AirPulse's frozen 30-day study covered 30,504 production responses. Mention status changed between consecutive observations 9.7% of the time. 35.9% of sufficiently repeated prompt-engine cells changed state at least once. Consecutive cited-domain sets had 0.396 mean Jaccard overlap. A separate rolling validation extracted on July 20, 2026 covered 30,146 responses: 10.0% flip rate, 33.9% mixed-state cells, 0.390 mean overlap. The two cohorts are never pooled.
AirPulse AI-Search Measurement Research
Every AirPulse study, method and public download, in one place. Each page states its cohort, its data-through date and what its result does not prove.
, updated
Research library
Public downloads
Aggregate statistics and templates only. No customer names, tenant identifiers, raw prompts or raw answers.
Aggregate stability study (CSV)
Frozen 30,504-response study and separate 30,146-response rolling validation, one aggregate row each.
Download the CSVStability study data dictionary (CSV)
Definition of every field in the aggregate stability file, including how flips and Jaccard overlap were computed.
Download the CSVPrompt-selection worksheet (CSV)
Scoring template for candidate prompts, with buyer stage, evidence source, six 0-2 criteria, owner and roster version.
Download the CSVMyChild question-neighbourhood aggregate (CSV)
Grouped counts from the post-hoc, single-brand MyChild analysis. No customer identifiers, raw prompts or raw answers.
Download the CSVTrial preregistration worksheet (CSV)
Fill-in template for the first-citation trial protocol, covering prompts, controls, engines, outcomes, decision rule, stop conditions and privacy review.
Download the CSVAI crawler audit worksheet (CSV)
Per-request log template with the operational role, the provider's documented verification method and a verified / failed / not-verifiable classification for each crawler event.
Download the CSVFirst 16 Perplexity observations (CSV)
The frozen citation-stability cohort, one row per valid observation with cited flag, citation-list position and exposed citation count. No raw answers or tenant data.
Download the CSVCore definitions
| Term | Definition |
|---|---|
| Prompt-engine cell | One exact prompt paired with one fixed answer engine |
| Valid run | One timestamped execution that returned a usable answer under the declared failure rules |
| Mention rate | Valid runs naming the brand divided by valid runs in the same scope |
| Own-domain citation rate | Valid runs citing the brand's registrable domain divided by valid runs |
| Source coverage | Distinct valid cited URLs or domains observed across repeated runs |
| Jaccard similarity | Intersection divided by union for two comparable source sets |
Full definitions, calculations and uncertainty rules are in the measurement methodology. Every study here is an observational production cohort, not a random sample, and none estimates the causal effect of one content change. To test one page change against a baseline and matched controls, follow the trial protocol.
Where the product applies it
- Prompt Visibility: repeated runs per prompt-engine cell.
- Citation Visibility: own-domain, third-party and source-history outcomes kept separate.
- Prompt Analysis: grouping and versioning the prompt roster.
- AI Traffic: referral sessions and verified crawler roles, separate from answer visibility.
- The free GEO Audit: the diagnostic entry point.
Update record
| Date | Change |
|---|---|
| July 13, 2026 | Frozen production study and corrected evidence strategy established |
| July 20, 2026 | Rolling validation added; first research cluster prepared for publication |
| July 27, 2026 | First-citation trial protocol and preregistration worksheet added to the library |
| August 3, 2026 | Crawler reference and audit worksheet added, after the protocol page's first observed citation on August 1 |
How this page was made
AirPulse analysed aggregate records from its production read replica in a read-only transaction. The public study excludes customer names, tenant identifiers, raw prompts and raw answers. Observations are grouped by brand, prompt and engine, ordered by time, and compared across consecutive valid runs. Harsh Songra wrote the page. The published numbers reproduce from the frozen evidence file.
Primary references
- Don't Measure Once: Measuring Visibility in AI Search (arXiv)
- Quantifying Uncertainty in AI Visibility (arXiv)
- How many AI-search runs are enough? AirPulse's frozen study
- Which AI search prompts should your brand track?
- How to prove a GEO change improved AI visibility
- Which AI crawlers visit your website, and what are they doing?
See the measurement running on your brand.
The free GEO audit runs your buyers' real questions against four engines. The research above explains how to read what comes back.