How to Create a Data Sources Slide with AI: Step by Step
A data sources slide is the one page in a data deck that says where every number came from. It lists the files, the date range, the definitions you chose, and the known limits. Build it by naming each source, stating its coverage, and pinning an as-of date that does not silently change.
What a data sources slide is, and what it is not
Every data deck makes claims. The sources slide is where those claims become checkable.
It is not a bibliography, and it is not a legal disclaimer. It answers four questions a skeptical reader will ask within ten seconds of doubting your chart:
- Which file is this? Not "our CRM" but the specific export, with its name and pull date.
- What period does it cover? A quarter-over-quarter chart built on a partial quarter is a different chart.
- What did you decide? Whether refunds are netted out, which date field defines the period, how you handled unmatched rows.
- What do you already know is wrong? Missing regions, a system migration mid-period, a category that was renamed.
Teams skip this slide because it feels like admin. It is the opposite. A deck that cannot answer those four questions gets re-litigated in the meeting instead of decided.
What belongs on the slide
You do not have to invent the field list. Google's published guidance for describing a dataset already names the properties that identify one, and most of them map cleanly onto a slide.
The Dataset structured data documentation treats two properties as required. One is name, defined as "a descriptive name of a dataset." The other is description, "a short summary describing a dataset." The recommended set adds creator, license, temporal coverage, spatial coverage, distribution format, and the variables measured.
Translated into slide rows:
| Field on the slide | Where it comes from | Example |
|---|---|---|
| Source name | The export or system, named specifically | orders_2026Q2.csv, pulled from the order system |
| What it contains | One line, not a paragraph | Line-level orders with SKU, quantity, and net amount |
| Period covered | Start and end, explicitly | 2026-04-01 to 2026-06-30 |
| Owner | Who to ask when a figure looks wrong | Revenue operations |
| Format | How it arrived | CSV export, 41 columns |
| Definitions applied | The judgment calls you made | Period cut by invoice date; refunds netted out |
| Known limits | What the reader should allow for | Two regions migrated systems in May |
One guideline from that documentation is worth borrowing directly. It recommends unique names for distinct datasets, using "Snow depth in the Northern Hemisphere" rather than two files both called "Snow depth." The same rule saves decks. Two rows on your slide both reading "sales export" is how the wrong figure survives review.
What you need before you start
Three things, and you probably already have two of them.
- The files themselves, with their original names and the date each was pulled.
- A note of your own decisions — which date field, which exclusions, how blanks were treated.
- The known caveats, which usually live in someone's head rather than in a document.
The third one is the reason this slide is hard to write after the fact. Reconstructing why you excluded 40 rows three weeks later is guesswork.
How to build the slide manually
The manual route works fine. It is just easy to get subtly wrong.
Option 1: A plain table on one slide. Insert a table with the seven rows above as columns, one row per source. This is the version most decks should use, and it takes about ten minutes if your notes are current.
Option 2: A footnote-per-chart approach. Put a small source line under each chart instead of gathering them on one page. This reads well for a short deck. It scales badly, because the same source gets restated five times and the five statements drift apart.
Option 3: An appendix page plus a footer date. Keep the detail at the back and put the as-of date in the slide footer so it appears on every page.
For Option 3, the mechanics are documented. Microsoft's guidance says that in PowerPoint you can "add or remove automatic slide numbers, hide numbering on the title slide, and move slide numbers." It adds that you can also put the date and time on slides.
There is one setting that matters more than it looks. When adding the date, you choose between Update automatically and Fixed. Microsoft gives a precise reason for choosing Fixed. Doing so means "you can easily keep track of the last time you made changes to it."
For a data deck, Fixed is almost always the right answer. An auto-updating date tells the reader when they opened the file, not when the data was pulled. Those are different facts, and confusing them makes the deck less trustworthy rather than more.
How to create a data sources slide with AI
Step 1: Upload the files the deck is built on
Drop the same exports you analyzed into Powerdrill Bloom — Excel, CSV, PDF, and documents are all accepted. The point is that the agent reads the actual files. Row counts, column names, and date ranges then come from the data rather than from memory.
Step 2: Ask for the sources slide in natural language
Describe what you want the slide to state. For example: "List every file I uploaded with its row count, column count, and the min and max date in the date column. Then note any columns with missing values above 5%."
That last clause is the part worth keeping. The known-limits row is the one people leave blank, and it is the one a reader most wants.
Step 3: Review the caveats, then export
Read the generated limits line by line and correct anything only you know — a system migration, a renamed category, a deliberate exclusion. Then export the slide alongside the rest of the deck. Set a scheduled task if this deck rebuilds monthly, so the as-of date and row counts refresh with the numbers.
Common mistakes
Writing "internal data" and stopping. It reads as though you would rather not say. Name the export.
Letting the as-of date auto-update. Covered above, and it is the single most common error on otherwise careful decks.
Listing sources you did not actually use. A source on the slide that contributed nothing to the analysis is noise that invites the wrong question.
Describing limits as strengths. "Data is directional" is not a caveat, it is a hedge. Say which regions are missing and for which weeks.
Rebuilding the slide by hand each cycle. If the deck is monthly, the sources slide should regenerate from the files rather than get edited in place. Hand-edited numbers drift from the data they describe.
A sources slide supports the rest of the deck rather than replacing its craft. If your slides are hard to read before anyone questions the numbers, fix readability first.
Conclusion
The sources slide is cheap insurance. It costs one page and about ten minutes, and it converts every "are you sure about that number" into a two-second glance.
The reason it usually goes missing is that writing it by hand means reconstructing facts about files you closed days ago. Generate it from the files instead, then spend your time on the caveats only you know.
Build one from the exports you already have. Try Powerdrill Bloom free and get the analysis, the charts, and the sources page from one upload. For recurring decks, Excel to PPT and the AI report generator cover the same files.
Microsoft and Google guidance quoted here was retrieved on September 2, 2026.
Frequently asked questions
What is a data sources slide?
It is a single slide listing every dataset behind a presentation. Each source is named, its period stated, the definitions applied, and the known limits noted. It makes the deck's numbers checkable without a follow-up email.
Where should the data sources slide go in a deck?
Either directly after the title slide or as the first appendix page. Put the as-of date in the footer so it travels with every slide, and set that date to Fixed rather than automatic.
What is the difference between a sources slide and a methodology slide?
A sources slide says what data you used. A methodology slide says what you did to it. Short decks usually combine them into one page with two blocks.
How much detail is too much?
One line per field is enough. If a row needs a paragraph, that paragraph belongs in an appendix and the slide should carry a pointer to it instead.
Can AI generate a data sources slide from my files?
Yes. Upload the exports, then ask for each file's row count, column list, date range, and missing-value rate in natural language. Review the limits section yourself, because context like a mid-period system migration is not visible in the data.