Three visible citation interventions
Six existing guides form three fixed editorial pairs. One guide in each pair receives a small visible evidence passage; its paired comparison retains its original body. The additions apply to every reader and crawler. They preserve the original text and its qualifications.
| Intervention | Changed guide | Comparison guide |
|---|---|---|
| Numerical comparison | A6 on A3 | A6 on A4 |
| Claim with conditions | A4 with bleed | Bleed and margins |
| Nearby source quotation | Actual-size printing | PPI and DPI |
The first addition makes a conditional numerical comparison explicit. The second keeps a claim, its conditions, arithmetic and boundary together. The third places a verified short Adobe quotation beside the printing setting it describes. These are adaptations of proposed generative-engine optimisation interventions, rather than established ranking rules.
Download the complete registration, source passages and fixed questions.
What the research does and does not establish
The original GEO study tested citation, quotation and statistical additions. Its Perplexity evaluation supplied source text as files. That is different from demonstrating that a newly published page earns live-web retrieval or a higher Google position.
A March 2026 structural-feature preprint motivates testing the presentation of unchanged facts. AgentGEO instead motivates targeted repairs and reports drawbacks from generic rewriting. An October 2026 observational study finds that retrieval explains much more citation variation than common page-formatting features. These findings leave meaningful uncertainty.
Text discovery and connected tools
An experimental llms.txt index points to selected text representations. The A3, A4 and A5 format pages also expose linked Markdown alternatives. The text comes from the same visible source as their HTML pages. The six citation-study guides are not mirrored in that Markdown cohort.
This follows the llms.txt proposal, rather than a universal search-engine protocol. Google explicitly says it ignores these special AI files for Search. Their use here tests compatibility and discovery by other tools; no Google ranking benefit is claimed.
The remote calculator interface provides a separate distribution experiment. Compatible MCP clients can request the actual calculations and receive structured results with units, methods and source URLs. A source URL returned by our own tool is not an independently earned citation.
How observations will be judged
The registration fixes twelve questions across the six guides. Observations must retain the engine, model, date, exact question, answer, citation URLs and a verifiable capture source. Clickable citations to the exact guide are counted separately from other links to this site and uncited brand mentions.
No pre-release answer baseline has been recorded, and no outcome rates are available. Missing observations remain missing. Audit scores, file downloads, our own test requests and tool-returned source links are not substitutes for independent search observations.
Three different-topic pairs provide an exploratory pilot, not a powered randomised trial. New-site indexing, simultaneous discovery changes, the future domain migration and changing search systems limit attribution. The reference methods remain the basis for checking the underlying print calculations.