How we write

Every article on this blog is written with AI assistance and reviewed before it is published. That is stated on each article, not buried here, because a reader who works it out for themselves has been given a reason to doubt everything else on the page.

Here is exactly what happens.

Where the facts come from

The pipeline reads RSS feeds from mining and crypto publications — Hashrate Index, Bitcoin Magazine, CoinDesk, Cointelegraph, Decrypt and Bitcoin.com News. It does not scrape search results and it does not republish anyone's text.

Each article names and links the outlet that did the reporting. If the sources are gone from the bottom of a page, that page is broken.

What the model is and is not allowed to do

Two rules are enforced in code, not just asked for in a prompt:

Nothing invented. The writer is given the source text and told to use only what is in it. A separate pass compares the finished article against that source and fails it if any figure, date, company or causal claim appears that the source does not support. A failed check does not publish — it becomes a draft a person has to open.

No pretended experience. We do not write "we tested", "we measured" or "in our lab". JouleBook has not tested the hardware in these articles and does not claim to have. An automated check rejects the article if that language appears at all.

That second rule is worth dwelling on, because the industry norm is the opposite. It is common for AI content pipelines to instruct the model to write as though the publisher broke the story themselves, and to avoid naming the outlet the facts came from. That produces a claim the business cannot stand behind, and it is exactly what Google's guidance on experience and trust is built to catch.

What we are not optimising for

We are not trying to make these articles undetectable as AI-written. That target is both dishonest and, in our reading of the evidence, wrong on its own terms: Google does not penalise content for being AI-written, it penalises content that is published at scale and adds little. The tricks people use to fool detectors — injecting casual filler, varying sentence length mechanically — reliably become their own tell within weeks, and they do nothing about the actual problem.

So the checks measure things a reader would notice. Stock phrasing. Paragraphs that run too long. Sentences that are all the same length. Whether the piece says anything the source did not already say better.

Volume

The pipeline is rate-limited on purpose, with a ceiling on articles per day. A publication that posts six thin pieces a day is a worse publication than one that posts three useful ones, regardless of who or what wrote them.

Corrections

If something here is wrong, write to support@joulebook.com with the article and the claim. We will correct it and say that we did.