Monday morning reporting often starts with three tabs, two exports, and someone asking which number is current. AI can help assemble that information into a dashboard. The useful part is deciding what the dashboard should answer before asking it to make charts.
Choose the decision first
Imagine a small service company tracking inquiries, estimates, and booked jobs. Its owner wants to know where follow-up is slipping. A dashboard full of website traffic would look busy without answering that question. A better first version would show inquiries awaiting a response, estimates still open, and booked jobs by source.
Write down the question in one sentence. Then define the numbers that support it. Does an inquiry mean a unique person or every form submission? Does a booked job require a deposit? Those definitions belong beside the report so everyone reads it the same way.
Bring a reference and a small data sample
A screenshot can communicate preferences such as a compact summary row or a table beneath a chart. Treat it as a layout reference. Ask for an original design using your branding and your actual reporting needs.
Provide a recent export with a manageable number of rows, explain each column, and identify the authoritative source. With Sites and suitable data access available, you can ask for a working page. A file-based version is a snapshot until you add and verify a refresh process; a moving chart does not make the underlying data live.
- List the fields needed to answer your question and remove unrelated personal data.
- Ask AI to inspect missing values, duplicate records, and conflicting dates before building.
- Agree on metric definitions, then build the smallest useful dashboard.
- Compare the displayed totals with the original export and open a few underlying records.
Give the dashboard a reality check
Suppose your sample has 40 inquiries and 10 booked jobs. A simple inquiry-to-booking ratio is 25%, but it can be misleading if those bookings came from last month’s inquiries. Ask the assistant to show which dates and records each calculation includes. If you want a true conversion measure, define a group of inquiries and track what happens to that same group.
Check a date range with no records, a missing source label, and a long customer name. Make sure chart values are accessible on a phone as well as with a mouse. Every page should show its reporting period and when its data was last updated.
Make the next version earn its place
Use the first version for a real reporting conversation. Note the question it cannot answer, then add the field or view needed for that question. This keeps the project focused and gives you a practical reason for each change.
If you later connect a CRM or spreadsheet, define who can view customer details and what happens when a refresh fails. An old but clearly dated report is more useful than a polished page that quietly presents stale numbers as current.
A prompt to make it your own
Use @Sites to build a reporting dashboard for [team or business]. The decision it should support is [question]. Use the attached screenshot only as layout inspiration and create an original design. My data is [file or authorized connection]. First inspect the fields, duplicates, missing values, and date coverage. Use these metric definitions: [definitions]. Show [three to five metrics], a relevant chart, and the underlying records. Label the reporting period and last data update. If this uses an uploaded file, label it as a snapshot. Reconcile the displayed totals with the source and report any unresolved discrepancies before calling it complete.
Start with one decision and a few numbers you can trace. Add more only when the first version is useful.
Use-case inspiration: 9x’s ChatGPT Work video and companion article. Examples, instructions, and prompts written for AI Prompts.
Use the prompts with the tools and permissions available in your AI environment. Some workflows require connected accounts or uploaded files.