Pre-registered a calibration study for my GEO tool. Result: null

Draft published with the note on geoready.dev. These files let a reader check the figures in the note.

PREREGISTRATION.md: the analysis plan, written before the full run. Its SHA-256 is 441fcbb422964fdf2bf2b63eb02181ccbea9e7c5590d3e770399d567cd373ff8 (check with sha256sum PREREGISTRATION.md).

results.csv: one row per page (341 pages), in shuffled order. No domain, address or headline.

analysis-results.txt: the output of the pre-registered analysis (run on 2026-10-17 on the exact rank, four decimals).

Columns of results.csv

Column Meaning

rank_band Band of the site's Tranco rank

ln_rank_1dp Natural log of the exact rank, rounded to one decimal (used as the popularity control)

score GEO score of the page, 0 to 100 (rubric v2, GEO Optimizer 4.19.3)

google_ai ... negative_penalty Category scores and the negative-signal penalty of the same audit

words Words on the page

blocks_ai_bot 1 if the site's robots.txt blocks at least one AI bot named in AI_BOTS

domain_cited_A, domain_cited_B 1 if the page's domain appeared among the sources Perplexity Sonar returned for question template A or B

page_cited_A, page_cited_B The same, for the exact page

Question template A: "What are the most reliable sources on: [headline]". Template B: "Explain this topic and cite your sources: [headline]".

Logistic regression of domain_cited_A on score / 10, ln_rank_1dp and log(words). On this table: odds ratio per 10 score points 0.962 (95% confidence interval 0.783 to 1.184), p = 0.717. On the unrounded rank the analysis gives 0.9625 (0.7827 to 1.1836), p = 0.717; the note rounds it to 0.96 (0.78 to 1.18).

One engine (Perplexity Sonar), one date (17 October 2026), one model version. The question is built from each page's own headline, so the outcome measures whether a page is cited when the question names its own topic. The sample favours reachable, English-language sites that publish a sitemap. Observational data: no causal claim. The sampling, audit and API-call scripts are not published.