How to Use Merlin AI with Files and Long Docs

November 17, 2026

Merlin AI reads the documents you upload — PDFs, Word files, plain text — inside its browser sidebar, then answers questions with the document as context. On the free plan file uploads are capped (a handful per day), Pro unlocks them properly. It is the fastest path to querying a long PDF without building a RAG stack.

Most people stop at “summarise this page.” The file mode is where the tool earns its subscription, and also where the limits bite.

How do you attach a document to the sidebar?

Click the paperclip in the chat box, pick the file from your machine, wait for the parse, then ask. Merlin shows a small chip with the filename so you know what is in context. You can stack several files in one conversation, though not all model tiers handle multi-file context well.

If the file is a scanned PDF — images of text rather than real text — expect garbage. OCR is not its job. Convert first, or use a dedicated OCR tool.

What are the practical size limits?

From real use, a 40-60 page text PDF works. A 300-page annual report usually trips the context window or the upload size cap, and you get a truncated answer that sounds confident. Split the document, or ask section by section.

The Pro tier raises both the file count and the context length, which is the one upgrade reason that is actually justified. Free plan users get a small number of file uploads per day — enough to test, not enough to work.

How do you get usable answers out of a long document?

  • Ask for structure first. “List the headings of this document” tells you what the model actually received.
  • Quote, don’t paraphrase. “Give me the exact sentence about the termination clause.” You can spot hallucination immediately.
  • One question per turn. Multi-part questions on long docs produce answers that skip the second part.
  • Turn on a stronger model for legal or financial reading. The cheap models skip qualifiers.

That last point matters more than it sounds. A summarisation miss on a marketing blog is annoying; a summarisation miss on a lease clause is a problem.

Merlin AI for files vs NotebookLM and ChatGPT Projects

NotebookLM is free and grounding-focused, with citations back to the source — better if you need to verify every claim. ChatGPT Projects keep files across sessions more neatly. Merlin wins on one axis only: it lives in your browser next to the page you are reading, so you can cross-check a document against a live website in one view.

If citation fidelity is the priority, NotebookLM is the safer pick. If speed of access is, Merlin is faster.

Where the file workflow falls down

Uploads are sent to Merlin’s servers and then to the model provider, so client documents with personal data need a policy check first. There is also no local mode. And the sidebar does not remember files across new conversations on the free tier, so you re-upload more than you would like.

For teams that need a document store with retention rules, this is not it. For a solo consultant who needs to interrogate a contract before a call, it does the job in four minutes.

Quick Summary

  • Merlin AI’s paperclip upload handles PDF, DOCX and text files in the sidebar.
  • A 40-60 page text PDF is comfortable; 300-page reports get truncated.
  • Pro is worth it mainly for file uploads and longer context.
  • Ask for exact quotes to catch hallucination on long documents.
  • NotebookLM is better for cited answers; Merlin is better for speed next to a live page.

Full pricing tiers and supported file formats sit on the Merlin AI tool page.