Unlocking the Power of Zendesk AI: Your Questions Answered

May 15, 2024

Unlocking the Power of Zendesk AI

Zendesk helps customer service teams scale—and stay nimble.

In an era of towering customer expectations, workforce limitations, and economic unpredictability, the power of AI can become your secret weapon to stay agile and exceed those expectations. Built on billions of customer-service data points,Zendesk AI can enhance smarter conversations and bots, boost productivity tools for agents, and provide new insights and instant actions for admins.

During our recent webinar, our team of experts covered everything you needed to know about Zendesk Generative AI, including:

  • Zendesk and Advanced AI Overview
  • Breakdown of all things Generative AI
    •  Generative AI for KB
    •  Generative replies for bots
    •  Generative AI for agents (expand, tone shift,  summarization)
    •  Generative AI for voice (call transcription, call summary)
  •  Zendesk AI Roadmap 

Did you miss it? No problem! You can watch the full webinar here

Everyone had great questions, and unfortunately we weren’t able to address them before we ran out of time. We saved them all however, and have provided answers to them below. 

Want to chat about all things AI? Don’t hesitate to get in touch. 


Do all of these AI options work with the classic widget? We have not yet migrated to the messaging widget.

To fully benefit from Zendesk AI, you need to switch to messaging.
To use generative replies, your account must meet the following requirements:

  • Agent workspace is activated.
  • Messaging is activated.
  • An active Zendesk knowledge base is connected to the bot’s assigned brand.
  • Is assigned a matching intent model (Intent-related features only).

More information here.


What strategies does Zendesk use for unsupervised learning and representation learning within Gen AI models, especially in scenarios where labeled data is scarce or unavailable?

All models developed by Zendesk are classification models – this means they are trained to read and classify inputs into one of a set number of categories created by Zendesk. Because these models are not generative, no content is produced by the model, and it is not possible for data to be reproduced by the model.

Additionally, before Service Data is used to train a generally available model, Zendesk applies aggregation and sanitation processes, as necessary. No fields designed to intake personal data or ticket attachments are used for model training. Zendesk is committed to ensuring that no Service Data will be reproduced by the model. See AI Data Use Information. Service Data will never be shared with third-parties to train external machine learning models.

Can users influence or guide the content generated by the bot to align with specific tones, styles, or themes?


No, the content generated by the bot will have your knowledge base content as its source and the tone that will be used by the bot is set inside your Admin Center by an admin. The way end-users interact with the bot will not alter that.


Can the Generative AI bot incorporate multi-modal inputs (e.g. text, images, videos) to generate richer and more contextually relevant responses?

In general, Generative AI can indeed handle multi-modal inputs, such as text, images, and videos, to create richer and more contextually relevant responses. However, the current Generative AI bot available in Zendesk is limited to generating text only.


For autoreplies, are we able to automate the first touch for tickets that need a little extra information so the agents don’t need to collect that initial information?

To better answer your question I’d need to know if you are referring to:

  1. Autoreplies with Articles, Advanced email,
  2. Autoreplies (based on intent or sentiment, for example), or
  3. A trigger configuration to reply back to the customer with extra questions when the message meets some requirements.

Number 1 and 2: You cannot control what is going to be auto replied. You set the configuration and the system will reply accordingly, depending on the message received, the KB articles, the sentiment, confidence, language, etc.

Number 3: You can control and can automate your own replies according to each scenario.