Tool · free · no sign-up

Learner

Give it a link to an article and get a prompt that makes an AI produce not a summary but material you can learn from. No API key, no subscription: the prompt works in any chat.

Build the prompt

A normal web page with an article. A paywall or a file to download is something the model cannot read.
Steps 2 and 3

Next — in the same chat

The analysis is only the start. Understanding shows up in the next two steps, and both have to be done by hand rather than read.

Step 2 · after you have answered yourself

Check your answers

The analysis will contain five questions without answers. Answer them yourself — out loud or in writing, without looking back at the article. Only then send this prompt.

Step 3 · the Feynman technique

Test your explanation

Explain the topic in plain words, as if you were telling someone with no background. Send this prompt, then your explanation. The AI will look for holes in it instead of praising it.

What is inside

Why not a summary

Asking an AI for a recap is easy and pleasant, and it adds nothing to your understanding. A clear explanation creates the feeling of understanding — and the difference shows up on the first real task. So the prompt is built to make you work rather than read.

A map of the text

Structure, key concepts and the links between them — before you read. You can see which three or four sections you actually need and which you can skip. The prompt also demands that the model name the author and their interest: a vendor text reads differently.

Inspectional reading — Mortimer Adler, How to Read a Book

Claims with what they rest on

Every claim the author makes is written out together with what holds it up: data, logic, personal experience or nothing. That one line exposes more than the rest of the analysis combined.

Critical reading

Five questions without answers

Re-reading barely works — trying to recall does. So the answers are not handed over straight away: you go back for them in a separate step, after answering yourself.

Practice testing — one of the two techniques rated highest in the review by Dunlosky and colleagues, 2013

Ten cards

The same half hour spread across a week beats an hour and a half in one sitting. The cards are short — a long answer on the back is not something you remember.

Distributed practice — the second highest-rated technique in the same review

Who disagrees

Every methodology has critics, and it is usually the critics who have written down where it breaks. The prompt demands strong objections rather than straw men, and forbids inventing opponents who do not exist.

Syntopical reading — Adler again

A task for practice

One concrete 30–60 minute task with a completion criterion and a prediction of what will turn out to be harder than the article makes it look. The gap between "that's clear" and "I've got it" only opens up here.

The practice check
Honest about the limits

What this tool does not do

It does not read the article for you

The article is opened by the model in your chat, not by this site. All that happens here is the text of the prompt being assembled — nothing is sent anywhere and your link is not stored.

It does not fully stop invention

The prompt has a dedicated "facts only" block and a requirement to mark anything the model adds itself. That lowers the risk without removing the need to check: verify numbers and source titles against the original.

It does not replace practice

No analysis produces understanding without an attempt to apply it. The last block of the prompt exists for exactly that, and skipping it means you have spent your time on pleasant reading.

FAQ

Frequent questions

Do I need a subscription or my own API key?
No. This site only assembles the text of the prompt; it runs in whichever chat you already use — ChatGPT, Claude, Gemini, Grok, Perplexity, DeepSeek, Alice or GigaChat. No sign-up, and the tool is free.
How is this different from "summarise this article"?
A summary creates the feeling of understanding rather than understanding itself — the gap shows up on the first real task. Instead of a retelling, the prompt demands a map of the text, claims with what they rest on, five questions with no answers, ten cards, the strongest objections and a practice task. You do the work instead of reading.
Does the link get sent to your server?
No. The prompt is assembled in your browser: the link is never sent anywhere and never stored. The article is opened by the model inside your chat, not by this site.
The chat says it cannot open the link. What now?
You need a model that can reach the web. Ask it to turn on web search, or paste the article text as your next message. Paywalled articles and downloadable documents are out of reach for the model.
Why are the answers a separate step?
Re-reading barely works; trying to recall does. So the five questions come with no answers: you answer them yourself first, without looking back at the article, and only then send the step 2 prompt to check. Practice testing is one of the two techniques rated highest in the Dunlosky et al. review, 2013.
Can the AI still make things up?
It can. The prompt has a dedicated "facts only" block and a requirement to mark anything the model adds itself. That lowers the risk without removing the need to check: verify numbers and source titles against the original.