A privacy policy can be long and still omit the features a careful reader looks for first: a dated update line, a numeric retention bound, a children’s statement, or a clear change-notice method. We froze a 12-dimension rubric, retrieved 24 purposively selected English privacy notices from a Cursor-hosted USA cloud VM, and scored what the retrieved text made observable.
The sample’s mean observability index was 80.5 (median 81.8). That index is the share of applicable rubric points that were clearly or partly met. It is Tidak a privacy grade, trust score, compliance certification, or claim about real-world logging.
Every notice clearly named data categories, recipient categories, and a rights-plus-request path under our definitions. The most common zeros were change notification and children’s data (each absent in 8 of 24). ProxySeller multi-exit conditions were blocked (no PROXYSELLER_* env vars). CometVPN appears as scored policy text and affiliate context only — no VPN tunnel.
Apa yang kita mengukur
We asked a document question only: which pre-registered features appear in the public notice a visitor can fetch without logging in? The sample spans five strata — VPN (7), proxy (5), privacy tools (4), mainstream platforms (4), and legal-tech vendors (4). Selection was purposive for FreePolicyTools readers, not a random web census.
After the first crawl we applied protocol-allowed URL repairs and two disclosed legal-tech substitutes where Cookiebot and PrivacyPolicies.com returned HTTP 403. NordVPN’s product HTML returned 403; we scored the official PDF linked from the Nord Security hub via pdftotext. The rubric and collector live in the downloadable package. Related reading: our guide to Apa yang harus dimasukkan oleh kebijakan privasi website.
Finding 1: every notice cleared the same three basics
Across all 24 scored notices, data categories, recipient/processor categoriesDan rights plus a request path each scored a clear 2. In this sample the floor was not “is there a policy?” but whether later dimensions stayed specific.
We did not submit access or deletion requests. Separate FreePolicyTools work on privacy-request forms before submit examines data leaving a form before the user clicks submit.
Finding 2: change notices and children’s statements were the widest gaps
Eight notices (33%) scored 0 on change notification. Another eight (33%) lacked a children’s / under-age statement under our definition. Sale/share-for-ads stance was absent in four notices. Numeric or event-bounded retention was missing or only “as long as necessary” in several rows.
These gaps describe wording observability on the collection date. A company may handle children’s data carefully without saying so in the fetched notice — and the reverse is also possible. Authoritative framing for notices includes the ICO right to be informed guidance Dan GDPR.eu overview.
Finding 3: VPN and proxy “no logs” language still varied
For VPN/proxy rows we scored network_telemetry separately. Specific statements (excluding traffic content, DNS, or connection timestamps, or stating IPs are not retained) scored higher than slogan-only “no logs” marketing.
Policy prose is only one input. Our proxy routing research Dan proxy encryption research measure client and authentication behaviour. When you evaluate residential or datacenter proxies, confirm the exact product path with the provider.
Finding 4: stratum averages differed without ranking privacy
Mean observability indexes: privacy tools 83.0, proxy 82.5, mainstream 81.8, legal tech 79.5, VPN 77.4. On shared dimensions only, VPN averaged 76.0 versus 81.8 for mainstream platforms. Proton VPN’s retrieved notice was comparatively sparse on children, cookies, change notice, and law-enforcement procedure despite strong IP-retention language — which lowered the observability index without judging Proton’s engineering.
Legal-tech vendors did not automatically top the index. Dense notices help readers; they are not proof of safer backends. For behavioural VPN-tunnel questions see Inside the VPN Tunnel.
Finding 5: high scores are not behavioural evidence
Top ranks in this purposive list reflect rubric coverage in the fetched text. They do not show that those organisations retain less data, honour deletion faster, or resist legal process differently. We did not audit production systems or claim any provider “lies.”
For VPN product context — including residential-IP VPN marketing — see our Ulasan KometVPN. CometVPN is VPN-only; this study did not tunnel through it.
A practical checklist from the rubric
- Specific effective or last-updated date.
- Named data categories beyond “information you provide.”
- At least one numeric or event-plus-maximum retention bound.
- Recipient categories with some specificity.
- Cross-border transfers plus a safeguard class, if transfers occur.
- Rights listed and an email, form, or account path.
- Legal-process language beyond “comply with law.”
- Explicit sale / share-for-targeted-ads stance.
- Cookies, pixels, or SDKs described in-body.
- Children’s age threshold and practice.
- How material changes are communicated.
- For VPN/proxy: specific connection/traffic retention claims.
Need a starting draft for your own site? Try the Penerus kebijakan privasi offison — it is not a substitute for counsel or for this observatory sample. Also see the FTC privacy and security guidance.
Bukti, eksklusi, dan pengungkapan
Public package under /research/privacy-policy-observatory/: protocol, sample, rubric, fetch log, redacted extracts, scores, analysis, URL amendments, SHA-256 manifest. Raw bodies stay private and untracked.
No CAPTCHAs were solved, no accounts created, and no privacy requests submitted. N=24 is purposive and small. Do not generalise to “the web.”
Charts and ranked sample
Mean observability index by stratum
Purposive N=24. Descriptive means only — not a ranking of real-world privacy.
Dimensions most often scored zero
Zero means the feature was not found in the retrieved primary notice under the frozen rubric.
| Layanan | Stratum | Index | Sum / max |
|---|---|---|---|
| Termly | legal tech | 95.5 | 21/22 |
| AndometVPN | vevpn | 91.7 | 22/24 |
| IPRoyal | Proksi proksi | 91.7 | 22/24 |
| Surfshark | vevpn | 91.7 | 22/24 |
| Betina | privacy tool | 90.9 | 20/22 |
| Microsoft | mainstream | 90.9 | 20/22 |
| ExpressVPN | vevpn | 87.5 | 21/24 |
| Proxy-Seller | Proksi proksi | 87.5 | 21/24 |
| 1Password | privacy tool | 86.4 | 19/22 |
| DuckDuckGo | privacy tool | 86.4 | 19/22 |
| Webshare | Proksi proksi | 83.3 | 20/24 |
| CookieYes | legal tech | 81.8 | 18/22 |
| mainstream | 81.8 | 18/22 | |
| Amazon | mainstream | 77.3 | 17/22 |
| Meta | mainstream | 77.3 | 17/22 |
| Bright Data | Proksi proksi | 75 | 18/24 |
| NordVPN | vevpn | 75 | 18/24 |
| Oxylabs | Proksi proksi | 75 | 18/24 |
| Didomi | legal tech | 72.7 | 16/22 |
| IVPN | vevpn | 70.8 | 17/24 |
| iubenda | legal tech | 68.2 | 15/22 |
| Isyarat | privacy tool | 68.2 | 15/22 |
| Mullvad VPN | vevpn | 66.7 | 16/24 |
| Proton VPN | vevpn | 58.3 | 14/24 |
download penelitian
Redacted public evidence only. Raw HTML/PDF bodies remain private. Manifest includes SHA-256 hashes.
Pertanyaan yang sering diajukan
Pengungkapan: This article contains ProxySeller and CometVPN affiliate links. FreePolicyTools may earn a commission if you purchase through them. These offers are separate from the research findings and did not change the study’s methods, scores, or limitations.