Skip to main content
A separate pass from the pre-publish checklist, because it tests something different: not whether the article is good, but whether a model can find one part of it, trust it, and quote it. The reasoning behind each group is in SEO and GEO concepts.

​
Can a passage stand alone?

An answer engine retrieves chunks, not pages. Each section has to survive being pulled out of context.
  • Every section answers its heading in the first sentence, before context or setup.
  • No section depends on having read the section above it to make sense.
  • Pronouns near the start of a section refer to something named in that section, not in the previous one.
  • Headings are specific enough to identify the section’s subject without the page around it.
  • The FAQ section exists, and its questions are ones people actually ask.

​
Is there anything worth quoting?

A model assembling an answer needs a fact, a number, or a phrasing it can lift. In controlled testing, adding quotations, statistics, and cited sources produced the largest measured gains in whether content was used.
  • At least one statistic that is specific, sourced, and dated.
  • At least one quotation from someone with standing on the subject.
  • Every claim that could be challenged has a citation next to it.
  • Original data, if we have it, is stated as a plain number in the body text and not only shown in a chart.
  • Definitions of key terms are written as definitions, in a sentence that would work quoted alone.
  • The article contains something a model could not generate without it.

​
Is the entity clear?

  • The company name is written exactly as it appears on every other profile we control.
  • The product name is consistent with the rest of the site, with no informal variants.
  • The author is a real named person, with a bio and a link, matching how they are named elsewhere.
  • Key terms the article assumes are defined in the article.
  • Organization schema with sameAs links is in place site-wide.
  • Article schema is applied, with the author and dates populated.

​
Is it fresh, and does that show?

  • The last-updated date is visible on the page, not only in the CMS.
  • The date reflects a substantive change, not a bump.
  • Every statistic carries the year it refers to.
  • Nothing in the article contradicts something we published more recently.
  • The next review date is set in the register.

​
Is it structured for extraction?

  • Roughly a quarter to a third of the content is lists or tables, where the content is genuinely parallel.
  • Tables have header rows and are not used for layout.
  • Heading depth is three to five levels overall, without skipping.
  • Emphasis is used on key terms rather than scattered.
  • Numbered lists are used for sequences, bullets for unordered sets.

​
Can it be crawled at all?

  • The page is not blocked in robots.txt.
  • AI crawlers are not blocked unless that is a deliberate decision the team has made: GPTBot, ClaudeBot, PerplexityBot, Google-Extended.
  • The content is in the HTML, not rendered only by client-side JavaScript.
  • The canonical URL points where it should.
Blocking AI crawlers is a legitimate position for some businesses, and it is a decision to make deliberately rather than discover. A site that blocks them is choosing not to be cited by those engines, which is a coherent choice if the traffic tradeoff works out and an expensive accident if nobody meant to make it.

​
Off the page

These are not article-level checks, but they decide how much weight the article carries. Review them per cluster rather than per article.
  • The company appears with consistent naming across the profiles we control.
  • The author has a presence beyond this article.
  • Something in the article, usually original data, gives a third party a reason to reference it.
  • The distribution actions from the brief have owners and dates.
Last modified on August 6, 2026