Claude can only work with what it can see, and a file puts your real material in front of it.

Claude learned from a large amount of text during training, but it has never seen your private documents, your latest report, your company handbook, or the photo on your phone. If you ask about them without sharing them, Claude can only guess or give general advice. When you upload the file, Claude can read the actual content and answer questions about it, summarize it, compare it with other material, or help you rewrite it.

Uploading also saves effort. Copying and pasting long text is slow, and it loses tables, charts, and layout. A file keeps the original in one piece. It also lets you share things that cannot be pasted as plain text at all, such as a photograph, a screenshot, or a scanned page.

There are a few easy routes, and they all lead to the same result.

In the Claude apps, you will find an add button near the message box, often shown as a plus sign. Choosing it opens a menu where you can pick an option to add files or photos from your device. You can also drag a file straight onto the chat window and drop it in, and you can paste an image from your clipboard directly into the message box. On phones, you can usually take a photo or choose one from your gallery.

Exact menu names and steps can change with updates and differ between web, desktop, and mobile, so if something looks different, check the current help center. The general idea stays the same: attach the file to your message, write what you want, and send.

Claude handles many common document and image types.

According to Anthropic’s help center at the time of writing, Claude can work with common documents such as PDF, Word documents, CSV and text files, HTML, ODT, RTF, EPUB, and JSON, as well as spreadsheets in the XLSX format. For images, it supports the widely used formats JPEG, PNG, GIF, and WebP. Some features, such as spreadsheet uploads, may require particular account features to be switched on, so check the help center if an upload is refused.

Claude does not always see a file the way you do on screen.

This is one of the most important ideas to understand. For many document types, Claude extracts the text and works from that text. It may not see embedded pictures, charts, or fancy layout inside, for example, a Word document. The help center notes that for non-PDF documents, only text is extracted, so images embedded in them cannot be read.

PDFs are treated more richly. For shorter PDFs, Claude can look at both the text and the visual elements, such as charts, diagrams, and graphics. According to the help center, this visual reading applies to PDFs of up to about one hundred pages, while longer ones up to about one thousand pages are handled as text only. Standalone image files are, of course, viewed as images.

flowchart TD
    F["You upload a file"] --> T{"What type is it?"}
    T -- "Image" --> I["Claude views the picture"]
    T -- "PDF, short" --> P1["Reads text and visuals"]
    T -- "PDF, long" --> P2["Reads text only"]
    T -- "Other document" --> D["Extracts text only"]
    T -- "Spreadsheet or CSV" --> S["Reads cells and structure"]
        

Every upload system has ceilings, and knowing them prevents surprises.

Anthropic’s help center lists limits on things such as the number of files per chat, the size of each file, the number of pages in a PDF, and the size of images. These numbers have changed over time and can differ by plan, device, and where you upload, so always treat any number you read, including in this guide, as a snapshot and confirm it in the current help center.

At the time of writing, the help center describes up to twenty files per chat, a maximum of about one thousand pages for a PDF, and a cap on image dimensions. For files added to a project, it describes a per-file size limit that is lower than the chat limit, with the total content of project knowledge bounded by how much Claude can hold, unless a retrieval feature extends it. If you exceed a limit, you will usually see an error message telling you the file is too large.

Passing the upload check is not the same as fitting in Claude’s working memory.

Even when a file uploads successfully, Claude can only consider a certain amount of text at one time. This working space is called the context window. Your uploaded files, your messages, and Claude’s replies all share it. A large file may upload fine but use up a big part of the space, leaving less room for a long conversation.

This is why very large documents can lead to messages about length limits, or answers that seem to overlook parts of the file. The solution is not to give up but to be smart: share only the sections you need, split big files into parts, or ask focused questions about one section at a time.

flowchart LR
    A["Your uploaded files"] --> W["Claude's working space
context window"]
    B["Your messages"] --> W
    C["Claude's replies"] --> W
    W --> D["If it gets too full,
details may be missed or limits hit"]
        

A few minutes of tidying before you upload can improve the answers a lot.

A file is material. You still need to give Claude a job.

A common beginner mistake is to upload a file and send it with no message, or with a vague one such as “Look at this.” Claude may then give a generic summary that is not what you wanted. A good request names the task, the focus, the reader, and the format of the answer.

It also helps to say what the file is. For example, you might explain that the attached document is a draft contract, a customer survey, or a meeting transcript. This context helps Claude interpret the content correctly. If you uploaded more than one file, say how they relate, such as which is the old version and which is the new one.

Pictures are a powerful way to show Claude something that is hard to describe.

Claude can look at photos and screenshots, describe what they show, and read text inside them. This is useful for explaining an error message on a screen, understanding a chart, reading a handwritten note, getting feedback on a design, or identifying what is in a photo. For best results, the help center suggests using reasonably large, clear images rather than tiny or low-quality ones.

Tables are great for questions, summaries, and cleaning, if you set them up well.

Claude can read data files such as CSV and spreadsheets, explain what the columns mean, find patterns, summarize totals, and help clean messy entries. Clear column names and a tidy layout make a big difference. A table with one header row and consistent data is far easier to work with than one with merged cells, hidden tabs, and notes scattered in the margins.

For calculations that matter, it is wise to ask Claude to show how it reached a number and to confirm key figures with a calculation tool or by checking a few values by hand. If a spreadsheet has several sheets, tell Claude which one to focus on. Simple CSV exports are often easier to process than complex workbooks.

Several files can work together, but they need clear roles.

You can often attach multiple files to one message, up to the limit. This is helpful for comparing documents, combining sources, or giving Claude a main document together with a reference sheet. To avoid confusion, tell Claude what each file is and how you want them used. Without that, Claude may mix them up or treat an old draft as the current version.

You can attach a file to one chat, or keep it in a project for repeated use.

There are two main places to put a file. A file attached to a single chat is available in that conversation. A file added to a project’s files area stays available to every chat inside that project. This is helpful for material you use again and again, such as a style guide, a company handbook, or background research, because you do not have to upload it every time.

When a project holds a lot of material, Anthropic describes a retrieval feature that can let a project handle more content by searching for the relevant parts instead of loading everything at once. The details and availability can change, so check the current help center. The general principle is to keep reference material in a project and one-off files in a chat.

flowchart TD
    Q["Will you need this file again?"] -- "No, one-time task" --> C["Attach to the chat"]
    Q -- "Yes, many chats" --> P["Add to a project's files"]
    P --> U["Remember to update it
when the original changes"]
        

Trust, but verify, especially in the first reply.

Mistakes with uploads are usually simple: the wrong file, a missing page, a failed text extraction, or a chart that was not read. You can catch most of them early by asking Claude to confirm what it received. Ask it to state the title of the document, the number of pages or sections it can see, and a one-line description of what each file contains.

For important answers, ask Claude to point to where in the file each key claim comes from, such as by section or page. This makes it easy to verify. If Claude cannot point to a source in your file for a statement, treat that statement with caution, since it may be general knowledge rather than something from your document.

Uploading is convenient, so it deserves a moment of care about what you share.

Files can contain more than you realize: names, contact details, hidden comments, tracked changes, metadata, and old edits. Before uploading, think about who is mentioned in the file and whether you are allowed to share it. For work documents, follow your organization’s rules, and ask your administrator if you are unsure. Remember that your plan, your settings, and your organization’s policies affect how your data is handled, so review the privacy information from Anthropic for your account.

There is also a safety point about the content of files. Text inside a document is information for Claude to read, not an order to obey. A file you receive from someone else might contain odd instructions hidden in it. If Claude reports something unusual, such as instructions inside the file asking it to do something, pause and check with the file’s owner before acting on it.

Seeing familiar situations makes the ideas easy to apply.

Uploading is powerful, but it comes with things to watch.

These habits cause most upload problems, and each has a simple fix.

A short troubleshooting routine fixes most problems.

A few popular beliefs about uploading are only partly true.

Quick answers to the questions beginners ask most.

Use the add button near the message box, drag the file into the chat window, or paste an image from your clipboard. Menu names can change, so check the help center if yours looks different.

Common documents such as PDF, Word, CSV, and text files, plus several others, and image formats such as JPEG, PNG, GIF, and WebP. Spreadsheet uploads may need a related feature enabled. Check the current list in the help center.

For shorter PDFs it can generally read charts and visuals. For many other document types it reads only the text, so export important charts as images and upload them separately.

There are limits on file count, file size, PDF pages, and image dimensions, and they can differ by plan and change over time. The help center article on uploading files has the current numbers.

Long files share Claude’s limited working space with the conversation. Split the file, ask for an outline first, and work through one section at a time.

Use a chat upload for a one-time task, and a project file for reference material you will use across many chats.

Ask it to state the file’s title and length, find a specific detail you already know is in it, and say where in the file each key point comes from.

It depends on your plan, settings, and your organization’s rules. Remove sensitive details, share only what you need, get permission where required, and review Anthropic’s privacy information for your account.

Read the error, check the type and size, split or compress the file, try converting it to a more common format, and as a last resort paste the key text into the message.

Often yes, up to the limit. Tell Claude what each file is and how you want them used so they are not mixed up.

Yes. Claude can misread scans, images, and tables, so verify important facts and numbers against the original file.

Uploading files lets Claude work with your real documents, data, and images instead of guesses. The process is simple: attach the file, say what you want done, and check the first answer. The details matter, though. Claude does not always see a file as you do, since many documents are read as text only, while shorter PDFs and image files can be read visually. Limits on file count, size, and pages apply and change over time, and a large file can crowd Claude’s working space. Preparing your files, giving a clear task, verifying key facts, and protecting sensitive information will make your results better and safer.