This page aggregates measured, corroborated evidence: donors' iPhones recorded apps contacting this domain, every claim below was reported by separate accepted donations and published under a human verdict (later rows for an approved app inherit its verdict). A contact is a measurement, not an accusation — it does not show what data was sent.
sp.yelp.com
classified as first-party · 2
What our research panel found AI-drafted · human-reviewed
We ask several independent AI models the same question and publish their answers separately — cross-vendor agreement is evidence against fabrication, and disagreement is worth seeing. 3 of 3 models answered, overall confidence medium — undisclosed split: anthropic=false, openai=true, gemini=false. A human reviewed this domain's claims before anything here published.
Anthropic Claude — first-party
sp.yelp.com is a subdomain of yelp.com, the app publisher's own domain, so this is first-party infrastructure; the 'sp' prefix hints at an event/telemetry endpoint but I cannot verify its function, so I do not classify it as analytics. Since the label data provided shows nothing declared at all (possibly incomplete rather than empty) and first-party contact does not by itself prove collection of labelable data, I cannot conclude the contact is undisclosed.
OpenAI — first-party
The domain is a Yelp subdomain and therefore most likely first-party. The app declares no data collection or sharing, so this observed network contact appears undisclosed.
Google Gemini — first-party
The domain sp.yelp.com shares the same parent domain as the app developer (Yelp), making it a first-party endpoint. First-party network contact alone does not inherently indicate an undisclosed data flow without evidence of undeclared data collection.
Apps measured contacting this domain
| App | Label gap | Donations | Contacts | Seen |
|---|---|---|---|---|
com.yelp.yelpiphone | — | 4 | 8 | 2026-07-28 → 2026-07-29 |
The corpus grows one report at a time: donate your App Privacy Report — anonymous by design.
Think something here is wrong?
Evidence and research improve when people push back. Tell us what to re-check — a misattributed domain, an outdated classification, a claim about your own app — and a human will review it. Responses can take up to 30 days.