@Jimmy Miller - Can I inquire if you’ve tested this with your emoji searching and if so, were the results as favorable using natural language than the tabled information?

Original post: Writing Content Intended to be Read by AI

@Gregg Frank  - So far, Answers won’t recognize emoji. This is a little frustrating, but I’m testing a new method - a return to ascii art.

Example: replacing  with :)

This seems to be a bit more recognizable by the language processor, but is more difficult for the end user to decipher apart from Answers.

TLDR - This tool doesn’t like emoji in any format.

Update - 

Test 4 - Hide the natural language in a dropdown menu

I “hid” the natural language list in a dropdown menu at the bottom of the card and put a user friendly spreadsheet in the card as an image.

The theory was that a user could access the card, and get the information they want from the image without seeing the natural language mess intended for the AI.

When asking Answers - 

“When is the next company holiday?” The system responded first with “Memorial Day is observed on Monday, May 29, 2023”. Still struggling with time and relationships

“When is Memorial Day?” The system answered with “Memorial Day is observed on Monday, May 29, 2023.”

“Can I have a list of company holidays?” The system listed all holidays as written.

Conclusion: Answers will read the dropdown menu successfully! If we can find a way to hide natural language in metadata, and show readers the more visibly appealing data when searching for specific cards, then this tool becomes INCREDIBLE

Hey @Jimmy Miller ! Thank you for sharing your breakdown and examples of how Answers work in your testing. Just wanted to share some more context that might help as you’re experimenting:

• The technology powering Answers is not aware of the current date or time, so asking questions about the next holiday or next event won’t result in an accurate answer. We don’t yet have a solution to this challenge, but one workaround could be including today’s date in the question like “What’s the next company holiday after June 7, 2023?”
• Card content is currently passed to Answers as plain text, but an improvement is coming soon to include markdown context like links, bullets, table formatting, etc. In your first example using a table, the table formatting is currently being removed when the card content is passed in, so Answers might not understand the relationship between all the text in the table. It’s possible this will work much better once we support markdown formatting. No promises since we are still working on it, but we will share an update once that change is out!

To your larger question about whether we should be writing cards to be read by the AI tool and not a person, our current position is that teams shouldn’t have to change the way they write content for Answers. By collecting examples like the ones you shared above, we will learn more about where the AI has challenges processing current card content and where we can make improvements. In the coming weeks, we will be adding a way to submit feedback directly on individual Answers so we can collect more feedback like this at scale. 

Let me know if there’s any more context that would be helpful as you’re using Answers and thinking about it’s impact on your workflows. Always happy to share!

This is very helpful!

Thank you so much for the insight. 

I’ll keep experimenting and trying to document as best as I can for this community,

@Brad Turner my latest iteration has been a huge success.

I used a PNG to show the information I want my people to see.

The text for Answers is below, hidden in a dropdown markdown. 

It works like a charm. My people get the pretty view, Answers gets the text it wants.