GeminiMultimodalLevel 3: Workflows

How Do You Get Gemini to Reason Across Images, Files and Text?

Attach the files together with Add files, name what each one is, and ask a question that needs all of them, such as where they disagree. Gemini can read images, PDFs, spreadsheets and most other file types in one prompt, up to 10 files at a time.

Prompt to copy
I've attached [number] files:
- [file 1 name]: [what it is, e.g. signed quote]
- [file 2 name]: [what it is, e.g. final invoice]
- [file 3 name]: [what it is, e.g. photo of the delivered goods]

Compare them and list every place they disagree on [items, prices, dates, quantities].

For each one, give the value from each file and where you found it (page or part of the image). Say how sure you are when you're reading from an image. Don't list things that match.
  1. 1Attach up to 10 files
  2. 2Name each file's role
  3. 3Ask where they disagree
  4. 4Check its citations
Rule of thumb, not a Google chart: name each file in the prompt, so Gemini knows what role each one plays.

Try it now

  1. 1Choose two or three related files, such as a quote, an invoice and a photo.
  2. 2Click Add files, then Upload Files, and select them. Add from Drive also works if Workspace is connected.
  3. 3Paste the prompt above with your file names, and click Submit.
  4. 4Check each difference it reports against the originals. You'll know it worked when every difference it lists is real, and it hasn't missed one you already knew about.

Use it when

Checking a delivery photo against an order, comparing a slide with the report it cites, reading a screenshot of an error along with the code, or matching a floor plan against a contractor's quote. Use it wherever the answer depends on two sources being read together.

Skip it when

When the data exists as text or numbers already, send that, because reading figures from an image adds a chance of error. Also skip it for anything you'd need to be exact, such as measurements from a photo, unless you verify them. Gemini Apps can make mistakes, and Google says to double-check.

Why it works

The value comes from comparing things you'd otherwise check by eye: a contract against an invoice, a photo against a spec. Google's upload page allows up to 10 files per prompt, with most types up to 100 MB. A Google AI Pro or Ultra plan raises upload limits, and the context window (1 million tokens, against 32k with no plan) is what lets several long files fit at once. On the API, Google says images cost 258 tokens if they're 384 pixels or smaller on each side, and larger ones are split into tiles at 258 tokens each, so a handful of photos is usually a small share of the window.

Example

Say a contractor's final invoice is higher than the quote. Attach the quote PDF, the invoice PDF and a photo of the work. Ask Gemini to list every line that differs, and whether the photo shows the items billed. You've turned an hour of reading across files into a short list. A photo can't confirm what was installed behind a wall, so treat that part as a question to ask the contractor.

Going further

Prompt order isn't settled across Google's docs. For a single image on the API, the image page says to place the text before the image. The PDF and video pages say to place the text after. With several files, naming each one matters more than order (rule of thumb, not a Google figure). In the app, you can also ask for a chart from an uploaded spreadsheet. Click Edit below it, change the chart type or labels, then click Update.

Common mistakes

  • Attaching files with names like scan1.pdf and no explanation. Say what each file is, so Gemini can tell which one is meant to govern.
  • Asking "what do you think?" A broad question gets a broad summary. Ask where the files disagree.
  • Uploading more than 10 files. Each file in a folder counts toward that limit, according to Google's help.
  • Believing a small number it read from a photo. Zoom in and check it yourself.

Sources

Checked 6 November 2026. Products change often, so confirm details with the vendor.

Related

Beyond the course