GEO build journal: the dated trace, incidents included
This journal documents, date by date, the transformation of buddyleader.com into a site that answer engines can cite and act on — the framework is described in the SMB Agentic Ladder. It publishes what was done, failures included — never the list of upcoming targets. Media cost: ~CHF 0 (working time, a few cents of API calls, existing tooling).
The trace, date by date
| Date | Event |
|---|---|
| 2026-07-06 | Machine layer: llms.txt, machine product feed (mcp.json), robots.txt allowing 21 AI crawlers, 11-URL sitemap. |
| 2026-07-07 | J0 baseline measured BEFORE any content deployment: on the internal panel, 3 of 11 queries already cite the site on Perplexity (brand family) and 1 consumer query (rates-data family) cites it as the #1 source; 0 citations in the other families; 0 on ChatGPT (Bing index not yet built) and Gemini. |
| 2026-07-08 | Incident: two producers were writing to the same repository → pages 404 for several hours, the won citation disappears. Guardrails adopted: a single write path, sitemap diff before any replacement, live verification after every deployment. Lesson: availability IS a GEO criterion. |
| 2026-07-09 | Rate barometer in 6 languages (42 pages), then 4 comparison pages ×6 languages, tester-feedback page, admin-burden data page, 2030 analysis and white paper: the site grows from 11 to 94 URLs, all hreflang-linked. |
| 2026-07-13 | 5 articles ×6 languages (94→124). Internal audit. Machine feed v1.1 (six trades, Barometer dataset). Cache fix: machine files were served stale by the CDN → Cache-Control 300 s set. Entity graph reinforced (brand + legal entity, namesake disambiguation). |
| 2026-07-14 | This journal + the “SMB Agentic Ladder” framework page (124→136 URLs). Measurement panel FROZEN (30 queries, 5 engines) — SHA-256 fingerprint published below. |
The measurement protocol, pre-registered
A panel of 30 queries was frozen on 2026-07-13, split into six families: brand (4) · proprietary data (4) · category best-ofs (5) · use cases (10) · trust & data sovereignty (2) · comparisons and alternatives (5). Engines tested: Perplexity, ChatGPT, Gemini, Copilot, Claude. Metrics: linked citation, entity and price accuracy, source position. Cadence: J0 baseline (07.07), J+7, J+30, then monthly. J+30 success criterion: at least 4 of 30 queries with a linked citation on Perplexity, at least 1 on ChatGPT, and zero entity confusion on brand queries.
The exact panel is kept internal. Its SHA-256 fingerprint, published here, proves it will not be modified after the fact:
Fingerprint of the frozen panel (v1, 2026-07-13)622ffa0c2de95159346388bae8ac4124d65c9e15c0e82c2dc36ffe3cb6eba857
When results are published, the won and consolidated queries will be revealed in clear text, with evidence; the rest stay aggregated by family. We publish the trace, not the target map.
The cost, honestly
“~CHF 0 media” does not mean free. The real denominator: working time (founder + AI assistant), a few cents of API calls per page batch, and tooling already in place (static hosting, code repository). No advertising, no purchased backlinks, no agency.
Why publish — and why not everything
Publishing the method and the trace makes the case verifiable — that is what makes it citable. Publishing the failures (the low baseline, the 08.07 incident, the stale cache) makes the rest credible: a method without failures is not a method, it is an advertisement. We do not, however, publish upcoming target queries or competitive analyses: the trace, not the roadmap. That restraint is itself documented — it is a position, not an omission.
Frequently asked questions
Why publish a public GEO journal?
Because a verifiable case is worth more than a pitch: the steps are dated, the volumes measurable, the incident documented, and the measurement protocol was frozen before the results (SHA-256 fingerprint published). That is the difference between a demonstrated method and an agency promise.
What is the SHA-256 fingerprint of the panel for?
It is scientific pre-registration: the 30-query panel was frozen and its cryptographic fingerprint published before any measurement. The queries cannot be changed after the fact to embellish results — the fingerprint would no longer match. The exact panel stays internal so as not to hand out a target map.
How much did this visibility cost?
About zero francs of media: no advertising, no purchased backlinks, no agency. The real cost is working time (founder + AI assistant), a few cents of API calls per page batch, and existing tooling. The volumes are in the dated trace above.
Why not publish the exact panel queries?
Because the list of queries an SMB is targeting is a treasure map for its competitors. We publish the query families, the full protocol and the fingerprint of the frozen panel; won and consolidated queries are revealed with the results. The trace, not the roadmap.