How Reffed Academy is built and maintained.
The standards every Academy lesson is held to, the source verification process, and our public commitment to honest framing of what we know versus what is still being learned in a field that is barely two years old.
During our own internal QA of Academy v1, we identified that several specific statistics across multiple lessons were either attributed to the wrong study or cited at incorrect numeric values. The underlying methodology and concepts were sound, but the specific quantitative claims that anchored them needed re-verification against primary sources.
Rather than ship partially-verified content to paying members, we paused new founding-member enrollment and ran a full claim-by-claim audit of every lesson, sales page, and blog post. That audit is now complete. Every numeric research claim has been verified against its primary source and cited inline, reframed as our own observation where it came from our experience rather than a study, or removed where it could not be substantiated. Founding-member enrollment is open again.
This page is the permanent public record of how we work. If you find any claim in any lesson that you cannot trace to a cited source, email matt@reffed.ai directly. We will fix it.
Every numeric claim traces to a primary source.
Most GEO content on the internet right now is a chain of secondhand citations. Article A cites Article B's statistic, which cites Article C, which paraphrased a primary paper, which the original article got wrong. The numbers drift further from the source at every hop. We hold the Academy to a stricter standard.
Numeric claims must trace to a primary source
If a lesson says "+22%" or "115%" or "87% of citations" — we name the study, link the paper, and quote the original authors. If we cannot find the primary source, we either remove the number or reframe the claim as our own observation.
Engine behavior claims reflect current testing
Statements like "ChatGPT does X" or "Claude weights Y" reflect engine behavior as tested in the current month, not behavior we read about in someone else's article from last year. AI engines change. The lesson timestamp shows when the claim was last verified.
Reffed audit observations are labeled as such
When a claim comes from analyzing thousands of Reffed customer audits rather than from an external study, we say so explicitly: "in our analysis," "based on Reffed audit data," "across the audits we have run." We never present internal observations as external research findings.
Opinion is labeled opinion
"Project pricing is usually better than performance pricing for new operators" is an opinion grounded in experience. We frame those as recommendations and trade-off discussions, not as universal truths backed by research that does not exist.
Lesson timestamps and source notes
Every lesson carries a last-verified timestamp. When a section of a lesson references a specific source or study, that citation is inline and clickable. When something changes — a paper retracts, an engine updates its behavior — the lesson is updated and the timestamp moves.
Honest about what we do not know
GEO is a field that is roughly two years old as a named discipline. Many specific questions ("does engine X weight signal Y more than signal Z?") do not have rigorously-established answers yet. Where evidence is thin, the lesson says so. Confident-sounding claims about specifics no one has measured are how the rest of the field misleads readers. We try not to do that.
Where our claims come from.
For full transparency, here are the source categories that count as authoritative in the Academy curriculum. Anything cited as a research finding traces to one of these.
- Peer-reviewed academic research. The Aggarwal et al. KDD 2024 paper ("GEO: Generative Engine Optimization") is the foundational research and is cited extensively. Other peer-reviewed GEO and retrieval-augmented generation papers as they appear.
- Engine operator documentation. Official robots.txt and crawler documentation from OpenAI, Anthropic, Google, Microsoft Bing, Perplexity, and Apple. Schema.org specifications. Bing Webmaster Tools documentation.
- Major industry citation studies with published methodology. When third-party studies cite specific numbers (Profound, Otterly, Brandlight, SparkToro, Semrush, Ahrefs), we cite them when methodology is public and reproducible. We do not cite vendor-claimed numbers as research findings.
- Live engine testing. For "engine X currently does Y" claims, the engine in question is queried during lesson maintenance and the output verified directly.
- Reffed customer audit data. Aggregated across the customer base, anonymized, and labeled clearly when used: "across the audits we have run," "Reffed customer data suggests," and similar framing.
- Practitioner experience. For operator business mechanics — pricing models, client engagement, scaling decisions — claims grounded in working GEO operator experience are labeled as such ("in our experience," "operators who scale to N clients typically").
v2 is live. Maintenance is continuous.
The v2 audit covered every lesson in Foundations, Quickstart, and Certification, plus all blog posts and sales-page claims. It is complete and live, and founding-member enrollment is open at the original prices. Going forward, the same sourcing standard applies to every new lesson and every revision: numeric research claims trace to a primary source and are cited inline, our own observations are labeled as such, and anything we cannot substantiate does not ship.
Members keep their access continuously. The quiz infrastructure functions throughout. Lesson dashboards remain available. As lessons are updated, they update for everyone — current members included — at no charge.