What We Are, and What We Aren’t

Last updated: June 2026

Most companies that use AI would rather you not look too closely at how it works. We’d rather you did. Learning Whistle is built on AI, and AI is genuinely useful and genuinely fallible — often in the same paragraph. This page is the plain-language version of what that means for you, with no marketing gloss. If you only read one page on this site before deciding whether to trust it, read this one.

What the AI is doing here

When you pick a topic, an AI model writes you a short, structured learning path — a handful of stations, each a few minutes to read, calibrated to the reading level you chose. It’s good at explaining things clearly, adjusting how complex the language is, and organising a subject into a sensible order. Think of it as a knowledgeable, tireless writer who can pitch the same idea to a 9-year-old or a graduate student on request.

On Premium paths, it does more: before writing, our pipeline goes and finds real, open-access sources — peer-reviewed research, government data, public-domain texts — and the writer is required to tie its specific factual claims back to those sources, with citations you can click and check.

What it isn’t

It isn’t a person, and it isn’t a search engine. It doesn’t “know” things the way you do — it predicts plausible text. That makes it fluent and confident even when it’s wrong, which is exactly the trait that makes AI worth being careful about. A clear, well-written explanation is not the same as a correct one, and we never want the polish of the writing to be mistaken for proof that it’s right.

It also isn’t a textbook or a citable academic reference. Standard pathscarry no citations and are written from the model’s own knowledge — a great place to start, not a source to quote in an essay. For anything you intend to cite, use a Premium path, and still check the original source.

What we verify

For Premium paths, we don’t just take the AI’s word for it:

• A separate AI pass — one that never saw how the station was written — checks each citation against the source it points to, and flags claims a source doesn’t actually support.
• Citations that point nowhere real, and quotations that don’t match their source word-for-word, are treated as zero-tolerance defects: they’re corrected or removed before the path reaches you.
• When a path is made public, we run an extra fact audit against Wikidata — an independent, openly-maintained knowledge base — as a second opinion.
• Every Premium path is scored against a quality gate before it’s delivered. If it fails for a reason we can’t fix, you aren’t charged — your Gold Tickets are refunded in full.

What we can’t

We can’t promise zero errors. Verification reduces mistakes; it doesn’t eliminate them. A source can be cited correctly and still be outdated, or a nuance can be lost in summarising. We can check that a claim is supported by a source — we can’t guarantee the source itself is the final word on a moving field.

We also can’t cover every topic equally. Open-access material is rich in the sciences and thinner in some humanities and pre-modern subjects. Where the public record is sparse, a Premium path will say so plainly rather than pretend otherwise — and if there genuinely isn’t enough to source a path, we stop before writing it and refund you.

What we won’t teach

We say “any topic,” and we mean it more literally than most platforms do — but not without limit. Every topic you enter passes a safety filter before anything is generated. Sexually explicit material is blocked, for everyone, always. There is no adult tier, no unlock, and no setting that turns this off.

What that does notmean is squeamishness about the human body. Human reproduction, anatomy, puberty, STI prevention, the endocrine system — these are legitimate subjects and they are deliberately allowed. The line we draw is about explicit or gratification-oriented framing, not about biology, medicine, or health. If a clinical topic is ever refused, that’s a bug worth reporting, not the policy working.

The filter errs toward caution. If it can’t confidently judge a topic, it declines rather than guesses — so you may occasionally be turned down for something reasonable. We think that’s the right trade for a platform built to be safe for a thirteen-year-old and a graduate student on the same day.

You may notice a Safety Filtersetting in your profile. To be straight with you: it currently has no effect on what you can generate — the limits above apply to everyone either way, and no setting unlocks a blocked subject. We’re either giving that control a real job or removing it, rather than leaving a switch that implies something it doesn’t do.

Where you bring your own judgment

Use Learning Whistle the way you’d use a sharp, fast, well-read study partner: as a starting point and a way to get oriented quickly — not as the last word. Every station is labelled as AI-generated, and Premium citations are there precisely so you can follow them to the original and decide for yourself. For anything that matters — a medical decision, a legal question, an academic submission — verify against the primary source and a qualified human.

If you find something wrong, we want to know. Report it on our errata pageand we’ll look into it. And if you want the full mechanics of how Premium paths are built and checked, that’s on our methodology page.