Skip to content

Issues with QuickSight Q Topics – Errors, Answer Accuracy, and Custom Instructions

0

Hi community,

We’ve been using Amazon QuickSight Q Topics and have encountered several issues. I’d like to know if others are facing the same problems and whether there are workarounds or best practices:

1. Errors After Dataset Changes When we update or delete Q Topics after modifying the underlying dataset query, we often hit errors. Even duplicating the topic doesn’t solve it. Has anyone found a stable way to manage Q Topics when datasets evolve? 2. Answer Quality Concerns QuickSight Q sometimes produces misleading or incorrect answers. It has also hallucinated UI elements or misinterpreted natural language queries. How are others handling the reliability of answers from Q? 3. Custom Instructions Not Working Well Even after setting clear synonyms, terminology, and context in custom instructions, Q frequently ignores or misapplies them. Is this a known limitation, or are there tips to make custom instructions stick better 4. Expectations vs. Reality We expected QuickSight Q to function as a true analytics copilot, but so far it feels limited to surface-level documentation-style answers rather than actionable insights. Has anyone successfully gotten more advanced use out of it? 5. AWS SDK Limitation When creating Q Topics via the AWS SDK, it only supports classical Q&A. There doesn’t seem to be support for generative Q&A yet. Is this intentional, or is generative Q support planned for the SDK? I’d really appreciate any insights, workarounds, or confirmation on whether these are known limitations/bugs.

Thanks!

1 Answer
0

I can address several of your concerns about Amazon QuickSight Q Topics, particularly around answer accuracy and custom instructions.

For improving answer quality concerns, QuickSight provides multiple ways to correct wrong answers. When Q misunderstands or misinterprets your question, you can:

  • Define terms in your question by linking them to appropriate fields
  • Adjust how Q uses a field to answer your question
  • Customize the visual you receive as an answer
  • Manually link an existing visual to a question

If Q gets a term wrong, you can choose that term in the Q bar and select "Skip this term." Users with author privileges can also add filters to datasets, link them to terms, or create calculated fields and assign them to terms.

Regarding Custom Instructions, this is a generally available feature for Amazon Q in QuickSight that allows Author Pros to curate responses by adding domain-specific knowledge. To improve their effectiveness:

  1. Open the QuickSight console
  2. Choose Topics and open the topic you want to modify
  3. Select the Custom Instructions tab
  4. Add topic-level guidance to help Amazon Q better understand context, terminology, or intent specific to your topic

Custom Instructions can include disambiguation tips, field relationships, definitions for terms that can't be captured in calculated fields or topic filters, or instructions for customizing relative date ranges. When writing custom instructions, it's important to match cell values precisely, be specific and quantitative, and use formatting for clarity.

It's worth noting that Custom Instructions have limitations - they improve Amazon Q's understanding of business context but don't add new capabilities. They won't change chart type selections, perform calculations, fill nulls, create new fields, control formatting, or alter narratives.

For verified answers, QuickSight allows you to review questions your readers ask and verify correct answers, which then appear at the top of the list when someone begins entering a question.

Regarding your other questions about dataset changes, SDK limitations, and expectations vs. reality, these appear to be beyond what's covered in the available documentation I have access to.
Sources
Improve Q&A accuracy with custom instructions - Amazon QuickSight
Correcting wrong answers provided by Amazon QuickSight Q - Amazon QuickSight
Correcting wrong answers in Amazon QuickSight - Amazon QuickSight
Custom Instructions for Topics now generally available for Amazon Q in QuickSight - AWS
Verifying Amazon QuickSight Q answers - Amazon QuickSight

answered a year ago

You are not logged in. Log in to post an answer.

A good answer clearly answers the question and provides constructive feedback and encourages professional growth in the question asker.