> For the complete documentation index, see [llms.txt](https://help.daasity.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://help.daasity.com/core-concepts/ai-analyst/how-your-views-are-built.md).

# How your views are built

Where the views your AI queries come from, what carries over from your model, what doesn't, and what happens when they're rebuilt.

Your AI doesn't query your warehouse tables directly. It queries **views** that Daasity builds from your data model, so every answer starts from the same definitions your dashboards use. This page explains where those views come from, what carries over from your model, what doesn't, and what happens when they're rebuilt.

## One view per Explore

Each Explore in your Daasity model becomes one view your AI can query. If you use Looker, that includes the Explores your team built. If you don't, your workspace gets Daasity's standard views.

A view carries:

* **Metrics** - sums, counts, distinct counts, averages, medians and calculated measures from the Explore
* **Dimensions** - the fields you group and filter by, including the values a field can take
* **Definitions** - the descriptions and metric definitions written in your model
* **Default filters** - filters the Explore applies by default (for example, excluding cancelled orders), written out in plain words so your AI applies them unless your question says otherwise
* **Labels** - the names your team gives fields, such as SKU attribute names

## Rates are calculated from totals

Some measures are written as a rate on each row, such as sales divided by spend, and then added up. A view calculates these from totals instead: total sales divided by total spend. That keeps ROAS, click-through rate and cost per click right at any grouping, whether by day, week or campaign. The number can differ from a Looker tile built on the same measure, because the tile adds up the row-level rates.

## What doesn't carry over

A view is built from the model, not from dashboards. These never carry over:

* Custom fields and table calculations added on a dashboard or in an Explore session
* Filters set on a dashboard tile
* Anything built outside your Daasity account, such as separate files or external tables

Some fields can't be carried over safely, so they're left out of the view instead of returning a wrong number:

* **Comparison measures** that depend on a selected period (vs. last year, vs. prior period). Ask your AI for both periods and the comparison instead; it can line them up for you.
* **Measures on a join that repeats rows.** When a join matches several rows to one record, sums from it would double count. Dimensions and distinct counts from that join still come through. Where it matters, a separate view at the join's level carries those measures, named after the Explore and the join, for example "Orders by Line Item".
* **Fields that point at data your account doesn't have**, such as a connector you don't use.
* **Rates that can't be split into two totals.** A summed rate whose top and bottom can't be separated is left out.
* **Measures that add up an ID.** A sum of a product ID or a similar identifier isn't a real total.
* **Measures your warehouse rejects.** If the SQL behind one measure has an error, that measure is left out and the rest of the view builds.

Every view lists what it left out and why. Ask your AI "What was left out of the Sales Report view, and why?" before concluding a field doesn't exist.

## Views built on a derived table

If an Explore in Looker runs on a derived table, the view is built from the same SQL. Its totals can differ slightly from a dashboard that reads a stored copy of that table, because the two are refreshed at different times. Use these views for rates, trends and changes, and tie out absolute dollars or counts against the dashboard. The view's health shows when this applies - see [Reading a view's health](https://help.daasity.com/core-concepts/ai-analyst/reading-a-views-health.md).

## Customer information

Names, email addresses, phone numbers, street addresses and IP addresses are never returned, whichever view or table a question reads. Order-level details such as order IDs, amounts, products, city, state and ZIP remain available.

## When views are rebuilt

Views are rebuilt when your data model changes, and you can rebuild them yourself from **AI Analyst > Connect Your AI > Refresh MCP Views > Refresh Now**. A few things to know:

* **Your current views keep answering while a refresh runs.** Each view is swapped for its new version as that version finishes building.
* **A refresh can take a few minutes to a few hours**, depending on how many views your model has. One refresh is enough; starting another doesn't make it faster.
* **Connect Your AI shows the refresh as it runs** and, under **Your views**, when each view was last built. A view that didn't rebuild keeps its earlier date, so you can see at a glance which views are behind.

## Related

* [Reading a view's health](https://help.daasity.com/core-concepts/ai-analyst/reading-a-views-health.md)
* [Ask your AI about your views](https://help.daasity.com/core-concepts/ai-analyst/ask-your-ai-about-your-views.md)
* [Give your AI standing context](https://help.daasity.com/core-concepts/ai-analyst/give-your-ai-standing-context.md)
