Why a POC, and why 90 days?

GEO suffers from a credibility problem: it is easy to promise and slow to verify. The 90-day POC exists to make the engagement falsifiable. Ninety days is the shortest window in which content can be published, indexed and incorporated into engine answers with measurable movement — local surfaces can shift in four to eight weeks, but a quarter gives the stochastic answer layer enough runs to show a trend rather than a coin flip. One brand, one market, one fixed question set: narrow enough to attribute cause, real enough to matter.

Week 1: Baseline

  1. Freeze the question set. Thirty to fifty real customer questions — category, brand, situation — agreed with stakeholders in writing. This set does not change for the duration; changing it mid-flight destroys the experiment.
  2. Run the baseline. Every question against the agreed engines (typically Google AI Overviews, ChatGPT, Perplexity, plus Gemini for local), fresh sessions, verbatim prompts, full screenshots archived with dates.
  3. Score it. Visibility, Citation Rate, Semantic Coverage, Brand Perception — 0-100 per dimension per engine, per the measurement standard. The baseline scores, with evidence, are the contract’s reference point.
  4. Diagnose. Every gap mapped to its class: indexation, content, entity, perception. The diagnosis determines the intervention; there is no generic playbook applied blindly.

Weeks 2-10: Intervention

Only what the diagnosis justifies, sequenced by leverage. Typical workstreams — chosen from, never all-of:

  • Entity remediation: NAP reconciliation, schema deployment, duplicate suppression, Bing Places and IndexNow where missing.
  • Answer content: answer blocks and FAQ pages for the highest-value unanswered questions; pricing, insurance, service-area and policy content where the question set demands it.
  • Review operations: honest cadence started, response routine installed — effects land in weeks, compounding after.
  • Corroboration: community and press artifacts, directory completion, forum-substrate triage.

Every intervention is logged with dates, so week-12 deltas can be read against what actually shipped. A light mid-point run at week 6 checks direction without contaminating the formal measurement.

Week 12: Re-measurement

The frozen question set, same engines, same protocol, same rubric — scored blind to the baseline where possible, then compared. The readout states, per dimension and engine: what moved, what did not, which interventions plausibly caused which movements, and what remains unfixed. Screenshots from baseline and week 12 sit side by side in the appendix. Then the decision: scale, adjust or stop — with evidence rather than faith.

What a POC does and does not prove

It proves whether specific interventions moved four-dimension scores on your question set, in your market, in that quarter. It does not prove permanent ranking (answers are stochastic and engines change), does not guarantee revenue (conversion is your site’s and sales team’s job), and does not generalize automatically to other markets — each new scope earns its own baseline. Any vendor showing you composite success numbers without a frozen question set, archived evidence and a named baseline is showing you theater.

Running it yourself

Everything here is public: the standards, the playbooks, the checklists. Teams with in-house capacity should run their own POC; our audit exists for teams that want the baseline and diagnosis done independently, and training for teams building the capability permanently. ChinaGEO is the international line of 海鹦云 (Haipengyun Holding Group). Contact: GEO@kchr.com.cn.

Frequently asked questions

Why publish your methodology if it lets others copy it?
Because the methodology is not the moat — disciplined execution is. Publishing the protocol also protects buyers: any GEO vendor, including us, can be held to a frozen question set, archived screenshots and a named baseline. Standards favor the honest.
What happens if the POC shows no movement?
That is a legitimate, valuable outcome and the report says so plainly: what was tried, what did not move, our read on why (wrong lever, too-competitive question set, entity problems needing longer), and whether continued work is justified. A POC that cannot fail is not a proof of anything.