voice and experience. We want to hear your perspective, not a response written by AI.Share your story right here in our community, then visit the Spark reward hub for full details and to make sure you are eligible for rewardsOnce we review and approve y

We are testing Knowledge Co-Pilot but we are finding that the ticket references it is using are inaccurate.  Usually only one ticket refers to the issue.  We have gone through the tickets to see if key words were triggering the response but we haven’t found any.  They are usually completely different and unrelated.  We also went through several examples and each time we found the same result.

To migrate to Zendesk means moving tickets, contacts, and knowledge base content into a new environment where Zendesk's AI (Intelligent Triage and Advanced AI) starts reading your data immediately. The data determines what the AI learns, and it starts learning the moment records land in the new instance.The main challenge is data quality, not data volume. Zendesk's AI runs a continuous pattern-matching loop on whatever it's given. Feed it three years of unresolved tickets, outdated pricing references, and low-rated interactions, and it confidently learns the wrong patterns. Because a full historical dump teaches AI your worst-case interactions as well as your best, teams split the migration into two ordered phases rather than moving everything at once.When an AI-ready migration to Zendesk makes senseTeams sequence their Zendesk migration around AI readiness in situations like:1) Enabling Intelligent Triage on day one: triage accuracy depends on help center freshness and clean intent si

The Zendesk Product Team is looking to speak with customers about how you manage organizational change today and gather feedback on an early Change Management concept. We're interested in learning whether the proposed experience aligns with your workflows for change policies, playbooks, approvals, change requests, and AI-assisted workflows. Your feedback will directly inform future product decisions and help shape the overall product experience. What's involved?20-minute unmoderated concept evaluation, followed by a 60-minute remote Zoom interview Between August 10th and August 21st $100 USD eReward (redeemable at retailers like Amazon) Interested?Sign up for a session via User Interviews.If none of the listed times work, please email me with your availability and we'll do our best to accommodate. Contact: Tara Rastogitara.rastogi@zendesk.com Additional InformationParticipation is subject to the Zendesk Privacy Notice (including the California Notice of Financial Incentive, where appli

Hello! I’m working on setting up some advanced AI agents right now for email. We have patients & staff both send questions to the same channel. Ideally, I’d like my AI agent to only respond to questions from patients, not from staff members. However, although I’ve set up instructions & use cases for this, it continues to just respond to every question during testing. Does anyone have suggestions on how I can get it to automatically send questions that don’t pertain to set up use cases directly to live agents?

Hey community,We want to open up a discussion around a major theme from our Relate 2026 roundtables: navigating team uncertainty. When we introduce AI agents, it's completely natural for frontline staff to feel a bit anxious about how their roles might shift. Change can be intimidating! But the teams hitting a 50% ticket deflection rate while keeping morale high aren't replacing humans. Instead, they focus on two key strategies: Elevating the human experience: Human service remains non-negotiable for premium, high-touch customer tiers. Because AI can struggle with vague queries, scaling successfully means building seamless, zero-friction "Can I talk to a human?" escalation paths the moment automation hits its limits. Guiding teams through the transition: Frontline adoption is a psychological milestone, not just a technical one. Leaders are smoothing out resistance by launching volunteer  tiger teams and internal champions, reframing AI as a career accelerator, and hosting casual win-

Hi everyone,We are running into a frustrating system constraint regarding how Zendesk’s native AI features (Auto-Assist and Ticket Summaries) handle inline (interposed) replies from customers, and we’re hoping to learn how other teams have solved or worked around this.The ProblemWhen we send a structured list of troubleshooting or diagnostic questions, many customers reply inline (e.g., typing their answers directly beneath our questions, sometimes changing the font color or bolding their text).Because of how email clients nest these replies, Zendesk's email parser flags the incoming text as empty or strips it as "quoted/duplicate text." The agent has to manually click "View original email" to actually see the customer's inline answers.The Impact on Zendesk AIThis completely breaks our automated AI workflows: Ticket Summaries: The built-in Ticket Summary function fails to register the inline answers. It summarizes the ticket as: "The customer has not yet provided this information. The

As part of the Custom Agents EAP, we're using a custom agent that reads the attachments of incoming tickets, summarizes them, and extracts data from them. For tickets that come in through a help center form for one of our brands, accessing the attachments fails reproducibly.Setup: A single Zendesk account with multiple brands. Affected brand: brand subdomain brand-group.zendesk.com, host-mapped domain service.brand.de. Example form: https://service.alvi.de/hc/de/requests/new?ticket_form_id=24976309304860 Problem: For tickets that come in via the respective brand webform, the agent cannot download or read the attachments. Instead of the file, an authentication/redirect page is returned. The attachments are located on the host of the brand. For tickets via the default brand (account domain), the same agent works perfectly — the attachments are read and processed correctly. Important: "Secure downloads" and "Private attachments from end users" are disabled. So these are not private

Hi, we are trying to connect our AI Agent to a custom crm api connection that we created within Zendesk, but the current documentation seems to reference how it was done before the merge of Advanced AI Agents into the base AI. Does anyone know how this is currently connected?  Thanks in advance John

I’m scanning through the recommendation from Admin Copilot and there are a number of “Create an article” suggestions. These are recommendations that I create articles for FAQs for which we already have articles. And when I click in the “Related tickets” for the recommendation, it’s quite clear that the AI Agent (Essentials) has replied appropriately to the end-user’s request based on the existing Help Center articles.Our Help Center is 98% locked down - the only 2 anonymously viewable articles are “How to Use this Support Site” and “Legal Notices”. Everything else requires authentication, which is required by our InfoSec team due to the subject matter in our Help Center. Is Admin Copilot basing its recommendations solely on anonymously viewable articles? If so, it’s basically useless for us, and in fact, a distraction (since most of its recommendations are to create articles that exist).

Hi Team, We’ve definitely started thinking differently about content now that AI is becoming a common way for customers to get answers. Instead of creating long articles, we focus on making content easier to scan, with clear headings, short sections, and direct answers to common questions.One thing that has changed is how often we review our help content. If we notice customers asking the same questions through tickets, chat, or community posts, we update those articles first. We also try to keep information consistent across all channels so AI tools can surface accurate answers.AI helps us with drafting and organizing content, but we still rely on human review before publishing. Accuracy is important, especially when product features or workflows change.To measure success, we look at customer feedback, article usefulness, reduced repeat questions, and whether customers are finding answers without needing to contact support. When fewer people ask the same question and feedback improves

We have noticed, that all the managed accounts are missing AI Agent’s Allowance usage -information in their Admin Center. As we are a Zendesk-reseller in EMEA, this feature would be very crusial for us and our customers, in order for them to track how many Automated Resolutions they currently have left. Based on that information, they would be able to decide when to purchase more Automated Resolutions, before they run out of them. Without the information, it’s also impossible for them to know whether they are currently running on the pay-as-you-go pricing (2€), or the regular pricing (1,50€) - this issue would also be solved by showing the Allowance usage -information in the Admin Center.Below are the screenshots how a regular account looks like, compared to how a managed account looks like.We hope this would be developed soon!  

Hi everyone,I’m looking for the best practice to route outbound WhatsApp replies to specific Use Cases within the Zendesk Advanced AI Agent.My Scenario:I have 3 different Meta WhatsApp Templates used for proactive outbound messaging. All three templates start with a "Yes/No" Quick Reply button. In the Advanced AI Agent (Bot Builder), I have created 3 separate Use Cases (one for each template) to handle the different follow-up flows.The Problem: When a customer clicks "Yes" or "No", the AI Agent fails to identify which Use Case it should trigger. Since the reply is a simple "Yes", the NLU (Natural Language Understanding) cannot distinguish which template the user is responding to, leading to incorrect routing or a generic fallback.What I’ve tried:Sending metadata during the outbound trigger. Identifying the Template ID in the API call. Despite this, the Advanced AI Agent doesn't seem to "read" this context to select the correct Use Case automatically.My Question: What is the most effect

Hi, we are using Zendesk Advanced AI and our chat bubble is an ongoing conversation (it doesn’t create a new one every time the customer opens it but it’s a continuous conversation). i noticed that dialogues or procedures don’t get applied when the chat conversation isn’t completely new. need advice on how to include existing or reopened conversations on procedures/dialogues. 

I know that these two are no longer distinct technically but reading the documentation it isn’t super clear to me if there’s actually a significant change as Zendesk would have you believe. I can see that the upgraded AI can access API’s, and that might be something that would be a game changer as I’ve just started messing around with managing our help center via API, but if anyone has done that already, I’d love to hear examples of how you’re using that. 

Let’s talk about Copilot Ticket Summaries!I didn’t really want to talk about them. In fact, we already did talk about them, tried them, and decided not to buy them. Then you decided to make it weird by forcing them on everyone as a freemium upsale, in the hopes we’ll love them so much we want to pay a severely discounted price for them for a short time before the AI compute bill comes due for us all.  How do we opt out of having this unwanted and already-rejected feature enabled for our account? In advance, if you please.I don’t want to have to wait for them to be switched on to switch them off. And if I may say so: This is a pattern of feature rollout with Zendesk that is starting to get annoying.