The question was small enough to survive contact with reality: could one public book record become an original landing page with traceable facts and a working reading link? No content-intelligence layer. No dashboard congratulating itself for having an address.

One URL was not the whole brief

The input was the Project Gutenberg record for Pride and Prejudice by Jane Austen, English-language eBook 1342. The brief explicitly requested an original English page, that record as the sole factual source, and a direct path to legitimate reading or download destinations. Additional instructions ruled out copied assets, private data, accounts, paid services, scripts and invented claims. This was a constrained authoring case, not a blind trial or a test of a commercial website builder.

One URL, plus instructions, source checking and an actual test. The marketing department has asked us to delete everything after the comma.

The result was one standalone HTML page with its own text and styling. Its main link opens the online reader; EPUB3 and plain text are secondary choices. The first generated file was preserved unchanged. No corrected second version was needed for the checks we ran. The download below contains that original page and notes describing the experiment.

What actually passed

  1. The recorded title, author, language and eBook identifier matched the source page.
  2. The main reading link and both download addresses matched the recorded destinations; HTTP requests returned successful responses of the expected types.
  3. Inspection found no scripts, forms, payment flow, analytics, invented testimonials or private contact details in the generated page.
  4. The original HTML fingerprint was preserved, so the first attempt was not silently replaced with a tidier story.

These checks support a narrow result: a small page with source-linked metadata and reachable destinations. They do not show that the page persuades anyone, works beautifully on every screen or makes a business. A successful network response has been politely declined as evidence of product-market fit.

There was already AI in the source

The source page labels its plot synopsis automatically generated. The landing page paraphrases that synopsis; we did not independently check the plot description against the novel. Matching a source is not the same as proving the source correct. Apparently the fact-checking department also needs a fact-checking department.

A checklist worth stealing

  1. Record the input page, the retrieval time and the exact instructions before judging the output.
  2. Keep the first attempt. Put any corrections in another file and explain them.
  3. Separate factual claims from your own interpretation, and check where each fact came from.
  4. Give the primary action a real destination. Engagement is not an address.
  5. Check the destinations as well as the button labels; a handsome dead link is still a dead link.
  6. Write down the untested part before announcing the miracle.

Browser layout, small-screen behavior, accessibility in use and reader comprehension were not tested. There is no audience feedback or demand data. The next useful question is whether a second, genuinely different source exposes a failure these checks missed. Adding six more pages about how impressive the first page was would be considerably easier.

The page exists. The links respond. The empire remains a staffing issue.

Sources