The RankPropel method: turn search improvements into reusable evidence

Better search visibility is useful when it helps the right person discover, understand, and use your product. Our method starts with that outcome and works backward to a change we can test.

This is a working method, not a claim of measured customer results. The example below is invented.

1. Define the buyer's decision

Choose one audience, one buying question, and one conversion event. A fictional scheduling app might ask: can a small team evaluate calendar compatibility and complete its first booking? That question gives its integration guide a purpose and its onboarding flow an outcome to measure.

2. Save a baseline

Record the pages, query or prompt set, dates, market, engine, and measurement method. Keep search clicks, AI mentions, linked citations, referral sessions, signups, activation, and revenue separate. A citation is not a visit; a visit is not a customer.

For an AI prompt panel, record successful runs and failures separately. Report mentions per successful run and linked citations per successful run, for each engine. Repeat the same panel under comparable conditions. This measures the panel, not total market share or every buyer's experience.

3. Fix the strongest supported problem

Check access, page accuracy, and the buyer's path to action. Compare initial HTML with the rendered page; Google can render JavaScript, while other crawlers may have different capabilities. Test the actual surface instead of treating a user-agent string as proof of how an engine sees a page. Google JavaScript guidance.

Google's current AI guidance favors foundational SEO and useful original content. Neither special AI schema nor llms.txt is a requirement for Google visibility. Use structured data to represent real visible information, and write for readers. Google AI guidance.

For the fictional scheduling app, a useful experiment would be to replace an unclear integration description with a tested setup walkthrough and a working example. The predicted outcome is more completed setups; the measurement should test that prediction.

4. Measure and retain uncertainty

Record what changed and when. Compare equivalent periods and, where possible, similar unchanged pages. Annotate product launches, campaigns, seasonal changes, and search updates. A single change reduces ambiguity but does not establish causality.

Record observed referral evidence separately from a user's answer to “How did you hear about us?” Deduplicate conversions before combining these signals, and retain an unknown category. Self-report and analytics each miss parts of the journey.

5. Turn the lesson into a reusable asset

A useful lesson can become a diagnostic check, a checklist, a book exercise, or a product feature. Give it a source, scope, limitations, and a review date. Publish general methods with original examples; client identities and results require a separate confidentiality review and permission where applicable.

Research is a source of hypotheses. The GEO research repository offers an experimental implementation, but historical benchmark results do not guarantee current rankings, citations, or revenue. Test applicability before adopting a tactic.

6. Use GitHub where it serves developers

A runnable example or SDK can help someone evaluate a developer product. Explain the problem, quick start, expected output, limitations, and support route in its README. Use relevant topics and link to useful product documentation. GitHub documents READMEs and topics as discovery and orientation tools; neither is a promise of Google ranking gains.

Measure whether repository visitors activate the product. Publish something worth using, then improve it from actual questions and failures.

A reusable experiment card

Field	Write down
Buyer and task	Who needs to accomplish what?
Evidence	What did we observe, when, and using which method?
Hypothesis	Why should this change improve the buyer's outcome?
Change	Which page or product behavior will change?
Primary metric	Qualified lead, activation, or another defined outcome
Guardrail	What must not regress?
Evaluation	Window, comparison, sample size, and known confounders
Decision	Keep, revise, revert, or collect more data
Reuse	Which general lesson is safe and useful to share?

Start with one card and one improvement. The result can be positive, negative, or inconclusive; all three can improve the next decision.