Stats
How to test if two metrics really move together
cross-domain-correlator a saved expert workflow your assistant follows, in Claude, ChatGPT or Cursor
Guided "does X move with Y" studies across data domains, vitals vs clicks, share vs traffic, ad spend vs organic, citations vs sessions, run properly: aligned windows, the right correlation method, the caveat every time, and the causal test design offered when stakes are real.
Try it
do these correlate
Private beta Part of the quattr-ai-search plugin, rolling out org by org. Request access ↗
Say any of these
does X correlate with Yis speed related to our trafficdo citations move sessionscorrelation between spend and organicdoes share track revenue
You never name the skill or a tool, the assistant routes on what you asked. These are the phrasings it recognises for this one.
And when it stays out of the way
Not for a single-domain trend (weekly-search-report), an is-it-real verdict (significance-referee), or a cause hunt (anomaly-investigator).
Use it instead of anomaly-investigator when the question is a RELATIONSHIP, not an event; use significance-referee instead for one metric's change.
What it runs
correlate_domains
1 governed analysis, composed and scoped for you. Named here for transparency, not because you have to ask for them.
The numbers behind it
Metrics it reads, each defined in the reference
AI referral conversion rateAI referral conversionsAI referral sessionsAI referral source splitLargest Contentful Paintp-valuePearson correlationSpearman correlation
What it refuses to do
- The output vocabulary is bounded: "correlates", "moves with", "consistent with", never "drives", "causes", "explains".
- An insignificant correlation is reported as "no detectable relationship at this power", not "no relationship".
- Every statistic shown (r, ρ, p) must come verbatim from a tool result. An unsupported metric pair yields a descriptive side-by-side and the honest note that the correlation read isn't available for that pair, never a hand-computed number.
What the answer looks like
Is speed related to our traffic?
Analysis plan, every step, resolved for you
- Routed to the governed workflow
cross-domain-correlator - Dates bound explicitly
Jun 1 to Jul 31, 2026 - 1 analysis queued
correlate_domains - Honesty gates armed
significance α=0.05 · scope echo · freshness
One source can't answer this, they join
- Lighthouse/CWVweekly LCP
- Google Search Consoleweekly organic clicks
Weeks with slower LCP do see fewer clicks, a moderate, significant negative correlation.
- Spearman ρ
- −0.58 rank-based, robust to outliers
- Windows aligned
- 9 weeks same Sun to Sat buckets on both series
Jun 1 to Jul 31, 2026organic
⚠ Observationalas of Aug 1, 2026 (GSC lags 1 to 2 days)IllustrativeOpen in Quattr ↗
The relationship is unlikely to be chance, but it is a candidate, not a conclusion.
- Verdict
- Real — significant
- p-value
- 0.031
Correlation ≠ causation — candidates are not conclusions.
Jun 1 to Jul 31, 2026organic
⚠ Observationalas of Aug 1, 2026 (GSC lags 1 to 2 days)IllustrativeOpen in Quattr ↗
If the stakes are real, run the causal test: fix LCP on a matched page set and watch the split.
A simulated conversation on fictional data. Scope, freshness and the ⚠ Observational marker are attached by the same director that renders every demo here, not written by hand.
See this Skill inside a real analysis
All runs →A recorded run where this Skill does the work, the prompts as typed, and the decision that came back.
Want this Skill running on your own accounts? Every Quattr customer works with an AI Search Strategist who sets it up with you.