Give ChatGPT Your Files

Upload documents to ChatGPT and extract, summarize, and transform them.

Introduction

30 Second Summary

You have documents all over your life: job postings, class readings, meeting notes, reports. When you need something specific from one of them, you sit down and read through it yourself.

In this project, you will upload a document to ChatGPT and turn it into something genuinely useful. Along the way, you will learn to verify whether ChatGPT's answers actually come from your source material.

What You'll Build

By the end, you will hand ChatGPT a document from your own life and walk away minutes later with a polished deliverable built entirely from that source.

By the end of this project, you'll have:

A side-by-side comparison showing the difference between ChatGPT guessing from general knowledge and ChatGPT working from your actual source material.
A file-based conversation where you have extracted, located, summarized, and transformed information from an uploaded document.
A genuinely useful deliverable (study guide, action plan, executive summary, or content brief) built entirely from your own file.
Secret Mission: Bring a second file from your own life and build a multi-file comparison that reveals what two documents share and where they differ.

Are there any prerequisites?

This project builds on Projects #1 through #5 in the series. You should be comfortable using ChatGPT for everyday tasks before starting.

Before We Start

Before we dive in, take a moment to think about what you are building and why it matters. Consider which document from your life would be most useful to hand to ChatGPT. A job posting, class notes, a report, a contract, or even an article you saved as a PDF.

See What ChatGPT Does Without Your File

You have a document ready to hand to ChatGPT. Before you upload it, there is something important to see first.

Without your source material, ChatGPT can only draw on general knowledge to answer your questions. Experiencing that generic response firsthand makes the improvement obvious once you give it the actual file in the next step.

In this step, get ready to:

Ask ChatGPT for help with a topic your document covers, without uploading the file.
Flag one or two details in the response that are guesses rather than sourced information.

Ask ChatGPT for help without your file

You already know what your document is about. Now you will ask ChatGPT about that same topic, but without giving it the file. This forces ChatGPT to rely on general knowledge alone.

Open chatgpt.com in your browser and sign in to your ChatGPT account.
Start a new chat.
Type a question asking ChatGPT for help with the topic your document covers. Do not upload the file.

What kind of question should I ask?

Ask about something your document would answer. For example:

If you have a job posting, try: "I'm preparing for a Data Analyst interview. What should I focus on to prepare?"
If you have class notes or a reading, try: "I'm studying for my exam on this topic. What are the key concepts I should review?"
If you have a report or article, try: "Can you summarize the main findings of a report like this?"

The goal is to see what ChatGPT produces from general knowledge alone, so you can compare it to a file-grounded answer later.

Before you send your message, take a guess. How specific do you think ChatGPT's answer will be about the details in your particular document?

Send the message and read ChatGPT's response.

The response probably sounds helpful. It might list reasonable advice or cover relevant topics. But look at the specifics. Nothing in that answer comes from your actual document. Every point is drawn from general knowledge.

Flag the generic details

ChatGPT's answer might feel useful on the surface. The real question is whether it contains anything specific to your document, or whether it could have been written for anyone asking about the same general topic.

Read through the response carefully.
Pick one or two details that feel generic. These are points where ChatGPT is clearly guessing rather than pulling from your source material.
Keep this chat open. You will compare it directly to a file-grounded response in the next step.

What counts as generic?

Look for advice that could apply to anyone. If you asked about a job posting, did ChatGPT mention the exact tools, qualifications, or requirements from that posting? Or did it list common skills for that type of role in general? The more interchangeable the advice feels, the more generic it is.

You now have a generic ChatGPT response on screen. You have flagged at least one or two details that are guesses rather than information sourced from your document.

You have seen ChatGPT's best attempt without your source material. In the next step, you will upload your file and watch how dramatically the answers change.

Upload Your File and See the Difference

In Step 1, ChatGPT gave you a reasonable answer drawn entirely from general knowledge. It sounded helpful, but it had no way to reference the specific details inside your actual document.

This step is where the transformation happens. You will upload your file, re-ask your question, and see the same kind of prompt produce a dramatically different answer.

In this step, get ready to:

Upload your document to the chat.
Re-ask your question and compare the file-grounded answer to the generic one.
Get a structured overview of what ChatGPT sees in your file.

Upload your document

You have seen what ChatGPT does when it has to guess. Now it is time to give it the real thing.

In the same chat where you received the generic response, click the + button in the message composer at the bottom of the screen.
Select Add photos or files.
Choose the document you picked in Step 1.

Your file appears as an attachment in the composer. Do not send a message yet.

What file types can I upload?

ChatGPT accepts most common document formats, including PDF, DOCX, PPTX, XLSX, CSV, and TXT. Each file can be up to 512 MB.

Google Docs files (.gdoc) are not supported. If your file is in Google Docs, export it to PDF or DOCX first.

How many files can I upload?

Free users are limited to 3 file uploads per day. Plus users can upload up to 80 files every 3 hours. This project only needs one or two uploads, so either plan works.

File not showing up as an attachment?

Make sure the file is in a supported format. If you are working from a Google Doc, export it first using File > Download > PDF Document in Google Docs.

If you are on the Free plan and your upload fails, you may have reached the daily limit of 3 uploads. Try again tomorrow or use a different file you have not uploaded today.

Help me upload a file to ChatGPT

Re-ask your question with the file attached

With your file attached, you need to tell ChatGPT to use it. A generic question without referencing the document will still pull from general knowledge, even with the file sitting right there.

Type a new message that directly references the uploaded file. Adapt one of these examples to fit your document:

Job posting: "Based on the job posting I just uploaded, what specific skills and qualifications should I focus on to prepare for this interview?"
Class notes or reading: "Based on the document I just uploaded, what are the most important concepts I should understand for an exam?"
Report or article: "Based on the report I just uploaded, what are the key findings and recommendations?"

Before you hit send, predict: will the response include specific details from your document, or will it sound similar to the generic answer you already have?

Press Enter to send the message.

Read the response carefully. ChatGPT is now pulling specific details: exact skills listed in your document, particular tools or technologies mentioned, specific qualifications or data points. Scroll up and compare this response to the generic one from Step 1.

What changed?

In Step 1, ChatGPT drew on its training data to give you a reasonable but generic answer. Now it is reading your actual source material and referencing the specific information inside it.

This is the core skill of the project: giving ChatGPT your information so it works from your source, not from guesses.

Ask ChatGPT what the document contains

Before you dig deeper in the next step, it helps to see what ChatGPT "sees" in your file. Asking for a structured overview tells you how it reads and organizes your document.

Type this message in the same chat: "What does this document contain? Give me a structured summary with the main sections and key points."
Press Enter to send.

ChatGPT responds with a breakdown of your document's main sections, themes, or structure. Scan through it and see if the summary matches what you know is in the file.

How does ChatGPT read your file?

On all plans except Enterprise, ChatGPT uses text-based retrieval. It extracts the digital text from your file and discards any images embedded in the document. Charts, diagrams, and photos inside a PDF or DOCX are invisible to it.

If your document has important visuals, you can upload them separately as image files (PNG, JPEG, or GIF). ChatGPT can analyze images when they are uploaded on their own.

You now have two responses side by side in the same chat: the generic answer from Step 1 and the file-grounded answer from this step. The specificity gap between them should be clear.

You can see exactly how much more useful ChatGPT becomes when it has your actual source material. Next up, you will dig deeper and extract specific details from your document.

Extract and Locate Specific Information

Your document is uploaded and ChatGPT has given you a structured overview of its contents. The specificity gap between the generic answer and the file-grounded response is already clear.

Getting an overview is just the start. The real power is being able to dig into a long document and pull out exactly the detail you need without reading every page yourself. In this step, you will send targeted extraction prompts and test whether ChatGPT can tell you when it is guessing.

In this step, get ready to:

Extract specific details from your document using targeted prompts.
Test what happens when you ask about information not in the file.
Practice verifying AI claims against your source material.

Search for specific details

A structured summary tells you what your document covers at a high level. Targeted extraction goes further. It pulls out the exact details you care about so you can work with them directly.

In your ChatGPT chat, send a message asking for a specific type of detail from your document. Be precise about what you want.

Here are example prompts you can adapt to fit your file:

"List every tool or technology mentioned in this document."
"Find any deadlines, dates, or time-sensitive requirements."
"What qualifications are listed as required versus preferred?"
"Pull out any specific numbers, metrics, or targets mentioned."

Pick one or two prompts that fit your document, adapt the wording, and send them.
Read the response carefully. Notice how ChatGPT returns specific names, values, or categories pulled directly from your file.

Why spot-check against the original?

ChatGPT can sometimes blend general knowledge with file content. A detail that sounds right might actually be something ChatGPT "knows" from training data, not something your document says.

By verifying a few details against your original file, you build the habit of catching these blends early.

Open your original document in a separate window or tab.
Pick two or three specific details from ChatGPT's response.
Search for each one in your original file and confirm it actually appears there.

If everything checks out, great. If you find a detail that does not appear in your file, you have already caught ChatGPT filling in from general knowledge.

Getting vague or incomplete results?

Try being more specific. Instead of "what are the key points," ask for a particular category of information like tools, dates, or requirements.
If ChatGPT missed something you know is in the document, quote a few words from that section and ask it to look there specifically.
ChatGPT extracts digital text from your file. If the detail you are looking for is inside an image, chart, or diagram, ChatGPT will not see it on non-Enterprise plans.
ChatGPT is not finding the details I expected from my uploaded document. What can I try?

Ask about something not in the document

You have seen ChatGPT pull real details from your file. Now you need to test the other side. What happens when you ask about something your document does not cover?

Think of a detail that would be useful to know but is unlikely to appear in your file.

For example, if you uploaded a job posting, you might ask about the salary range. If you uploaded class notes, you might ask about a topic the notes do not cover. If you uploaded a report, you might ask about data from a year the report does not include.

Before you send the prompt, pause and predict: do you think ChatGPT will tell you the information is not in the file, or will it try to answer anyway from general knowledge?

Send the prompt and read the response carefully.

Compare what happened to your prediction. ChatGPT might clearly say the information is not in the file. It might also provide a plausible-sounding answer drawn from its training data without flagging that the answer is not from your document.

If ChatGPT gave you a specific answer, send a follow-up asking it to confirm the source. Try something like: "Is that detail actually stated in the document I uploaded, or are you estimating from general knowledge?"

Why verify the source of ChatGPT's answers?

Knowing the difference between ChatGPT working from your source and ChatGPT filling in gaps from general knowledge is the most important habit you build in this step.

When you can spot the difference, you can trust the grounded answers and question the rest. This becomes especially important when you start building deliverables from your documents in the next step.

You have now extracted specific details from your document and tested what happens when you ask about information that is not in the file. You know whether ChatGPT clearly flags when it is pulling from general knowledge instead of your source.

You can now locate and verify details inside your document. Next up, you will take this a step further and transform your source material into something entirely new.

Summarize and Transform

You can now extract specific details from your document and spot when ChatGPT is drawing from your file versus filling in from general knowledge. That is a powerful skill on its own.

Extracting details is just the start. The real power of file uploads is transforming your document into something entirely new: a study guide, an action plan, an interview prep checklist. In this step, you will reshape your source material into a deliverable you can actually use.

In this step, get ready to:

Summarize your document for a different audience or purpose.
Build a complete deliverable from your uploaded file.
Verify a specific claim in the deliverable against the original source.

Summarize for a different audience

Your document was written for a specific context. Reshaping that same information for a different audience tests whether ChatGPT truly understands the content or just parrots it back.

In your chat on chatgpt.com, send one of the following prompts (pick whichever fits your document best):

"Summarize this document in 3 bullet points for someone who has 30 seconds to read it."
"Explain the key requirements in this posting as if you were telling a friend over coffee."
"Write a one-paragraph executive summary of this report."

Read ChatGPT's response.
Open your original document in a separate tab or window.
Compare the summary to the source. Check whether it captures the most important points and whether it left out anything critical.

What is transformation?

Transformation means reshaping information without changing its meaning. A 10-page report becomes 3 bullet points. A technical spec becomes a plain-language explanation.

The content stays faithful to the source while the presentation shifts. That is what makes it different from asking ChatGPT to generate something from scratch.

Build your deliverable

Summaries are useful, but the biggest payoff is turning your document into something entirely new. This is where file uploads go from convenient to genuinely powerful.

Pick the path below that best matches your uploaded file, then send that prompt to ChatGPT:

Career (job posting): "Based on this job posting, create an interview prep checklist. For each required skill, note whether it is something I should prepare to demonstrate with examples, study up on, or highlight from my background."
Student (reading/class material): "Turn this into a study guide with key concepts, definitions, and 5 practice questions I could use to test myself."
Business (report/memo): "Create an executive summary with three specific recommendations based on the findings in this document."
Personal (any document): "Organize the key action items from this document into a prioritized to-do list with deadlines where mentioned."
Creator (article/source material): "Create a content brief based on this source. Include the main argument, key supporting points, potential angles for a new piece, and any gaps I could fill."

Read through the deliverable ChatGPT produces.

You should see a structured, detailed output that draws on specifics from your uploaded file. Notice how the deliverable references details that only your document contains.

Make it your own

These prompts are starting points. Feel free to modify them or write your own from scratch. The more specific you are about the format and purpose you want, the more useful the deliverable will be.

Deliverable feels too generic?

If the output reads like generic advice rather than something built from your file, try adding a follow-up: "Use only information from the document I uploaded. Do not add general advice that is not in the file." This pushes ChatGPT to stay grounded in your source material.

help me get ChatGPT to build a deliverable that stays grounded in my uploaded document instead of adding generic advice

Verify the deliverable against the source

The deliverable looks polished, but you need to confirm it actually reflects what your document says. ChatGPT can blend source material with general knowledge without telling you.

Before you check, take a guess: do you think every detail in the deliverable came directly from your file, or do you suspect ChatGPT added something from general knowledge?

Pick one specific claim or detail in the deliverable.
Open your original document in a separate tab or window.
Search for that claim in the source. Check whether it actually appears and whether it is accurate.

If you find a detail that seems off or unfamiliar, go back to the ChatGPT chat and ask it directly where in the document that information appears.

Type a prompt like: "You mentioned (the specific detail) in the deliverable. Where exactly does that appear in the document I uploaded?"
Read the response carefully. If ChatGPT can point to a specific section, the claim is grounded. If it hedges or admits the information is not in the file, you have caught a gap.

Why does this matter?

ChatGPT is trained on enormous amounts of general knowledge. When it builds a deliverable from your file, it sometimes fills in small gaps with plausible-sounding details that are not actually in the source.

This is not a bug. It is a feature you need to manage. Building the habit of spot-checking one or two claims keeps your deliverables trustworthy.

You now have a deliverable built from your source material, with at least one detail confirmed against the original document. Whether ChatGPT got it right or you caught it filling in from general knowledge, you have practiced the most important skill of working with file uploads: verification.

Not sure what to check?

Look for the most specific detail in the deliverable. Numbers, names, percentages, or exact requirements are the easiest to verify. If a claim is too vague to check ("strong communication skills are important"), pick something more concrete.

help me figure out which claim in my ChatGPT deliverable to verify against my source document

You have turned a raw document into something genuinely useful. Ready for a bonus challenge? The Secret Mission takes this further with a multi-file comparison.

Secret mission

Build a Multi-File Comparison

You have turned a single document into a useful deliverable. But what happens when you give ChatGPT two documents at once? Upload a second file and build a structured comparison that reveals what your documents share and where they differ.

Clean Up Your Resources

This project runs entirely in your browser on chatgpt.com. There are no paid cloud resources or ongoing costs to worry about.

Your uploaded files are automatically saved to your ChatGPT Library. Decide whether to keep your files for future reference, remove sensitive documents, or delete everything.

Resources you used:

Uploaded documents saved to your ChatGPT Library (your original file and your second file from the Secret Mission).
Chat conversation(s) on chatgpt.com containing your prompts, responses, deliverable, and multi-file comparison.

Keep everything running

No action needed. Choose this if you want to reference your uploaded files and chat history in future conversations.

Your files stay in your Library and can be added to new chats using the + menu and selecting Add from library.
Your chat history stays in the sidebar for you to revisit anytime.
You can check how much Library storage you have used under Settings > Storage. Free users get approximately 500 MB. Plus users get 20 GB.

Pause - I'll come back to this later

Remove any sensitive documents from your Library, but keep your chat history so you can pick up where you left off.

Open your Library from the left sidebar.
Review the files saved there. If any contain sensitive information you do not want stored, select the file and click Delete (or the trash icon next to it).
Keep your chat conversations in the sidebar. You can return to them later and continue asking questions about any files still in your Library.

Deleting a chat does not delete your Library files

Chats and Library files are managed separately. If you delete a chat, the files you uploaded during that chat remain in your Library until you delete them manually.

Delete - I don't want to use this again

Remove all files and chat history from this project.

Open your Library from the left sidebar.
Select each file you uploaded during this project and click Delete (or the trash icon next to it). Repeat for both your original document and the second file from the Secret Mission.
Return to your chat history in the sidebar. Hover over each conversation from this project, click the three dots next to the chat title, and select Delete. Confirm when prompted.

What happens after you delete?

Deleted files and chats are removed from your view immediately. They are scheduled for permanent deletion from OpenAI's systems within 30 days.

Remember that deleting a chat does not delete your Library files. You need to delete them separately from the Library tab.

Nice Work!

Nice work! You just turned a document from your own life into something genuinely useful by teaching ChatGPT to work from your source material instead of guessing.

You've learned how to:

See the difference between ChatGPT guessing from general knowledge and ChatGPT working from your actual source material.
Extract specific details from a document, verify whether ChatGPT's answers are actually grounded in the file, and catch it when it fills gaps from general knowledge without flagging it.
Transform document content into a genuinely useful deliverable like a study guide, prep checklist, executive summary, action plan, or content brief.
Secret Mission: Compare multiple files in a single conversation, build a structured comparison, and spot-check cross-document claims against both sources.

Ready to quiz yourself?