01Publication standard
What a study must satisfy before it goes live
This is the standard Searlixa research must meet before publication. It is separate from our client measurement methodology, which describes how we measure a single brand. This page describes how we publish findings about many.
Nine requirements
A stated question
What was measured and why it matters, written before data collection begins rather than fitted to the result afterwards.
Dataset disclosure
Collection source, date range, sample size, inclusion and exclusion rules. A reader must be able to judge whether the sample supports the claim.
Repeatable method
Described in enough detail that an informed reader could run it again. If we will not describe how we got a number, we will not publish the number.
Stated limitations
Where results should not be generalised, what the method cannot see, and what would change the conclusion. This section is mandatory and is never a single sentence.
Evidence labels on every figure
Verified, Observed, Estimated, Demo or Updated. An unlabelled number does not publish.
Accessible charts
Unit, timeframe, sample and source on every chart, plus a text alternative that carries the same information for anyone who cannot see it.
Named author and named reviewer
Two different people. The reviewer checks the dataset, the coding rules and every Verified or Observed claim. Both are listed on the authors index.
Declared interest
Where a finding supports something Searlixa sells, that is stated on the page. It usually does, which is exactly why it needs saying.
Correction and versioning policy
Material corrections are published with a dated note describing what changed. We do not edit a claim out of a page and leave the page looking as though it always said the new thing.
What we will not do with our own research
- Turn an estimate into a headline. If the number is modelled, the headline says so.
- Publish a benchmark with no dataset behind it. A benchmark without data is a marketing asset wearing a lab coat.
- Compare non-deterministic outputs as if they were rank positions. Sampled distributions are reported as distributions.
- Recycle third-party research as our own. Where we cite others, they are named on the page and the claim is labelled Verified against their publication, not Observed.
Pre-registration
For studies where the result could plausibly flatter us, we publish the protocol before collecting data. The AI Visibility Index protocol is the first of these. It costs us the option of quietly abandoning a study that produces an inconvenient answer, which is the point.
→Next step