LEARN

A Checklist You Can Actually Tick

Four concentric validation rings, each with unique aperture criteria; only particles passing all four converge into a dense central cluster, symbolizing the checklist's four layers of verifiable criteria.
The checklist is arranged in four layers (capturability, retrievability, content & evidence, continuity), each requiring verifiable criteria; its value lies in criteria, not the number of items.

IN ONE SENTENCE

The checklist runs four layers — crawlable, retrieved, content and evidence, conversion — and every item needs a verifiable test.

Four layers: can it be crawled, is it retrieved, content and evidence, and conversion. Every item needs a test — 'we think we did that' does not count.

OUR POSITION

A checklist's value is in its tests, not its item count. 'Content optimised' is not a test; 'sample ten pages at random, each carrying a sourced statistic and a quotable opening line' is.

01

Layer one: crawlable

Retrieval crawlers allowed — test: 200 responses from those agents in server logs.

No additional CDN-level blocking — test: requests reach the origin in the logs.

Key content not dependent on client-side rendering — test: request the page without executing scripts and find a verbatim sentence from the body.

Structured data in server-rendered HTML — test: the same request contains application/ld+json.

02

Layer two: retrieved

The sitemap declares this domain's own indexable pages — test: sitemap domain matches page domain, and it contains no redirecting URLs.

No same-intent duplicates — test: no two pages' target questions can be described in one sentence.

hreflang x-default points at one fixed version — test: every locale declares the same x-default.

03

Layers three and four: content and conversion

Every page carries the three elements — test: sample ten pages at random; each has a sourced statistic, cited sources, and a quotable opening line.

Every page takes one position — test: you can state that page's judgement in one sentence.

AI channel isolated — test: analytics shows a distinct AI channel rather than Direct / Other.

A baseline exists with its definition archived — test: you can produce the question set, platform list and measurement date.

04

What you can skip

llms.txt: Ahrefs' study of 137,000 domains found 97% received zero requests in May 2026, and Google says it does not use them. Cheap to publish, but not a completed item.

AI-specific rewrites and AI-oriented chunking: named unnecessary in Google's guidance.

Data behind this page

30–40%

Relative visibility lift from adding statistics / citations / quotations

Source：Princeton GEO paper, KDD 2024, GEO-bench 10,000 queries，2024

97%

Share of published llms.txt files with zero requests in May 2026

Source：Ahrefs, across 137,000 domains，2026

Sources

  1. [1]Optimizing your website for generative AI features on Google Search．Google Search Central．2026-05-15
  2. [2]GEO: Generative Engine Optimization．Aggarwal et al., KDD 2024．2024

Updated 2026-08-10