Source-backed writing is easy to praise and hard to operationalise. This Research Note looks at how one public GEO handbook organises that discipline into files, chapters, and verification hooks. The repository is ferinazumaDEV/generative-engine-optimization-handbook, licensed CC BY-SA 4.0. Passages below that are inspired by that work are shared under the same license; the rest of mikaovo.ai is not relicensed.
What we observed from the public tree
The handbook presents GEO as a reference topic for answer engines: how to structure and publish so systems can understand and cite content. Publicly visible structure includes numbered Markdown chapters under docs/, a graded claims table, contribution rules that insist on “cite or flag”, citation metadata (CITATION.cff), JSON-LD about the work, an llms.txt map, changeloged releases, and CI that checks claims and links.
That combination matters more than any single tip. It treats trust as a publishing system: claims have maturity language, links are tested, versions are archived, and machines get a table of contents that points to human chapters—not a dump of marketing copy.
Graded claims instead of ambient certainty
Inspired by the handbook’s approach: every substantive claim should either carry a source or be marked as needing verification. Primary sources are preferred; secondary industry sources are labelled as such. A separate claims document holds the vocabulary and the graded table the rest of the work rests on.
For B2B product sites, the transferable idea is not “add more footnotes for SEO”. It is to stop mixing verified specs, hopeful forecasts, and hearsay in the same paragraph without labels. Auditors can then ask: which sentences are evidence, which are opinion, and which are still unchecked?
Chapter architecture that answers one job at a time
The public chapter list moves from foundations and engines into content, technical surfaces, authority, measurement, research cases, ethics, and a glossary. Inspired by that sequencing: start with definitions and mechanism, then writing patterns, then crawl/render/structured-data concerns, then trust signals, then measurement—before ethics and vocabulary cleanup.
Each chapter can still be long; the design win is navigational honesty. Readers (and agents following llms.txt) can enter at the right altitude instead of scrolling a single undifferentiated essay. Mika’s Knowledge Center uses a similar “one question per article” bias for the same reason.
Machine-readable guides beside human prose
The handbook ships llms.txt as a curated map of identity, start-here links, core chapters, and meta files. It also exposes structured about-metadata and citation files so “Cite this repository” style workflows have something durable to point at. CI link checking and a claims maturity gate turn documentation quality into something closer to software quality.
Inspired takeaway for audit sites: if you publish research notes, give crawlers and agents a small, honest index of what exists—canonical URLs, language alternates, and what each page is for—without pretending that a text file replaces robots rules or guarantees inclusion in any answer product.
What Mika adapts (and what we refuse)
We adapt the discipline: visible sources, dated review, no citation guarantees, and separation of verified vs. needs-check language. We refuse the anti-patterns this handbook also pushes against in spirit—hype, unsourced platform claims, and treating GEO as a secret ranking lever.
We do not scrape the handbook into dozens of thin mirrors. One Mika note about structure is enough to teach the pattern; readers who need the full reference should go to the original repository under its license terms.
License note for this article
This page credits Fernando Aporta Franco (ferinazumaDEV) and the Generative Engine Optimization Handbook. Commercial use and adaptation of that work are allowed under CC BY-SA 4.0 with attribution; derivatives must remain CC BY-SA 4.0. Passages in this article that are inspired by that handbook are shared under CC BY-SA 4.0. Other Mika site materials are not converted to CC BY-SA by this note.