Citeora self-study · launch baseline

Building a crawlable evidence layer from an 18-URL launch site

This is a first-party implementation record, not a ranking success claim. It documents the measured starting point, what changed on 10 October 2026 and what must still be observed before outcomes can be reported.

By Citeora7 min read

18

URLs in the starting sitemap

10

Substantive English resources added

72

Prompts in the registered GEO protocol

0

Invented rankings or customer outcomes

The starting point

At the beginning of the review, the public sitemap exposed 18 URLs. They represented roughly six route types across English, Traditional Chinese and Simplified Chinese: home, pricing, about, privacy, terms and a synthetic product-demo report.

The technical base was crawlable, with robots, canonical URLs, hreflang and a sitemap already present. The content layer was the constraint: there were no substantive guides explaining AI SEO, generative engine optimization, AI visibility or citation monitoring, and no published measurement method.

The implementation decision

Instead of generating thin translations, the first release adds ten English resources from one typed content manifest. The manifest drives routes, metadata, schema, internal links and sitemap entries, which reduces the chance that a published page becomes orphaned or receives a mismatched canonical.

The four pillar topics deliberately answer different questions. AI SEO describes the operating workflow. GEO describes how source information is prepared for generative retrieval. AI visibility describes observations across a fixed prompt panel. Citation monitoring records the sources exposed with an answer.

What was implemented

LayerChangeEvidence
DiscoveryExpanded the sitemap and added a global Resources linkProduction sitemap and navigation
ContentPublished four pillars, four practical guides, a hub and a methodologyTen unique canonical pages
StructureAdded visible breadcrumbs, internal links and matching JSON-LDCollectionPage, TechArticle, HowTo and BreadcrumbList
MeasurementRegistered a fixed 72-prompt, three-engine monitoring protocolVersioned prompt IDs, hashes and validation tests
SafetyKept live recurring model spend fail-closedAuthenticated preflight, dry-run default and hard caps

What the numbers do—and do not—show

The figures above are implementation counts taken from the code and sitemap, not search-performance results. More URLs do not automatically create authority. The new pages have to be crawled, judged useful, earn legitimate references and accumulate repeated visibility observations.

No uplift in ranking, traffic, mentions, citations or revenue is claimed. The public methodology pre-registers the sample and formulas so later results cannot redefine success after the answers are collected.

Measurement and next checkpoint

The registered weekly panel contains 72 fixed prompts across English, Traditional Chinese and Simplified Chinese. A full wave across ChatGPT, Gemini and Perplexity represents up to 216 provider attempts. Failed calls remain failures; they are not converted into zero mentions.

The next public checkpoint should include only validated observations with the prompt version, engine, collection period, eligible denominator and failure rate. Until the asynchronous runner and recurring spend ceiling are approved, the scheduled route performs a zero-spend preflight rather than claiming a live baseline.

Review the reproducible method

The methodology publishes prompt allocation, run conditions, formulas, quality gates and limitations before any benchmark result is released.