Method first.
Findings when we have them.
Two studies in progress, both published protocol-first: the design, sample and reporting plan are live before any data is collected. It is a slower way to publish research and the only way that makes the eventual numbers worth citing.
What protocol-first publishing means
Protocol-first research publishes the study design — sample, method, coding rules, reporting plan and stated limits — before collecting data. Because the design is public and dated in advance, findings cannot be quietly reshaped to suit the publisher, and readers can judge whether the method is sound independently of whether the results are flattering.
Two studies, both open
Protocols are live and readable now. Participation is free and does not influence any result.
Protocol liveCross-brand AI citation studyAI Search Visibility Index
Which B2B brands AI assistants name and cite when buyers ask for recommendations. Fixed prompt set, five engines, five runs per prompt — with mention, citation and prominence coded separately.
Read the protocol
Accepting dataDistributions, not averagesSaaS Organic Funnel Benchmark
What B2B SaaS organic funnels actually convert at, segmented by ACV band, motion and stage, with every term defined before collection so contributors map onto the same meanings.
Read the methodSix rules we publish under
Written down because research standards that live only in someone's head tend to relax under deadline pressure.
Design before data
Protocol published and dated before collection begins. Any revision is noted on the page rather than made silently.
Limits stated up front
What a study cannot prove is declared alongside the design, so it cannot be omitted from the write-up later.
Distributions over averages
Skewed data reported as medians and quartiles. A single average across a skewed distribution describes almost nobody.
No estimates dressed as findings
If we have not measured it, the page says so. Placeholder sections stay visibly empty rather than being filled with plausible numbers.
Consent before attribution
Brand-level results published only with permission. Naming a company’s poor performance without consent is not research.
Inconvenient findings published
Including results that undermine our own commercial positioning. That commitment is why the protocol goes public first.
Why this section has no statistics yet
It would have been straightforward to publish plausible-looking numbers. Here is why we did not.
Statistics with no origin
- Figures with no stated sample or method
- “Our research shows” with no research behind it
- Numbers traced back to a competitor’s blog post
- Findings that always favour the publisher’s services
- Charts with no axis, source or date
Method now, numbers later
- Full protocol published before collection
- Empty slots left visibly empty
- Every external figure sourced and dated
- Commitment to publish unflattering results
- Aggregate dataset released with findings
About the research
Why publish a study protocol before you have results?
Because publishing the design first is what makes the eventual findings credible. A protocol registered in advance cannot be quietly reshaped to fit a flattering result, and it lets anyone assess whether the method is sound before the numbers arrive to distract from it.
Why not just publish an estimate in the meantime?
Because an estimate presented as research is a fabrication with extra steps. Marketing sites are full of statistics that trace back to nothing, and each one makes the category harder to trust. If we have not measured something, the page says so rather than filling the space.
Will you publish findings that are inconvenient for you?
That is the commitment, and it is the reason the protocol is public. If the AI Search Visibility Index shows citation rates are largely explained by existing domain authority rather than anything a GEO programme influences, that is a genuinely useful finding and we will publish it.
Can we participate or get early access?
Yes. Brands can request inclusion in the tracked set for the AI Search Visibility Index, and SaaS companies can contribute anonymised funnel data to the benchmark. Contributors receive findings ahead of publication and their own segment cut alongside the aggregate.
Will the raw data be available?
Aggregate data and full methodology, yes. Brand-level results will only be published with permission, because publishing a named company’s poor AI visibility without consent would be a hostile act dressed up as research.
Method and application
Be in the first edition
Request inclusion in the AI Search Visibility Index tracked set, contribute anonymised funnel data to the SaaS benchmark, or simply ask to be notified when either publishes.