Ask about one dataset, starting from questions built for it

The hardest question is the first one

You have just uploaded a sales export or connected a Google Analytics report. You know there are answers in there, but an empty box asking "What would you like to know?" is not much help when you are not sure what the data can tell you yet.

DashboardGPT now starts the conversation for you. Open any file, sheet or report and the assistant beside it suggests questions that fit what is actually in it: your money columns, your categories, your dates. Choose one and you have a chart in seconds. Then follow up in your own words.

New to asking questions of your data? Chat with your data: the complete guide covers the basics first.

Ask on the page of the data you mean

Every file, sheet and report you bring in has its own page under Datasets. The assistant on that page is about that data and nothing else.

• Answers from the data you are looking at. Questions asked on a page always use that page's data, even when your workspace holds a dozen other files. The box you type in shows which data it is using.
• A chat for each one. Each file, sheet or report gets a chat of its own. Move to another one and back again while you work, and your conversation is where you left it. Earlier chats stay on the Conversations page.
• A fresh start with suggestions. Until you ask something, the panel shows questions built for that data under Ask about, followed by its name.

Step by step

1. Open Datasets and choose the file, sheet or report you want to explore.
2. Look at the assistant panel: it lists up to five questions built for that data. Choose one and it is answered straight away.
3. Prefer to tweak the question first? Choose Ask about this at the top of the page. It puts one of the suggestions in the box for you to edit, and a different one each time you click.
4. Follow up in your own words: "Only last quarter", "Show it as a bar chart", "Which of these grew fastest?"

Where the suggestions come from

The questions depend on what kind of data you are looking at.

• Google Analytics and Search Console reports suggest the questions that come with the report you picked, such as which channels bring the most sessions or which queries bring the most clicks. They are the same questions shown under Ask about this data on the report's page. A custom report suggests questions that suit any website report.
• Uploaded files and Google Sheets get questions built from their columns: what you measure, what you can break it down by, and when it happened.

Questions built from your columns follow a few common-sense rules, so they read like something an analyst would ask:

• Money comes first. Revenue, sales, cost or price lead the way, then counts such as orders, visits or sign-ups.
• Rates are averaged, never added up. A conversion rate or a score is asked about as an average, because a total of rates means nothing.
• Labels stay labels. Numbers that only name something, such as a zip code, a phone number or a year, are never totalled.
• Breakdowns you can read. Categories with a manageable number of values, such as region or plan, are preferred to ones with hundreds.
• More than totals. Alongside "Show total revenue by region" you will see questions such as "How did revenue change month by month?", "What are the top 5 products by revenue?", "Which region grew the most in revenue?" and "Compare revenue and cost by region".

Clear column names make better suggestions. How to chat with a CSV file has a two-minute checklist for preparing a file before you upload it.

Suggestions written by your own AI provider

If your organisation connects its own AI provider, such as OpenAI or Anthropic, that provider writes the suggestions for your uploads, Google Sheets and custom Google Analytics and Search Console reports. It reads what your columns mean and a few example values, so its questions can be more specific to your business than rules alone. Ready-made Google Analytics and Search Console reports keep their own questions.

1. An owner or admin opens Settings, then AI.
2. They add your organisation's AI provider and its key.
3. Open any file or sheet. The first time, the suggestions can take a few seconds while your provider writes them.

• Written once, kept for later. Your provider writes the suggestions the first time someone opens the data, and DashboardGPT keeps them, so coming back doesn't ask your provider again.
• Always something to ask. If your provider doesn't answer in time, you see the questions built from your columns instead.
• Only your provider. Without your own provider, the suggestions come from the rules above.
• A gentle nudge. When the suggestions on an upload or Google Sheet were built from columns because no provider is set up, a note under them says so. Owners and admins get a link to Settings, and everyone else is asked to check with an owner or admin.

To write suggestions, your provider sees the name of the data, a description of its columns with a few sample values and the number of rows, not the whole file. New to using your own provider? How to get an AI API key walks you through it, and bring your own key explains why teams choose it.

Everywhere else in the workspace

Pages that aren't about one file, sheet or report, such as the overview, the Datasets list and Insights, suggest general questions that fit any data, such as "What are the main trends in my data?" and "What should I look at first?" On a canvas, select a chart and ask about it to keep working with the data behind it.

Ready to see what your own data has to say? Sign up and upload a file, compare what each plan includes on the pricing page, or contact us if you would like a hand getting your team started.