The Differentiation Standard

E-ITA.org recovery audit · Phase 2 companion document · issued 2026-09-26

This is the written specification every rebuilt country page must meet before it ships. It turns the Phase 2 requirements in Part 6 of the Recovery Plan into section-by-section rules, thresholds, and a pre-publish checklist. The two model pages, /italy/ in English and /es/estados-unidos/ in Spanish, apply it end to end. Clone their shape; don't invent a new one.

1. Why a standard, and why this one

Driving documents are a YMYL topic by the rater guidelines' own example list, which means raters hold these pages to the highest scrutiny.

Today 184 of 195 English country pages have a sibling above 0.90 similarity, and translated country pages run 0.97 to 0.99 median (Part 3.2). A rater who opens several of them sees one page with the country name swapped. That is the pattern the guidelines tell raters to mark Lowest:

The target is low effort and low originality at scale, not AI use as such. The guidelines say so directly (SQRG 4.6.6, p.43: “the use of Generative AI tools alone does not determine the level of effort or Page Quality rating”). So the standard tests what a page carries that a template can't produce: facts no sibling page has, primary sources, named people, and dated verification.

2. What the leading pages carry

Measured on the live pages on 2026-09-14. These are benchmarks for the bar to clear, not sources to cite (the Part 6 blacklist bans citing rival vendors on rebuilt pages).

These measurements set the thresholds below. Word count alone doesn't differentiate: one leader runs an identical 1,278-word page for Japan and Spain. So the unique-fact floor and the overlap ceiling are the primary tests, and word count is secondary. None of the three names an author, a reviewer, or a visible date on a country page. That leaves named authorship with dated verification open for E-ITA to take.

3. Page anatomy: 12 sections, in this order

Every country page, in every locale, carries these sections in this order. The acceptance test in the right column is a publish blocker.

The order button sits after section 9 and never above it. It offers the booklet as a translation aid, with no claim of validity or acceptance beyond what the page states.

4. Thresholds, per page

These restate the Part 6 per-page requirements with the reason behind each.

5. Sourcing rules

1. Hierarchy. Convention texts and the treaty depositary outrank government pages, which outrank automobile clubs, which outrank rental-company terms, which outrank editorial sources. When sources conflict, the page states the conflict and the more authoritative source wins.
2. Every number gets a source. Speed limits, fines, validity periods, fees: each is backed by a primary source listed in the Sources section with an accessed date. No exceptions. The sources sit at the end of the page so the body reads clean.
3. Acceptance claims trace to convention status or an issuer of record. Never to marketing copy. No primary source supports “175+ countries”; the official US figure is “more than 150 countries” (7 FAM 1436 a), and it applies to convention IDPs, not to a private booklet.
4. Contradictions get resolved on the page, not hidden. Japan cannot be accepted on the homepage and excluded in the FAQ. A multi-year product term cannot sit next to official-permit language as if it were permit validity.
5. US buyers get the FTC frame plainly. AAA and AATA are the only issuers the US Department of State authorizes. The page either says so or states exactly what the product is and is not. There's no middle version that depends on which page a visitor landed on.
6. Check for exceptions before stating a requirement. A country's law can defer to international agreements, so a rule that reads universal may not apply to every visitor. For each main visitor nationality, check that nationality's own government travel guidance and state any exception on the page. Example: Italy's art. 135 asks non-EU drivers for a permit or translation, but the UK government says UK photocard licenses need no IDP for Italy.
7. Translations carry the same citation set as the English master and are reviewed by a named fluent speaker.

Coverage rule

Every country page draws from at least these 6 source types. The approved domains for each claim type are the Part 6 whitelist; the banned source types are the Part 6 blacklist (rival vendors, lookalike operators, Wikipedia and Wikisource, blogs, forums, stale aggregators, sites that sell coverage).

Re-check every URL live before citing it. A source that won't load is not cited from memory.

6. Authorship and review

1. Every country page has a named human author and a named human reviewer. Each has a bio page stating relevant credentials: travel law, automotive-club work, rental operations, or documented travel in the country.
2. The reviewer signs off with a visible review date. Both names appear in Article structured data.
3. Names come from E-ITA's own team or named contractors. A page without real names doesn't ship; the byline slot stays empty until it's filled.
4. One change-log entry per substantive update. Bulk date touch-ups with no content change are not allowed.

E-ITA's registered business category is translation services, yet none of the 454 audited URLs names an author (Executive Summary, ranked cause 3).

7. Per-page automation disclosure

Where automation contributed to a page, the page carries its own short disclosure: who was involved, how automation was used (drafting, translation, data assembly), what human review followed, and why. A single site-wide sentence doesn't meet this, because it can't say which pages it covers:

8. Translation and locale rules

The locale findings behind these rules are in Part 3.3.

1. Every locale page is reviewed by a named fluent speaker whose credit appears on the page.
2. One register per page. Spanish pages use formal “usted” throughout and never mix it with “tú”.
3. One product term per locale, held consistently. In Spanish, “Permiso Internacional de Conducir” is the official permit and “Licencia Internacional de Conducir de E-ITA” is the product.
4. No English title case in non-English headings.
5. A localized title tag, meta description, and H1 on every locale page, with the brand suffix once.
6. The same facts and citation set as the English master. Contradictions get fixed at template level so they can't propagate across locales.
7. Complete, reciprocal hreflang at every level, re-crawled after each fix.
8. Localized language-switcher chrome, including the “Login” label.
9. Order of work: the locales that still carry traffic first (es, fr, it, ar), then de, then a fix-or-retire decision for the rest.

9. Update cadence

• Sitemap lastmod equals the date of the last substantive change, nothing else.
• Priority-market pages are re-verified monthly; every other country page at least quarterly.
• Any change in a cited source (law, fee, convention status, rental policy) reaches the page within 30 days, with a change-log entry.

10. Pre-publish checklist

A page ships only when every row passes. The rating column is an analyst read of how the guidelines classify a page that fails the row, not a Google score. Rater data isn't used directly in ranking, but it describes what the ranking systems are built to reward and demote.

11. How to measure a page: the audit's own method

Rewritten pages are measured with the same instruments the audit used, so every result compares like for like with the Part 3 baselines. Run all three measurements on the main content only: strip navigation, header, footer, scripts, the country select lists, and any sitewide blocks before counting.

11.1 Similarity: three numbers per page

Classes are the verdict sheet's page classes: English country pages are measured against English country pages, and translated country pages against the translated pages of their own locale. Re-run the measurement for the whole class each time a batch ships, because every new page changes its siblings' scores too.

11.2 Quality: the 7-criterion rubric

The audit scored pages 1 to 10 on seven criteria and recorded the evidence behind every score. Score rewritten pages with the same rubric and the same discipline: one score per criterion, one or two sentences of evidence from the page, harsh by default. These are rubric judgments, comparable within this program. They are not Google scores and not rater scores.

12. Score targets

Every rewritten page reaches column (a) on every criterion before it ships. Column (a) beats every scored reseller competitor; column (b) approaches the official issuer's page, which scored 8, 4, 6, 7, 9, 8, and 7 on the seven criteria in order. Aim for (b) on the priority markets.

Why these bars: in this niche, identity and intent move scores more than volume. A competitor's 195-word legal page scored higher overall than the market leader's 3,280-word Japan guide (Part 3.5). Citation is the cheapest criterion to win, because the whole niche cites almost nothing.

13. Writing brief: hand this to a writer or an LLM

This section is self-contained. Copy it, fill the inputs, and give it to the person or model producing the page, together with sections 3 to 6 and 8 of this standard and the Part 6 whitelist and blacklist.

When the page comes back, the reviewer checks every ledger row against its source, the measurements in sections 11 and 12 run, and only then does the page ship.

14. The model pages

Two pages apply this standard in full. Each shows all 12 sections with the copy drafted, every fact sourced with an accessed date, and a visible slot for the author and reviewer names E-ITA supplies before publish.

• /italy/: the English model and the largest single page to recover.
• /es/estados-unidos/: the Spanish model. It carries the translation rules in section 8.