Robert Ninness
ServiceNow Employee
ServiceNow Employee

Catalog item generation leverages Now Assist to rapidly create service catalog items from natural language descriptions, enabling users to define item purpose and questions saving time and making catalog item creation more accessible to non-technical users.

 

This quick start and prompting guide, will walk you through how to activate Catalog item generation and how to fine tune your prompts to reduce hallucinations, and improve the accuracy of generated catalog items.

 

Install and activate the Catalog item generation skill

 
To install and activate the Catalog item generation skill, see the installation instructions in the docs site(docs.servicenow.com).
  1. Install/upgrade Now Assist for Creator parent application.
    • Catalog item generation will be installed as a dependency.
  1. Activate the skill in Now Assist Admin.
    • Find the Catalog item generation skill under the Creator list of skills.
  • (Optional) Choose the LLM.
    • Under Now Assist Admin > Settings > Manage LLMs scroll to Catalog item generation and choose a Provider.

 

Now that the Catalog item generation skill is activated, developers will need access to Catalog Builder and the "now.assist.creator" role to use Now Assist in generating catalog items. After selecting a catalog template in the new catalog item flow, creators can utilize Now Assist to build their catalog items.
 
RobertNinness_0-1750925548780.png

 

What catalog item features can Now Assist generate?

 

Questions

  • Text
    • Single-line text | Multi-line text | Rich text | Masked | URL
  • Toggle
    • Yes/No | Checkbox
  • Choice
    • Doprdown with fixed values | Dropdown with values from a table | Requested for | Record reference | Radio | Multi-select | Radio with Values from a tableNumeric scale
  • Date/Time
    • Date | Date and Time | Duration
  • Display Text
    • Plain text | Rich text
  • Attachment
  • Choices
  • Labels
  • Conversational label
  • Mandatory, hidden, read-only

Default values

  • Specify a default value to pre-fill the question

Auto-populate

  • Reference a dependent question, and the mapping of the field to value

Item descriptors

  • Item name
  • Short description
  • Item details

Record Producer

  • Question to field mapping

 

Prompting tips

 

General

 

Be as specific or as vague as you want. Sometimes you know exactly what you want and need things to be built to a spec. Other times you might be ideating or have been given broad strokes to work with. In either case, Now Assist will do its best to stick to your specifications, or fill in gaps if you aren't quite sure on the implementation. This is the power and flexibility of Generative AI, lean into it.
 
Here's an example of varying levels of specificity
 
Create a reference field called "Destination", set the reference to the location (cmn_location) table.
Create a field called "Destination" for capturing the destination location
Create the following fields: Departure, Destination, Length of stay

 

In the first instance the user has been very specific about the implementation of the destination field.
 
In each subsequent example, Now Assist must work out some of the details using context from the rest of the prompt, and its knowledge of the instance.
 

Lean into AI general knowledge

 
If you're unsure, ask Now Assist to give it a go. A Catalog generation prompt can be as simple as the following:
 
Using your knowledge of the Development Application process for seeking approval for new builds, create an exemption application form for requests to bypass the approval process based on the most common exemption requests.
 
This prompt leans on Now Assist's general knowledge and creativity to create a form based on an idea or use case, as opposed to a fully spec'd requirements document.
 
The same technique can be used at an individual question level, for example when used to generate choices.
 
Create a choice list to pick their favorite football team from the last 5 winners of the FIFA World Cup
 
Caution! Now Assist training data may not include the latest news or current world events. Your millage may vary whenrelying on knowledge of popular culture.
 

Choices

 
Experiment with different list notation to structure choice options. Markdown style works well, as does JSON For example,

 

Add a choice field for preferred delivery window, the options are:
- Morning
- Midday
- Afternoon
- Evening

 

Add a choice field for the preferred delivery window: [Morning, Midday, Afternoon, Evening]

 

Ask Now Assist to order the choices e.g.
 
Create a choice list to pick a favorite football team from the last 5 winners of the FIFA World Cup ordered by world cups won
 

Auto-populate

 

The Auto-populate behaviour pre-fills a question by dot walking into a reference and copying a field value from the record.
 
Create a field called "Departure" which references the location table. Auto-populate it with the requested user's location

 

Synonyms

 
Some of the features can be described in many ways. For example, you can refer to a Text question type as a "String" field, or a toggle as a Boolean or true/false. The LLM will do its best to interpret and convert your prompt into your intended behaviors. Don't worry about semantics and try to write naturally!
 
An exception (there's always one)! Be careful about how you describe Auto-populate and Default value. These are similar terms and the LLM could interpret synonyms such as "prefill", "autofill", "pre-populate" either way. Try to use the exact words here when describing how to handle these question behaviors.
 

Refine text

 

Many text fields have the option to refine using Now Assist. Elaborate, shorten or re-generate the text using quick actions.

RobertNinness_1-1750926063521.png

 

This UI action is available against the following text fields: Name, Short description, Description, Conversation label, and Instructions. Users can elaborate, Rephrase, or Shorten the content of these fields

 

Additional Resources

 

 

Share your prompt tips below

 
If you've found other neat ways to achieve a specific goal with your prompts, let us know in the comments. We'd love to know how you are structuring your prompts to get the most out of Generative AI.

 
 

Comments
CCZMAX1
Mega Sage

Hi Robert, I’ve noticed that even when I provide a specific question, the system tends to reword it. For example:

  • 'Request type' becomes 'What is your request type?'

  • 'Name' becomes 'What is your name?'

  • 'Prefix' becomes' 'What is your request type prefix?'

I understand it's trying to be helpful, but we've carefully designed these questions in collaboration with clients, so I need them to remain exactly as entered.

 

Do you know of a way to stop it from rephrasing or modifying the questions?

Thanks

Max

 

 

Version history
Last update:
‎06-26-2025 01:35 AM
Updated by:
Contributors