Research
How Lutiq Research measures things
Lutiq Research publishes original measurements and worked arithmetic about paid landing pages — not recycled listicles. This page is the shared method language behind those pieces.
What counts as a research page
A Research URL should answer a measurable question with:
- A clear claim in plain language.
- Method (who, when, what traffic, what was held constant).
- Numbers (n, rates, lift, or calibrated priors).
- Caveats (what the design cannot prove).
- Links to related Learn definitions and the product.
If we cannot attach method and caveats, it belongs in Learn — not Research.
Holdouts over “B beat A”
Relative winners inside a closed A/B or multi-variant race answer which option won among those tested. They do not prove the program beat the status quo. Where possible we report holdout comparisons: a reserved baseline experience versus the new system. See What is a holdout test? and the named standard: Lutiq Holdout Protocol v0.1.
Sample size and confidence language
We prefer stating n, observation windows, and confidence or uncertainty language over “statistically significant” as a slogan. Underpowered tests are labeled as such. See Landing page test sample size and CRO effect-size calibration.
Contextual uplift
Some gains appear only under routing rules (who sees which page) and cancel out when you only compare two pages in aggregate. Those designs are documented with worked examples — see What A/B tests cannot see.
What we will not do
- Invent customer logos or unmeasured case studies.
- Report lift without n or a clear comparison baseline when the claim requires one.
- Equate “AI generated a page” with “the system improved conversion.” Generation is an input; measurement against a holdout is the claim that matters.
Product relationship
Research is independent of marketing slogans. Product claims on lutiq.com and /about/ must stay consistent with how we measure: generate → approve → predict per click → holdout.