Moving from Legacy Topics to Dataset Q&A in Amazon Quick

[alaresch]

Moving from Legacy Topics to Dataset Q&A in Amazon Quick

Are you using a legacy Topic for your natural-language Q&A experience today? There's a better way — and it's available now.

With the new dataset enrichment experience, you can add business context (descriptions, synonyms, and instructions) directly to your dataset. This unlocks access to our latest agentic models with significantly improved answer accuracy — no legacy Topic required.

If you're working with multiple datasets, you can now create a multi-dataset Topic that lets you define joins and relationships, enabling accurate answers across your data — something that wasn't possible before.

---

What's changing?

Legacy Topics are now the previous generation of the Q&A setup experience. Going forward, the recommended approach is:

• Business context lives on the dataset — not in a separate Topic configuration.
• Multi-dataset relationships are defined in the new multi-dataset Topic experience — with support for natural-language join definitions.

By moving to the new model, you get access to stronger foundation models and improved answer quality immediately.

---

How to upgrade

If you have a single dataset in your legacy Topic

1. Create a new dataset and enrich it — migrate your business context (friendly names, field descriptions, synonyms, calculated fields, and instructions) directly onto the dataset.
2. Your enriched dataset is ready for the latest Q&A experience — no Topic needed.

Here's what dataset enrichment looks like in practice — you can add descriptions, geospatial types, synonyms, and behavioral instructions directly on each field:

If you have multiple datasets in your legacy Topic

1. Create new datasets — one per source — and enrich each with its business context.
2. Create a new multi-dataset Topic — define joins and relationships across your datasets, either manually or using natural language.

Here's what defining relationships looks like in the new multi-dataset Topic — the visual editor shows how your datasets connect:

You can also add custom instructions to guide how the model reasons across your joined datasets:

---

Why upgrade now?

Benefit	Legacy Topic	Enriched Dataset + Multi-Dataset Topic
Model quality	Previous-generation models	Latest agentic models with stronger reasoning
Answer accuracy	Baseline	Significantly improved
Cross-dataset answers	Not supported	✅ Define joins across datasets
Business context location	Locked in Topic config	Reusable, on the dataset itself
Natural-language joins	Not available	✅ Describe relationships in plain English

Prefer to automate?

You can create datasets and add metadata (descriptions, custom instructions) programmatically using the CreateDataSet API — see the SemanticMetadata parameter for adding enrichment context directly via API.

About the Author

Amy Marvin is a Sr. Technical Product Manager for Amazon Quick, focused on AI-powered chat analytics capabilities. She is passionate about removing barriers between people and their data, making it possible for anyone to ask a question and get a trusted answer in seconds. Outside work, she enjoys exploring the DC restaurant scene and biking on nature trails.

Have questions about migrating your setup? Drop them in the comments — we're happy to help.

[Giridhar.Prabhu]

Hi @alaresch

This is interesting.

We have Operational dashboards with mutiple datasets for different areas of the business. Can we create visuals in sheets from different datasets and will they interact based on the relationships in the topic?

Can you elaborate a bit more about this?

[alaresch]

yes! you can create a visual with fields from different datasets once you have defined the relationships of those datasets in a new topic

[Giridhar.Prabhu]

Hi @alaresch

I also read the following blog. So the feature looks cool. Something we have been wanting for long.

Build a unified semantic layer across datasets with multi-dataset Topics in Amazon Quick - Learn / Blogs - Amazon Quick Community

While I understand this requires the Admin/Author to upgrade to Pro licenses, it is not clear whether the reader’s should also be upgraded to Pro licenses. The licensing page gives me an impression that the Reader role can consume all kinds of stuff. Would you be able to clarify?

[Bhargav_s]

Hello @alaresch ,
Our production application embeds Amazon Q using GenerateEmbedUrlForRegisteredUser with GenerativeQnA and InitialTopicId. With the introduction of Dataset Q&A and semantic datasets, what is the recommended replacement architecture for embedded applications? Specifically, how do we embed Dataset Q&A, and what changes are required in our backend APIs and frontend embedding code?

[romez]

Hello @alaresch,
I have been using Quicksuite for some time, and I have a few questions.

I’m interested in chat availability during indexing. In the legacy topic, the chat was unavailable while indexing was in progress. How does this work with the new topic?

Does the new topic still have its own indexing process, or is indexing now handled by the dataset? How long does the process usually take?

Are there any limitations or recommendations regarding dataset size to ensure that the topic can handle it? For example, can a 200 GB SPICE dataset be used in a topic, or would it be better to split it into several parts and connect them to the topic as multiple datasets?

Best regards, Roman