---
slug: chat-fragen
title: "How do I ask questions about my own data?"
category: berichte
locale: en
updated: 2026-10-17
keywords: ["chat","AI","questions","assistant","language model"]
questions: ["What can the portal chat do, and what can it not?","How many questions can I ask?","AI chat in the portal","Ask questions about the data","What can the assistant do?","Can I ask the chat how to use the product?","What data goes to Google when I ask?","Are order numbers passed to the language model?","Why does the assistant not list every order?","Can I ask questions in the demo?","Which model does the chat use?","Which time zone does the chat use?","Why does the chat give a different figure for today than the overview?","Why do I get an error instead of an answer?","Can the chat tell me the contribution margin?","Why does the assistant say a tool does not provide something?","Can the chat tell me visits and time on site?","Why does the chat not name individual pages?"]
url: https://bench.metrunio.com/en/help/chat-fragen
---

# How do I ask questions about my own data?

**What can the portal chat do, and what can it not?**

Under “Ask your data” a language model answers questions about your measurement data and about how to use the product. It reads only through defined tools, states only figures that came from a tool call, and openly reports which tools it used.

> **Note:** In the demo nobody asks the language model. “Ask your data” shows three prepared example answers there, with the period, the model and the report the figures come from; each is labelled as a prepared answer. What is prepared is the wording, not the figures: they are calculated from the example data, from the same reports the answers lead to, and change with it. To ask your own questions you need an account of your own.

> **Note:** On the free Start plan the assistant answers ten questions per calendar month; from Pro there is no monthly limit. A question only counts once an answer has come back. If the language model fails, nothing is used. How many are left is shown above the input field.

| Tool | Answers |
| --- | --- |
| metrics | totals and metrics for a period |
| breakdown | split by channel, campaign, ad, creative, placement or publisher site |
| timeline | development over time |
| modelComparison | the same conversions across several attribution models, from the Pro plan; before that, the one model of your plan |
| conversions | individual completions |
| journeys | individual journeys with touchpoints; if asked, only those with a viewable impression, those starting via a particular channel or those with a touchpoint from a particular campaign |
| surveyAnswers | answers to the post-purchase question, labelled self-reported, with the answer rate |
| commerce | commerce as in the portal: new-customer cost, net revenue, contribution margin, repeat purchases and the product report, for one campaign if you like |
| web | web analytics: page views, visits, visitors, time on site and bounce rate with the previous period, per channel, source and campaign or per device, browser and operating system; measured for visitors who consented. The chat does not name individual pages, paths or real time, because a page path can carry things that do not belong in a language model. They are in the web analytics area |
| dataQuality | coverage, discarded events, findings |
| help | articles from this help centre, with source URL |

The “help” tool is why you can ask for a number and for its explanation in the same field: a question does not know which drawer it belongs in. “Why is half of my conversions unattributed?” is both: a number from your data and an explanation from an article.

- All tools are read-only; the model cannot change anything.
- If your question names no attribution model, the assistant uses your default model and your post-view setting from “Settings”, like the overview and the reports. If you come from a report in which you chose a different model, it takes that one as its default; the address of the “Ask your data” page carries it, and it is shown below the input field. If you name one (“by first touch”), it uses that one, provided your plan includes the model comparison. Below the answer it says which model and which period were used. The sources beneath it open the report with the same period, model and environment.
- Periods are whole calendar days in your account’s time zone (under “Settings”). “Today” runs from midnight until now, “the last 7 days” are seven whole days including today. That is the same rule as in the reports’ period picker. Every answer shows the period with its time zone underneath.
- Every figure in an answer has to appear in a tool result, 0 and 1 included; only list numbers, dates and times are exempt. If a figure appears nowhere, you get an error instead of an answer. Which metric a figure belongs to is not checked: a 0 from a different field would pass.
- If two figures differ, the assistant states both with their periods. It names a cause only when a tool supplies one, for example events still waiting to be processed. It calculates nothing itself: it reads unattributed revenue from the tool instead of forming a difference.
- If no tool provides what you ask for, the assistant says exactly that: the tool does not provide it. It does not claim that your data is missing. One example is the contribution margin as a total per campaign: “commerce” gives the contribution margin of all campaigns and, per campaign, only the product report, just like the “Commerce & products” page.
- The tenant comes from your session, not from something stated in the conversation.
- There is no memory on the server: the conversation lives in your browser, and each question carries at most the last twelve messages. Earlier answers that contained individual completions or journeys are not sent again. If you follow up, the assistant reads the rows afresh, as they are today. If you switch tenant, the conversation starts afresh.
- The request runs through Vertex AI in Google Cloud, region europe-west3; the region is also shown in the portal. Google processes it as a processor and does not train on it. If abuse is suspected, Google may keep requests for up to 90 days and review them.
- Google receives your question, the conversation and the results of the tools. Those are mostly metrics; if you ask about individual completions or journeys, those rows too, at most 300 per question. Order numbers and journey identifiers from your data only leave as codes that apply to this one request; the portal puts the real numbers into the answer afterwards. If you type a number that does not exist in your data, it leaves as you typed it. Orders your shop reported without consent are not listed individually; the assistant gives their count, below five only as “fewer than 5”. Raw events never leave.
- Email addresses, phone numbers and IBANs in your question are replaced by placeholders before it is passed on. Names are not: there is no pattern that reliably recognises a name. So keep your customers’ names out of the question.

> **Note:** Answers about the product name the URL of the article they come from. If the model finds no matching article, it says so rather than inventing one.

> **Legal note:** Where the answers contain data about your visitors and customers, Google processes it as a further processor; the entry is in the annex to the DPA. If you do not use the assistant, none of it goes to Google.

## Also asked

- How many questions can I ask?
- AI chat in the portal
- Ask questions about the data
- What can the assistant do?
- Can I ask the chat how to use the product?
- What data goes to Google when I ask?
- Are order numbers passed to the language model?
- Why does the assistant not list every order?
- Can I ask questions in the demo?
- Which model does the chat use?
- Which time zone does the chat use?
- Why does the chat give a different figure for today than the overview?
- Why do I get an error instead of an answer?
- Can the chat tell me the contribution margin?
- Why does the assistant say a tool does not provide something?
- Can the chat tell me visits and time on site?
- Why does the chat not name individual pages?

## Related

- https://bench.metrunio.com/en/help/mcp-anbinden
- https://bench.metrunio.com/en/help/tagesgrenzen-und-zeitzone
- https://bench.metrunio.com/en/help/csv-export
- https://bench.metrunio.com/en/help/berichte-ueberblick
- https://bench.metrunio.com/en/help/demo-ansehen