Skip to content
← All help

Courses

Sources approval before writing

A course here teaches from documents a reader can open, so the list of documents comes first. Before any lesson of a new course is written, the person who owns the course approves the list of sources it will cite. If a course is being built for you (by the platform team, a co-author or an assistant), that person is you. If you are building a course for a school, it is the teacher the course is for. Nothing is drafted on a source the owner has not seen.

What you receive

A sources packet: one short document, sent as a PDF and an EPUB, with one entry per source:

  • the full reference (title, author or publisher, date, link);
  • what kind of document it is (a primary record, an official body, a peer-reviewed or edited publication, an archive or dataset, a book, a news report);
  • what the course will use it for (which lessons, which claims);
  • whether a learner will be able to open it (a hosted copy, a link, or a book to obtain);
  • a one-line note on where it came from and the date it was opened;
  • the checks it passed (the eight below), with a note wherever one needs explaining.

Answer each entry with approve, request a change (a better edition, a source you trust more, a source you want left out) or drop. Changes come back to you before writing starts. If a source turns up while the course is being written, it is added to the packet, marked as awaiting your answer, and the course stays in draft until you have given it.

How a source earns its place

Every source in a packet is held to the same eight checks. One that fails is marked unverified: it is replaced, or the claim that needed it is cut.

  1. Kind. A primary document, an official body, a peer-reviewed or edited publication, an archive or a dataset come first. A secondary account is fine when it says where it got the fact. An encyclopedia entry or a listicle is a pointer to follow, not a source.
  2. Author. A named person or institution stands behind it.
  3. Date. The page or record carries a publication, revision or record date.
  4. Stable link. It opened on the date noted, at a link, DOI, handle or shelfmark that can be opened again. A page that could move is saved to the Wayback Machine first.
  5. Editorial process. The publisher's about page or masthead was read: who runs it, whether anyone edits before publishing, how errors are corrected.
  6. Cross-checked. Each fact the course leans on was confirmed in a second, independent source, named in the note.
  7. No content-farm signals. No author, no date, generic stock images, every page the same length, no sources of its own, a domain registered recently: one is a flag, two or more fail the check.
  8. Not an AI. Not a chatbot transcript or a chat link, not an AI-written summary, and not a page whose only source for the fact is an AI. An AI tool is never a source here.

"How do you test whether a source is AI generated?"

Honestly: no automated AI-text detector is used. The largest independent test so far, of 12 public tools and the two systems universities buy (Turnitin and PlagiarismCheck), found the tools "neither accurate nor reliable", biased toward calling text human-written, and worse still on paraphrased or translated text (Weber-Wulff et al., 2023). Vanderbilt University switched Turnitin's detector off on the same evidence: at the vendor's own claimed 1 percent false-positive rate, about 750 of the 75,000 papers it had submitted in a year would have been wrongly flagged, and detectors flag non-native English writers more often (Vanderbilt University, 2023). A detector's score would tell you nothing you could stand behind, in either direction.

What is done instead is the provenance check above, which asks a better question: not "did a machine write this?" but "who stands behind this, and can it be checked?" A page an AI wrote and nobody checked fails the author, editorial and content-farm checks. So does a page a person wrote carelessly, and the course is no worse off for not knowing which.

The tools behind the checks are ordinary ones: the source itself, read in full, never a search snippet; Crossref for a DOI's real record and Unpaywall for a free copy; library catalogs and official registers for books and law; the Wayback Machine for a page's history; a WHOIS lookup for a domain's age; the publisher's own about and corrections pages; and a second independent source for every fact that matters.

Where this sits in the process

Brief approved, then sources approved, then writing, then the content guards, then review and vetting (see Vetting a course, and the "Coming soon" page and Reviewer sign-off). A reviewer can ask for the packet: it is the quickest way to see what a course rests on before reading a lesson.

References

Vanderbilt University. (2023, August 16). Guidance on AI detection and why we're disabling Turnitin's AI detector. Brightspace. https://www.vanderbilt.edu/brightspace/2023/08/16/guidance-on-ai-detection-and-why-were-disabling-turnitins-ai-detector/

Weber-Wulff, D., Anohina-Naumeca, A., Bjelobaba, S., Foltýnek, T., Guerrero-Dib, J., Popoola, O., Šigut, P., & Waddington, L. (2023). Testing of detection tools for AI-generated text. International Journal for Educational Integrity, 19, Article 26. https://doi.org/10.1007/s40979-023-00146-z

Video tutorial scriptnot written yet

Nobody has written the narration for this one yet. It is listed here so the gap is visible rather than forgotten; an invented script would read as approved copy.