Holiday Calendars for multiple countries

[Patricia Romeo]

Hello!  We have just launched Virtual Agent using the Azure LLM and Genius search results.   The most common question asked so far is around the holiday calendar.   Some questions are "Is Veteran's Day a firm holiday?" or "Do we get off for Columbus Day?".   Those are not firm holidays and the VA response is the standard "I can't find anything on that directly but I can help you explore other options."  In addition, when users ask for "holiday calendar" the VA cannot provide a link to a very specific article entitled "2026 Holidays" that utilizes blocks for each country and is a table format.   I have tried changing the format of the table multiple times.   I cannot figure out if the block is preventing results to render or is there some additional formatting necessary on the table in order to answer specific holiday questions.   How are others formatting a similar article so that Virtual Agent can provide specifics regarding firm holidays across multiple countries?

[MaryG]

@Patricia Romeo I’d look at this as two separate issues:
1. Is AI Search retrieving the right article
2. Can the model easily interpret the answer once it gets there?

Knowledge Blocks themselves should not prevent the article from being searchable. They are supported by AI Search and can be used with Now Assist in Search/Virtual Agent on supported releases.

The table, however, may be part of the issue. ServiceNow recommends minimizing complex table formatting for generative AI because the model has to work out the relationship between the rows, columns, headers, countries, holiday types, etc. That gets harder with questions like “Do we get Veterans Day off?” where the answer depends on both the country and whether that day is considered a firm holiday.

For this use case, I’d make the content much more explicit in natural language instead of relying only on the table. For example:

United States – 2026 Holidays

Veterans Day – November 11, 2026
Veterans Day is not a firm/company holiday. Employees should follow their normal working schedule unless otherwise directed.

Columbus Day / Indigenous Peoples’ Day – October 12, 2026
Columbus Day is not a firm/company holiday.

Then repeat that same structure for each country.

I’d still keep the table if it’s useful for people reading the article, but I wouldn’t make the table the only place where the relationship between the holiday, country, and holiday status is expressed.

I’d also test queries like “holiday calendar,” “Veterans Day,” and “Do we get Veterans Day off?” directly in AI Search Preview. If the 2026 Holidays article isn’t being returned there, I’d work on retrieval first before spending more time changing the article formatting.
Short description, article content, and meta/tags all contribute to relevancy.

My recommendation would probably be one 2026 Holidays overview article with clearly separated country sections, or separate country-specific articles if there is enough content to justify it.

For this use case, I’d make each holiday, country, and holiday status explicit in natural language rather than asking the model to infer those relationships from a complex table.

Remember: 

“Veterans Day | US | No” is data.

“Veterans Day is not a firm holiday for U.S. employees” is knowledge.