The five-stage maturity model
Adoption is easier to reason about as stages than as a percentage. Each stage is defined by what the organisation can do, not by what it says it believes.
- ›Stage 1 — Ad hoc. Someone pastes the brand name into ChatGPT occasionally and screenshots the result.
- ›Stage 2 — Measured. A prompt corpus exists and is sampled on a cadence. Nobody is accountable for the number.
- ›Stage 3 — Owned. A named person owns Recommendation Share and reports it alongside pipeline metrics.
- ›Stage 4 — Optimised. Entity, content and corroboration work is planned against measured gaps rather than intuition.
- ›Stage 5 — Compounding. Answer-layer performance feeds product, PR and partnership decisions, not just marketing.
What separates stage 2 from stage 3
The jump from measuring to owning is where most programmes stall, and the blocker is almost never tooling. It is that no single function believes the metric belongs to them: SEO sees a content problem, PR sees a coverage problem, product sees a positioning problem.
The organisations that cross it do one unglamorous thing — they put Recommendation Share on the same weekly dashboard as pipeline, with one name against it.
Category patterns
Categories where the buyer researches conversationally and the purchase is considered — B2B software, healthcare, financial services, professional services — show the fastest adoption. Impulse and habit categories show the slowest, because the answer layer intercepts less of their demand.
Regulated categories are a special case: they adopt late but move fastest once they start, because they already maintain the canonical entity data that GEO depends on.
Observed failure modes
Three patterns account for most stalled programmes: publishing volume without structure, optimising for one engine and assuming the others follow, and treating a single week's sample as a trend when engine responses are inherently non-deterministic.