Does the opening answer the question in 40–60 words?
Read the opening, the TL;DR on ftes.ai, and check that it answers the piece's primary question directly in 40 to 60 words, beginning with a definition or a verdict sentence rather than with an introduction or a promise of what follows. It should be liftable verbatim into an AI Overview and natural when spoken aloud. Count the words; the verifier does.
Verify: the first sentence could stand alone as the answer, and the count is in range.
- 1. Answer-first, 40–60 words, definition or verdict first
Is every heading a question or a claim, and every section self-contained?
Check that every heading is a question the section answers or a claim it supports, never a bare label. Then read a sample of sections with the rest of the page hidden: each names its entity in the first sentence, answers its heading, contains no cross-reference such as above or below, and runs roughly 120 to 220 words. A section that depends on the page around it is not a chunk a retriever can cite.
Verify: the verifier passes the headings and flags cross-references; you read two sections in isolation.
- 2. Headings are questions or claims 3. Sections stand alone
Are the fan-out sub-queries covered, and is there information gain?
List the five to nine sub-queries the primary question fans into, what, how, cost, risk, versus, when, who, setup, and confirm each is answered in a section or an FAQ on this page or in a linked piece. Then name the one thing the piece adds that the index does not already have: a mechanism, a structural insight, a labelled worked example, a comparison nobody has drawn. If you cannot name it, the piece restates the consensus and should not ship.
Verify: the sub-query list is written down with a location for each; the gain is written in one sentence.
- 4. Fan-out sub-queries each have a home 5. One piece of information gain, named
Is every number sourced, shown, or a product mechanic?
Read every number and every claim of fact. Each must be a product mechanism the platform actually has, arithmetic shown as a labelled example with inputs left to the reader, or a named public source. Any statistic, study, customer, case study, testimonial, or quotation without a source is a fail, and the most serious one, because engines cross-check and an uncorroborated figure marks the page and the domain as unreliable. The fact gate blocks these in drafts from a Digital FTE; a person checks the rest.
Verify: every figure traces to one of the three kinds; the verifier's invented-figure heuristic reports no hits.
- 6. Every number sourced, shown, or a mechanism; zero invented figures
Are the FAQs voice-ready and the H2s video-ready?
Check there are four to six FAQs, that at least two answers are thirty words or fewer and read naturally aloud, and that the FAQ block is visible on the page so the FAQPage schema mirrors something real. Then read the H2s as a list and ask whether they would work as chapter titles for a video on the same subject; if they would, the piece is ready for a video to be planned to it, and if not, the headings need work anyway.
Verify: the verifier counts the FAQs and the short answers; you read the H2 list aloud.
- 7. Four to six FAQs, two under thirty words 8. H2s work as chapters
Are the three related links and the metadata complete?
Check the piece links three related pieces that resolve, at least one in the cluster's hub, so it is a spoke rather than an orphan, and that it links the relevant role or pricing page where genuinely useful. Then check the metadata: title as a plain string, an excerpt of the right length, keywords, tags, ISO dates for published and modified, and, on ftes.ai, the content kind set so the cover and kicker are right.
Verify: the verifier resolves the links and checks the fields; you confirm the kind is the honest one.
- 9. Three related links that resolve 10. Title, excerpt, keywords, tags, dates, kind
How do you use this checklist?
Run it on every piece before approval, in order, and stop at the first fail among items one to six; those are gates, and a piece that fails one does not ship until it passes. On ftes.ai the library verifier runs items one, two, three, seven, nine, and ten mechanically and flags likely fails on six, so a person's time goes to items four, five, and six, which need judgement. A Digital FTE's drafts arrive having passed the fact gate, so item six is usually already clean.
The checklist is the SPEC-035 standard in ten lines. Nothing in it is a stylistic preference; each item is a surface the piece has to win on.