Attach and analyze files

Turn ChatGPT into a document analyst

ChatGPT reads and analyzes files you attach to the conversation: documents, spreadsheets, presentations, images, and, on paid plans, audio recordings. You upload, ask your questions, and the analysis focuses on the actual content of the document rather than a summary you might have provided. Document, spreadsheet, presentation, and image uploads are available on both Free and paid plans, with limits specific to each tier: on Free, expect a limit of three files per day.

Uploading is straightforward: click the + button in the input area, then select Add photos & files, or drag and drop directly into the input box. Write your prompt in the same message and send. Files you import or that ChatGPT creates are stored in your Library, or in Space if your account has access (Lesson 3), where you can reuse them without re-uploading.

Supported formats and limits

Google Docs files do not import directly: export them as PDF or DOCX first. Regarding size, a document or presentation can be up to 512 MB, within a limit of 2 million text tokens per file; a spreadsheet, about 50 MB; an image, 20 MB; an audio file, 512 MB. You can upload up to 80 files every three hours, though OpenAI may lower these thresholds during peak times.

These caps don’t tell the whole story. OpenAI warns that an accepted file may still be too long, too complex, or too poorly structured to be analyzed in full. If a response seems to ignore part of the document, ask it to target a specific section, pages, sheets, or columns, or split the file.

PDFs: the most common use case

Three approaches cover nearly all needs. First, summarization, where value depends entirely on prompt precision:

[Attached PDF] “Summarize this document in 10 points. Highlight key figures and main conclusions.”

Next, targeted questioning, which treats the document as a knowledge base: “In this contract, what are the conditions for early termination? Are there penalties?” You get in thirty seconds what a careful reading would have taken fifteen minutes, and you can follow up with more questions. Ask it to cite the relevant passage: verification becomes immediate.

Finally, structured extraction, which converts continuous text into usable data:

[Attached PDF] “Extract all dates mentioned in this document and present them in a chronological table with the associated event.”

Spreadsheets: analysis goes beyond reading

With an Excel or CSV file, ChatGPT doesn’t just read cells: it writes and executes Python code to calculate, group, and plot charts (Lesson 8). A request like “Analyze this sales file. Give me total revenue, the top 5 products, the monthly trend, and a chart of sales evolution” triggers real processing, not a rough estimate. Review the displayed code and assumptions before trusting the result.

Data cleaning follows the same mechanism: “This file contains duplicates and missing values. Identify the issues and propose a cleaned file.” And nothing stops you from sending multiple files in the same conversation to cross-reference them:

[File 1 + File 2] “Here is our customer list and our order list. Cross-reference them to identify customers who haven’t ordered in 6 months.”

Presentations, audio, and video

A PowerPoint presentation can be summarized slide by slide, but it can also be critiqued: “Is the narrative flow coherent? Is there too much text on the slides? Suggest improvements.” This second use is the most valuable, giving you the outside perspective you always lack on your own work.

On paid plans, an audio recording can be transcribed, summarized, or queried: “Summarize this meeting: decisions, open questions, next steps.” OpenAI warns that transcription may contain errors and speaker identification may be unreliable: verify important details in the recording. Videos can also be attached, even on Free, but their analysis may be partial.

Projects for reference documents

If you frequently work with the same documents, store them in a project (Lesson 3) rather than in each conversation: all conversations in the project can use them. The number of files per project depends on your plan: 5 on Free, 25 on Go and Plus, 40 on Pro, Business, Enterprise, and Edu.

Your turn

A PDF, then a spreadsheet. Attach a PDF of your choice and request a five-point summary, then ask three specific questions requiring cited passages, and finish by extracting key information into a table. With a spreadsheet, first have it describe the structure (sheets, columns, data types), then request a trend analysis and a chart. Finally, verify a figure from the result directly in your file.

Official sources: Uploading files and audio to ChatGPT (OpenAI Help Center: plans, formats, sizes, quotas, files per project) · Data analysis with ChatGPT (spreadsheets, Python, partially analyzed files) · ChatGPT Image Inputs FAQ (images and videos) · Using Library to manage files in ChatGPT

Key takeaways

• ChatGPT reads PDFs, Word docs, spreadsheets, presentations, and images on all plans; audio on paid plans
• Always include a clear prompt with your file, and ask for cited passages
• An accepted file isn’t necessarily read in full: target sections or split the file
• For spreadsheets, ChatGPT executes Python to calculate and plot charts
• Imported and created files are stored in the Library; reference documents go in a project