How-To: Connect AI Tools to Your Own Documents

Step 1. — Claude Projects with files

Claude (Pro and Team plans) lets you create Projects and add files directly to the project's knowledge base.

1. •Create a project and give it a clear name (e.g. "Annual Report Analysis").
2. •Upload PDFs, Word docs, code files, or plain text — up to your plan's context limit.
3. •Set project instructions telling Claude how to use them: "Cite the page number for every claim; flag anything not in the sources."
4. •Ask questions in plain language within the project. Claude reads the uploaded files and answers grounded in their content.

Claude's large context window makes this especially good for long documents — hundreds of pages of contracts, research papers, or technical specs.

Step 2. — ChatGPT file uploads

Both free and paid ChatGPT users can upload documents directly into a chat. Click the paperclip / "+" icon, select a file, and ask a question. The AI reads the content and answers within that conversation.

• •Works well for: one-off questions about a single document, quick summaries, formatting conversion.
• •Limitations: file-size caps, and the upload is session-only — a new chat forgets the file. For repeat work, use a Project instead (see the ChatGPT Projects how-to).

Step 3. — Google NotebookLM

NotebookLM is a free Google tool designed specifically for grounded research. You upload up to 50 sources, and it becomes a personal AI research assistant that answers only from what you've added.

1. •Go to NotebookLM and create a new notebook.
2. •Add sources — PDFs, Google Docs, websites, YouTube URLs, or pasted text (up to 50 per notebook).
3. •Ask questions. Every answer includes inline citations you can click to jump straight to the source passage.
4. •Use the built-in summaries and "study guide" generation for fast overviews of an unfamiliar corpus.

For a deeper look at NotebookLM, see our companion post on using NotebookLM for research. It's free, it cites sources, and for many research tasks it beats ChatGPT outright.

Step 4. — Perplexity Collections

Perplexity (the AI search tool — see the Perplexity how-to) lets you save sources into Collections. Create a collection for a research topic, add URLs and uploaded files, then ask questions that are answered using only that collection's sources. It's ideal when you want AI answers scoped to a vetted reading list, not the open web.

Step 5. — A practical research workflow

A workflow that combines the strengths of all these tools, for a literature review or report:

1. •Collect the source papers and upload them to a NotebookLM notebook.
2. •Ask grounded questions — "What do these papers agree on about X?" — and copy the cited answers.
3. •Paste the answers and notes into a Claude project set up as a writing assistant.
4. •Have Claude draft the summary, citing page numbers from the originals.
5. •Manually verify the citations in the source PDFs before publishing.

Step 6. — Privacy considerations

When you upload a document to a cloud AI, that service can read it. For most everyday research this is fine, but for anything sensitive — client data, contracts under NDA, personal health or financial information — think before you upload.

• •Check the tool's data-retention and training policy. Some vendors let you opt out of training on your content.
• •For confidential material, consider running a local model. Tools like Ollama let you run capable models on your own machine; your documents never leave your laptop.
• •As a rule: nothing you wouldn't forward to a stranger should go into a public AI tool.