> For the complete documentation index, see [llms.txt](https://dots.gitbook.io/dots-docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://dots.gitbook.io/dots-docs/discovery/ask-ai.md).

# Ask AI

Ask AI is your natural-language assistant for exploring and analyzing your entire dataset — both qualitative and quantitative — through a simple back-and-forth conversation. Instead of writing queries or building filters by hand, you ask a question in plain language and get back a written answer that is **grounded in your own data**, with **references** to the exact passages it used.

It's especially useful when you want to combine qualitative stories with supporting quantitative data, or explore trends across data collection rounds, forms, or respondent segments. You'll find it under ***Ask AI*** in the main left panel.

It only answers from **your own datasets** — not outside or general knowledge — and is deliberately built **not to make things up**. If the answer isn't in your data, it will tell you rather than guess.

**Why use Ask AI?**

* **Ask in plain language** — no queries, code, or manual setup. Just type a question the way you'd ask a colleague.
* **Qual + quant together** — it pulls qualitative passages and can count and summarize structured fields in the same answer.
* **Grounded and verifiable** — every answer is assembled from real passages in your data, each linked back to its source.
* **Explore trends fast** — compare across rounds, forms, segments, or regions without leaving the chat.

#### Asking a question

<figure><img src="https://2208645187-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FgczwnwqbMALFEXVtg0t1%2Fuploads%2F2oHLh53QUTUA6T5imI6h%2Fask%20ai%20(1).gif?alt=media&amp;token=978b74c9-90d1-4e20-b314-60f4e25b91e0" alt="" width="563"><figcaption></figcaption></figure>

* A single text box sits at the bottom (*"Ask any question…"*). **Enter** sends; **Shift + Enter** adds a new line.
* Answers come back as **formatted text** — headings, lists, tables — and when your question is about counts or breakdowns, the answer may include a **simple bar chart**.
* **Follow-ups work.** The chat remembers the conversation, so you can ask *"why is that?"* or *"what about the other region?"* without repeating yourself.
* **A short wait is normal.** A status line ("Looking through your data…") cycles while it searches and reads.

**For sharper answers:** be specific — *"What frustrations did people mention about shared toilets?"* works far better than *"Tell me about the data."* If an answer feels thin, rephrase or add detail rather than repeating the same question.&#x20;

#### Filtering — set this up before you start

By default, Ask AI searches all of your datasets. Before you send your first message, open the **Filters** tab in the left sidebar to focus it. There are three things you can set:

* **Content Types (datasets)** — choose which datasets the AI should search. This is the biggest lever: narrowing to the right dataset(s) makes answers sharper.
* **Filters** — narrow the data further by tag, theme, category, date, or a numeric range. These are generated automatically from whatever datasets you selected.

{% hint style="info" %}
Filters and the chosen engine **lock once you send your first message** in a chat (you'll see a lock icon). This is intentional — it keeps the whole conversation consistent. **To change datasets or filters, start a new chat.**
{% endhint %}

#### The answer and its references

Every answer is grounded in your data — assembled from real passages in your datasets, not invented.

<figure><img src="https://2208645187-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FgczwnwqbMALFEXVtg0t1%2Fuploads%2FRg3TV9DU7M9J6svvwRTe%2Fimage.png?alt=media&amp;token=03675bef-5641-4b1f-a806-4b693673eede" alt=""><figcaption></figcaption></figure>

Below the answer, a **Top References** strip shows a few of the sources. Click any of them (or **View all references**) to open the **References panel** on the right, which lists **every** source the answer used:

* **Passages / highlights** — the actual text snippets (annotations and content) it drew from.
* **Document links** — jump straight to the source document, opened at the exact field.

References are the **trust mechanism**: they let you verify any claim by going back to the source. This is the main reason to trust an answer — always check them before quoting.

#### Copying and giving feedback

* **Copy** — copies the formatted answer so you can paste it elsewhere.
* **Feedback** — 👍 / 👎 on each answer. A quick *"Thanks for your feedback!"* confirms it, and it helps improve the system.

#### Managing chats

* The **Chats** tab lists your past conversations, each labeled by its first question.
* **Start A New Chat** begins a fresh conversation — and lets you set new datasets and filters.
* Delete any chat from its menu.
* Chats are **per-user**: you only see your own.

**Tips for getting the most out of Ask AI**

* **Be specific.** It searches for relevant material, so focused questions beat broad ones — *"summarize everything"* works less well than a pointed ask.
* **Filter first.** Narrowing to the right dataset(s) before you start is the single biggest lever on answer quality.
* **Use follow-ups.** Treat it as a conversation — drill in with *"why?"* and *"what about…?"* instead of re-asking from scratch.
* **Always check the references.** They're how you confirm a claim is real; sense-check before quoting.
* **A "no evidence" answer is working as intended.** If the data isn't there, Ask AI says so rather than guessing — that's the design, not a failure.