Public research log

Show the work. Keep the caveats.

Observations from testing pages and AI engines, published with enough context to know what they do and do not prove.

How we treat evidence

We separate page readiness, observed citations, and category discovery because each answers a different question.

A readiness score evaluates the page. A citation check records what an engine returned for a chosen prompt at a chosen time. A category-discovery test asks whether the brand appears when it was not named in the question.

Really Good GEO methodology note

Branded questions measure entity retrieval, not discovery

Really Good GEO uses branded and unbranded prompts to measure different things. “What is [brand]?” can show whether an engine recognizes and retrieves information about a known brand; it does not establish that the brand enters the consideration set for a buyer who did not name it. We label branded and unbranded results separately and do not report a high branded citation rate as evidence of category visibility.

Really Good GEO methodology note

Readiness scores are not probabilities of citation

Really Good GEO treats readiness (page composition) and citation observation (engine behavior) as separate measurements. A page can score low on readiness and still be retrieved for a brand-specific question — because known-entity retrieval and category recommendation are different tasks. Readiness diagnoses content gaps; citation testing observes outcomes at a point in time. We do not present a readiness score as a probability of citation.

Really Good GEO's minimum viable citation test

The following protocol is Really Good GEO's methodological choice. It is not a standard published by any AI product.

  1. Define the commercial question before seeing results.
  2. Use a mix of brand-free discovery, comparison, and known-brand prompts.
  3. Run the same questions across named engines within a short time window.
  4. Record whether web search fired, which domains were cited, and what position the brand occupied.
  5. Repeat after the page change and again later to distinguish movement from noise.
  6. Publish the prompts, dates, exclusions, and limitations with any result.
What we will not claim: that a single test proves a permanent ranking, that a readiness score guarantees a citation, or that one engine's behavior represents every AI product.

Research agenda

  • Measure branded versus unbranded retrieval separately.
  • Track citation stability across repeated runs.
  • Compare page changes against a no-change baseline.
  • Document when search is used and when it is not visible.
  • Build category-level samples large enough to report responsibly.

Contribute a page to the evidence.

Start with a readiness audit, apply one clear change, then record what happened.

Run a baseline audit