ChatGPT is the biggest name in AI right now. From recipes to speechwriting, settling historical debates or giving you driving directions, people around the world are finding endless uses for the new tool from OpenAI.

As a Zendesk consultant and writer, I have been curious how well ChatGPT could explain some of the concepts that I write about. Can its limitless ability to collect knowledge and regurgitate it in human-friendly sentences turn it into a tool Zendesk admins can rely on for helpful tips and answers to technical questions?

I’ve set out to answer that exact question, using the topic of a recent article of mine as inspiration.

Question 1. “All” vs “any” conditions in triggers and automations

When working with triggers or automations, it's essential to understand which conditions to put in the “all” section any which belong in the “any” section. I’ve covered this topic in detail in one of my articles, Zendesk Triggers and Automations: when to use ALL vs. ANY

Let’s see how ChatGPT fares when asked about the difference between these two.

Prompt

In the persona of a Zendesk administrator, explain to me when I should use "all" conditions and when I should use "any" conditions in a Zendesk trigger.

Response - part 1

‍

‍

Analysis

This is just the first half of ChatGPT’s answer, focussing on the “all” conditions. Overall, I’d say this is a very clear and relatively accurate response. The way ChatGPT explains the logic behind “all” conditions is really easy to understand. It has made quite a big error in its example, however. The final condition (“Hours since…”) is time-based. Only automations allow you to use time-based conditions; triggers do not. ChatGPT has incorrectly said that you can set up a trigger with this condition. So close, but not quite there!

Let’s see how it covers the “any” conditions in the second half of its answer

Response - part 2

‍

‍

Analysis

This is also a pretty solid answer, although it’s missing quite a key nuance that could mislead a lot of people. In its second bullet point, ChatGPT says “...if any one of the conditions within the trigger is true, the trigger will activate and perform its actions.” This statement is only true if you only use “any” conditions.

If you have a trigger or automation that uses “all” conditions and “any” conditions, however, then this statement would be incorrect. In that case, you’d need at least one condition from the “any” group to be true, plus all of the conditions in the “all” group. Let’s ask it a follow up question to see if it can get that crucial bit right.

‍

Question 2. Using “all” and “any” conditions simultaneously in a trigger or automation.

Following on from the last point, let’s test if ChatGPT can give a more detailed answer that is correct when it comes to using both “all” and “any” conditions simultaneously. This is a common use case so it’s important to get this right.

Prompt

In the same ChatGPT conversation, I followed up with:

What if I want to use "all" conditions and "any" conditions together. How do they interact?

Response - part 1

‍

‍

‍

Analysis

A good start - ChatGPT’s point #1 is correct. Unfortunately, there is a massive error in point #2. ChatGPT tells us “They provide extra flexibility and allow the trigger to activate if at least one of the "ANY" conditions is true, regardless of whether the "ALL" conditions are met”. That final part is totally incorrect. When using “all” conditions and “any” conditions in a trigger or automation, all of the “all” conditions must be met.

When I wrote about this previously, I created this image to highlight how Zendesk’s logic works:

‍

‍

Response - part 2

‍

‍

‍

Analysis

Interestingly, despite explaining its answer incorrectly previously, the explanation it gives for its example here is pretty good. That said, the example itself demonstrates ChatGPT’s lack of understanding regarding general customer service intricacies as well as how Zendesk works.

First of all, it’s very unlikely that you’d ever want to send the same automated response to a VIP customer as you would any ticket that is assigned to a “top-tier agent”. Tickets assigned to “top-tier agents” could be regarding any matter. Similarly, VIP customers are arguably the last customers you should be sending automated messages to. Instead, these are the customers that are most worth your agents’ time to keep happy.

Secondly, whilst we might be able to guess what ChatGPT means by “Top-Tier Agent”, the concept doesn’t exist within Zendesk, and certainly not one that is selectable when using the “Ticket: Assignee” condition in a trigger or automation, as ChatGPT has suggested. Whilst you can select a ticket’s group in a trigger, when it comes to selecting an assignee, we can only select a single assignee name at a time.

ChatGPT has an impressive grasp on some relatively complex Zendesk topics. Its understanding of when to use the “all” conditions vs “any” was quite impressive, however it fell short at the final key hurdle just about every time.

The phrase “a little knowledge is a dangerous thing” comes to mind. Whilst ChatGPT certainly has more than just “a little knowledge”, its understanding of Zendesk intricacies is far from complete. This fact, married with its excellent answers in other areas of Zendesk, could unintentionally lure users into believing that ChatGPT is a great authority to be relied on for understanding Zendesk.

Given that it requires an expert eye to spot these mistakes, and that most people who will be asking ChatGPT questions like these would probably not class themselves as experts, it might be hard to spot when it has got it wrong.

That said, the way it broke down some of the basic logic into clear language was very impressive. Trying things out for yourself within Zendesk, and asking ChatGPT to explain individual concepts that you can’t crack - this could be a combo that accelerates your Zendesk learning.