The twelve checks, and what each is worth.

What does each check measure, and what is it worth?

Weights are read from the same module the published tool runs, so this table cannot drift from the scorer. The last three columns are measured across the 91 content pages on this site.

Tables shows no fail count because its absence is recorded as an unscored note: plenty of good pages have nothing worth tabulating, and a check that concedes as much should not deduct points.

How should you use these pages?

Run the checker on a page you care about, then read the page for each check it flagged. Work down by weight rather than by how easy each fix looks: unsourced numeric claims is worth 18 points and is usually satisfied by attaching a source to a figure that is already there.

• Unsourced numeric claims 18 pts

  Why a figure with no source beside it is the hardest thing for an AI answer engine to repeat, how the check decides which sentences count, and how to fix them.

• Direct answer near the top 14 pts

  Why the first paragraph decides whether a page can be quoted, what length works as a standalone answer, and how to restructure an opening that buries the point.

• External citations 12 pts

  How many outbound sources a page needs to read as evidenced rather than asserted, what the check counts as attribution, and why linking out does not cost you rankings.

• Named author 8 pts

  Why a byline is one of the cheapest trust signals to add, what patterns the check recognises, and when attributing to a team is more honest than naming a person.

• Publication or update date 8 pts

  Why an undated page is easy to skip, what counts as a full date, and why rolling the date forward on every deploy is worse than having no date at all.

• Definitions 8 pts

  Why a page that defines its terms is easier to quote, what phrasing the check recognises, and how to define without padding the page.

• Heading structure 8 pts

  Why question-shaped headings make each section independently retrievable, how many the check expects, and why "Our Approach" is the worst heading you can write.

• Answer-shaped paragraphs 6 pts

  Why paragraphs over 400 characters get summarised instead of quoted, what proportion the check expects, and how to break up prose without making it choppy.

• Sourced figures 6 pts

  Why three sourced figures is the bar, why the check counts sourced numbers rather than measuring density, and why the absence of figures is never a failure.

• Methodology section 5 pts

  When a page owes the reader an account of how it knows what it claims, what wording the check recognises as a methodology heading, and why body-text mentions score less.

• Entity clarity 4 pts

  Why naming your subject consistently matters more than avoiding repetition, how the check identifies the page subject, and what pronouns cost you in retrieval.

• Tables 3 pts

  Why a structured comparison is unusually quotable, and why the absence of a table is recorded as an unscored note rather than counted against the page.

How did we measure this?

Every figure on this page comes from running the Gadex scorer over the built HTML of this site, reading each page's main content region only — nav, header and footer are excluded, because a visitor pastes an article rather than a whole document. Legal boilerplate, the 404 page and generated tool result renderings are left out as pages where a score would be meaningless. Per that run, 91 content pages are counted. The numbers regenerate on every deploy and the build fails if they stop matching, so nothing here can quietly go stale.

The reasoning behind why these properties matter draws on the platforms' own guidance, not on Gadex's opinion alone:

• Google Search Central: creating helpful, reliable, people-first content
• Google Search Central: Article structured data
• OpenAI: bots and crawlers

What none of this can tell you is whether an answer engine will cite a page. Citation depends on the model, the prompt, the index, the competing sources and your domain's standing — none of which is visible in markup.