Which tool to choose
They are not equivalent: each is built around a different priority.
| Tool | Underlying logic | When it's best | Free cap |
|---|---|---|---|
| Claude | Reads the document like a human eye: text and figures together | Manuals with diagrams, tables, charts | Up to 20 files per chat, 30 MB and ~100 pages each |
| Gemini | Bets on capacity: swallows huge documents | Very long booklets or ones already on Google Drive | ~32,000 tokens (about 50 pages); up to 10 files at a time |
| ChatGPT | The tool you already have open, handy for low volume | A single short manual | 25 MB per file, 3 files a day |
Claude also reads the figures and diagrams, not just the words, and this makes it the only one suited when the meaning is inside a diagram. Keep in mind that the file-upload limit applies on the free version too: your PDFs it reads anyway.
Paid Gemini reaches 1 million tokens, about 1,500 pages, so a 200-page booklet only the paid version digests; the free one stops much earlier.
ChatGPT handles the single short manual well, but with multiple documents the three files a day get tight fast.
The practical rule: document with images → Claude. Giant document → Gemini. Short document already at hand → ChatGPT.
How to do it
From a browser or from the app the path doesn't change:
- Open the chosen tool and start a new conversation.
- Upload the document with the attachments button (the paperclip icon or the "+" next to the writing bar). If it's short text, paste it directly into the message.
- Right below the file, write the prompt (the request) you find below.
- Read the answer and, if a point stays unclear, reply in the same chat: "Step 4 isn't clear to me, explain it as if to an absolute beginner."
The prompt to paste as is:
You are a patient technician explaining things to someone who has never
used this tool. Rewrite the attached instructions, turning them into a
simple guide. Rules:
- use numbered steps, one for each concrete action
- explain every technical term in parentheses, with common words
- remove useless cross-references ("see paragraph 3.2") and put the
information right there for me
- at the end, list the most common mistakes and how to avoid them
- do not add anything that isn't in the document; if a piece of data is
missing, write "the manual does not specify it"
The last line is the most important defense: it stops the AI from filling the gaps with invented information.
Check: a good answer has numbered steps, zero cross-references to paragraphs, and every acronym explained. If you still see phrases like "consult the previous section," the prompt wasn't followed: repeat the request adding "do not refer to other sections, write everything here."
A concrete example
Marta bought a boiler and the booklet reads: "Set the heating curve by adjusting parameter 1.4 according to the external compensation factor." She doesn't understand any of it. She photographs the page with her phone, opens Claude, uploads the image, and pastes the prompt above, adding: "Just explain to me how to adjust the heating to make the house warmer."
Claude returns: "The heating curve decides how much the boiler heats based on how cold it is outside. To get more heat: 1) press the menu button, 2) look for parameter 1.4, 3) increase the number by one point, 4) wait an hour and feel whether it's enough." On top of that, it warns her that raising it too much consumes more. Marta does it, the house warms up, problem solved without calling the technician.
When it does NOT work (and how to fix it)
If the PDF is a scan and the AI says it can't read it
It happens with old photocopied manuals. A scanned PDF is just an image: it has no text layer, so the AI finds nothing to read, regardless of the file size. Two ways out:
- Photograph the pages with your phone and upload them as images: Claude and Gemini "see" them.
- Turn the scan into text with OCR (optical character recognition). Upload the PDF to Google Drive and open it with Google Docs: it reconverts it into selectable text, which you then paste into the chat.
If the manual is too long and the AI "forgets" the beginning
Every tool has a context window, that is, the amount of text it can keep in mind at once. On free Gemini you exceed it beyond ~50 pages, on Claude you saturate it with very heavy documents. When you go over, the answers start to ignore pieces or lose the connections.
Fix: split the PDF by chapters with a splitter (Smallpdf online, or Preview on Mac or Adobe Acrobat) and work on one chapter per chat, uploading only the part you need each time.
If the file is too heavy and the upload fails
Above 32 MB or 100 pages, Claude refuses the upload with an error. Large PDFs are almost always bloated by embedded images: run them through a PDF compressor, which reduces the size while leaving the text's readability intact. If even so it stays over the threshold, split it by chapters as above.
If you've used up the day's free uploads
On ChatGPT's free plan the three files run out fast. No need to pay: for documents made mostly of text, select the sections you need from the PDF and paste them directly into the message. The file upload isn't used at all, so the daily cap doesn't affect you.
A tip from someone who really uses it
After the AI has simplified the instructions, ask a second question the original manual never answers: "What happens if I get this step wrong? Is it dangerous or reversible?" That's where you recover the confidence the printed sheet doesn't give you. And when you perform a physical action (adjusting an appliance, assembling a part), keep the chat open and describe what you see: the AI corrects you as you go, something a printed booklet can't do.
Frequently asked questions
Can I use a photo of the instructions instead of the PDF?
Yes, and it's often the fastest route for paper manuals. Claude and Gemini read images: you snap the page, upload it, and ask for the simplification. For multiple pages, take multiple photos and upload them together. On free Gemini you upload up to 10 at a time.
Does the AI also translate a manual in a foreign language while simplifying it?
Yes, the two things are done in one go. Add to the prompt: "Translate into Italian and simultaneously simplify for a beginner." It works on all three tools and returns the numbered steps already in your language.
If the simplification contains an error, will I notice?
This is the real risk to keep in mind, more than the page limit or the file size. The AI can rephrase clearly but wrongly: a value inverted, a step skipped. The prompt line that forbids inventing ("if a piece of data is missing, write that the manual does not specify it") greatly reduces the problem, but doesn't eliminate it. For operations that affect safety, warranty, or money, always compare the critical numbers (pressures, voltages, codes, sequences) with the original page. The simplification makes you understand what to do; the manual remains the source for which exact value to use.