Generative replies let you use AI-generated responses in your conversation bots to address questions from your end users. These responses use generative AI to evaluate articles in your linked help center, then use that knowledge to provide concise answers in conversations with your customers.
You can enable generative replies as part of a bot’s general setup configuration.
This article includes the following sections:
About generative replies in the EAP
The generative replies feature is available as part of the Generative AI early access program (EAP). See the Zendesk generative AI capabilities overview for more information on this EAP.
This section includes the following topics:
Overview of the feature
Generative replies can respond to a customer question when:
- No intents are detected for that question
- An intent matching that question is configured to generate a reply
If your account meets the eligibility requirements for using intents, you will gain access to intents when added to the EAP. Intents are included in the Advanced AI add-on. See Using AI-powered intents with conversation bots for more information.
Please note that when the EAP closes, you will need to purchase the Advanced AI add-on to continue using both generative replies and intents, if you haven't already added it to your account.
When activated for your bot, generative replies benefit both admins and customers. Customers can gain useful knowledge from your help center articles right away, without leaving the conversation. Admins can rely on AI-generated automated replies based on help center content, making it easy to deliver useful information to customers without building custom answers.Generative replies impact your conversation bot in a number of ways. This list describes how some common bot behaviors are affected:
- Bot reply presentation. When a relevant help center article is identified, generative replies are displayed as a single message with a link to the source article. These messages are short (generally under 100 words), and reflect the bot persona, if applied.
- Reply feedback. Generative replies are always followed by a standard request for end-user feedback (e.g. “was this helpful - yes/no”).
- Reply fallback message. When there's no relevant answer or help center content to generate a reply, the bot responds with the fallback message they've configured.
-
Languages. The generative reply content uses the language applied to the bot.
Account requirements
To use the generative replies EAP, your account must meet the following requirements:
- Agent workspace is activated
- Messaging is activated
- An open Zendesk knowledge base is connected to your account. If your help center requires users to sign in, it won't be available for generative replies.
- (Intent-related features only) Meets the requirements for using intents
EAP limitations
Because this is a pre-release version of the feature described in this article, there may be functionality limitations, including:
- Restricted help center articles won't be available via generative replies.
- Images can't be included in AI-generated answers.
Joining the EAP
This EAP includes generative AI features used in a number of functional areas, such as agent productivity, bots and automation, knowledge base management, and voice support.
When you sign up for the EAP, you can select the features you want to try. You are not required to apply all EAP features to your account. To use generative replies, you’ll need to opt-in to Generative AI for bots and automations. This also allows you to select and apply a bot persona, which determines the voice of AI-generated messages, including generative replies.
See Zendesk generative AI capabilities overview for more information on these features.
You can request admission to the EAP by filling out the sign-up form.Activating and configuring generative replies
When you are accepted into the generative AI EAP and opt-in to the generative replies feature, the option to activate generative replies appears in a conversation bot’s setup tab.
To turn on generative replies
- In Admin Center, click
Channels in the sidebar, then select Bots and automations > Bots.
- Under Conversation bot, click Manage bots.
- Click the bot you want to use with generative replies, or create a new bot.
- In the Setup tab, click to expand the section If the bot finds relevant articles.
- Select Generate a reply. If you’re creating a new bot, this option is selected by default.
When you select this box, any intents without an assigned answer setting will be updated to Generate a reply
- Click the Test bot button and enter phrases the bot can match to existing help center content, to view samples of replies your end users might receive.
- Click Publish bot to apply your changes.
Using generative replies with intents
To manage generative reply settings for specific intents
- In Admin Center, click
Channels in the sidebar, then select Bots and automations > Bots.
- Under Conversation bot, click Manage bots.
- Click the bot you’re using with generative replies.
- In the Intents tab, for each intent use the drop-down to select an option:
- Generate a reply (default) lets the bot create a response to a customer’s matching free-text entry.
- Don’t generate a reply prevents the bot from responding to a customer’s matching free-text entry. Instead, it replies with the standard fallback response (“Sorry I couldn't find a good answer to your question. Here are some topics that might help though”).
- Show an answer lets you connect the intent to an existing answer, or create a new answer for it.
Use the filters at the top of the Topic list to locate specific intents more quickly.
You can also choose whether you want to generate a reply for any question asked by an end user that does not have a matching intent. By default, a reply will be generated for any customer free-text entry that matches an intent if that intent is not associated with an answer in the conversation bot.
To manage generative replies to questions without matching intents
- In Admin Center, click
Channels in the sidebar, then select Bots and automations > Bots.
- Under Conversation bot, click Manage bots.
- Click the bot you’re using with generative replies.
- In the Intents tab, use the drop-down at the top of the page to select Generate a reply or Don’t generate a reply.
These changes are applied when the bot is published.
31 Comments
Does the phrase "The generative reply content uses the language applied to the bot." mean that we are unable to use generative AI to support more than one language? Does the function "Automatically translate answer" under the Language settings of a bot not work if you're using generative replies?
Hi Cloudhuset, generative bot replies can work in more than one language. If automatic translations is enabled, and the bot is responding in a second language (that is not the default bot language) then the generative replies will be consistent with that language, provided there is help center content in that language from which to derive a reply.
Is this feature to be added to auto reply for articles via email?
Does this mean that restricted article availability is on the roadmap for full release?
For context, our widget is only available when signed in, and we use SSO, so allowing the bot to access restricted articles will be key for us.
Is there a way to exclude certain help center articles from being the main source, or to rank support articles to make them more likely to pull from?
Use case: All of our products require installation, so the word "install" is present in dozens of our support articles. When I tested a question asking about installation, the answer pulled was from an article that needs to be public, but would be the wrong answer 99% of the time.
I could configure tags to exclude an article from generative replies, or if there was a setting within guide that allowed an article to be excluded from generative AI responses, that would be even better.
I am hoping this is a configuration issue on my side, but I cannot seem to get generative replies to see any of my KB articles. Whether they are public or private. Any question I ask it returns no results. Any of my Answers where I push an article works fine. I do not see where I am doing anything wrong in the setup.
Is there any explore reporting available during EAP that will help us assess how effective the generative replies are in deflecting tickets?
How does this work with flow builder? We have some questions where we know the customer will need to open a ticket. Are we able to use flow builder and the generative AI tools at the same time?
I cannot turn on generative reply on and existing BOT because the Setup tab the section "if the bot finds relevant articles" is not present
I find the section only if I create a new BOT but that would mean to recreate all the answers of the existing BOT
Hi Ashley Moore yes extending generative replies to autoreply on email is on the roadmap for H1 2024
Thanks for the feedback Alan Pugh. Yes supporting restricted content for authenticated users is on the roadmap for Q1 2024.
Hey Trudy Slaght, no there isn't currently any explicit features to exclude, rank or filter articles for use with generative replies. Thanks your feedback, we'll consider it as we plan the future roadmap. One suggestion in the mean-time is to review the content of articles relevant to the broader "installation" intent and encourage users to write complete phrases so the bot can find the best article without the need for tags or filters e.g. "how do I install A" vs "how do I install B" rather than just a single word like "install". The bot typically performs better across the board at recognising user intent and responding when the phrase is greater than two words.
Hi Sean Morrissey, I know we corresponded about this via email, however I'd just like to confirm were you able to resolve the issue? Can you share what was/is preventing generative replies from working with your help center?
Hi Austin, unfortunately no there isn't any Explore reporting for generative replies at this stage. However, we are actively planning for how we might best provide insights to customers relating to generative replies performance and expect to introduce them in H1 2024.
Hi Hayley Johnson, today the bot will respond to a user message with either a generative reply or a custom answer built in flow builder. And if it doesn't find either it will display a fallback message. If the user receives a generative reply and needs to continue on to talk to an agent this is possible today if they select "no" when the bot asks for feedback "was this helpful?" and the bot presents an option for a custom answer flow built for agent transfer. That said, we are considering how we can improve this experience and provide more flexibility for admins to configure how the bot should follow up from generative replies. Does this help? I'd love to hear more about the functionality you had in mind?
Hi Elisa, I've reached out to you directly to assist with this issue.
Is there a way to ignore selected articles based on labels?
We have articles for both clients and providers on our help centre. The longer term plan is to split this into two different help centres but for now I'd like to only generate articles that apply to clients.
would you please fix this ?

Hi Pete Holborow no there isn't a feature to filter help center articles through labels or similar means currently. However, we plan to release an enhancement to support restricted content in Q1 2024. With that, you could use Guide user segments and through authentication, the bot would only access content relevant to clients if the user is logged in and identified with the appropriate user permission. Will this meet your needs?
Hi Deiaa Eid thanks for the feedback. Good timing, we're actively working on improving the bot's comprehension of small talk such as your example here "thanks" cc/ Gin Atkins
Thanks Daniel Aron - yep Deiaa Eid we're going to start with dealing with 'hello' or greetings, and thank you type messages to avoid that situation. We're in discovery on this now, exact release date timing is TBC but expect it early next year.
Thank you Daniel Aron. Follow up question, where can I edit the "knowledge is power" message? I can't find that text anywhere in the settings.
Hey Pete, that message isn't editable currently, however we're considering how we might enable customization of those feedback collection messages. Can you share why and how you'd like to edit it?
The tone just doesn't really fit the rest of the answer and tone I'm going for. I'd prefer to be able to customise it to be a little more appropriate for our use case (medical related questions).
Hi. I am having the same issue as Sean Morrissey. Did you find a solution?
Hi Havard, we're actively working on a solution to enable restricted content to be referenced for generative replies. We're expecting to release this in January 2024.
Hi @Daniel Aron. My content is not restricted though. It is open to all, yet the generative replies will not work. It only says it could not find a solution.
Hello, we are experiencing the same issue as Håvard on our unrestricted help center. I created a ticket about this topic with Zendesk support today.
I had a call with the dev team last week. Appears to be a q1 2024 release for an update.
Please sign in to leave a comment.