Google Cloud DLP recognises 213 kinds of sensitive text. Omit recognises 191 of them, on your own machine, with nothing uploaded. The difference that decides this is not the size of the list: it is that a cloud scanner has to receive your document before it can tell you the document was sensitive.
| How they compare | Omit | cloud DLP |
|---|---|---|
| Does the document leave your machine | No. Detection runs on the computer the file is already on | Yes. It is uploaded to the vendor cloud to be classified |
| Kinds of sensitive text recognised | 191 of Google's 213 text types, on a default install | 213, the published Google Cloud DLP catalogue |
| Accuracy where both tools compete | Matched or beat Google on all 21 of the types both attempt | Beat Omit on none of the 21, in either configuration |
| PII actually redacted, across all 34 types tested | Caught 30 of 34, on the tier bundled with every install | Caught 16 of 34 at best, though 13 misses ship no detector |
| Watching a whole organisation at once | Nothing. It protects the one desktop it runs on | Monitors many cloud apps and data stores from one console |
| Dashboards, case management, data lineage | One local report per run, and no central console at all | Central dashboards, incident workflows, and lineage |
| What it costs | One-time licence, with nothing metered | Typically per seat or per volume, billed on data processed |
| Works with the network unplugged | Yes, and you can prove it on an isolated machine in two minutes | No. Classification needs the DLP cloud |
| Rows where cloud DLP leads are marked plainly and explained below. We concede what we do not win. Verified 13 August 2026. | ||
These two sound like competitors and mostly are not. Cloud DLP answers the question 'what sensitive data does my organisation already hold, and where'. Omit answers 'do not let this document leave my desk with names still in it'. If you need the first question answered across dozens of SaaS apps from one console, buy a cloud DLP platform, because Omit does none of that and is not trying to. Where the two really do compete is the desktop, and specifically the case where sending a file away to be inspected is itself the disclosure you were trying to prevent.
Two different claims, measured two different ways. Coverage is counted from a mapping file inside our own engine against a committed snapshot of Google's published catalogue. Accuracy is a benchmark we built and ran against the live Google Cloud DLP v2 API on 22 July 2026, giving Google two runs: once on its default infoTypes, and once told exactly which infoTypes to look for.
Verified 13 August 2026.
All 213 map to an Omit detector in a data file inside the engine, and a test fails the build if one is left unmapped. But the mapping is not uniform: 171 are one to one, 34 are caught at a coarser label, and 8 only through an open-ended catch-all. 22 of the 213 need the opt-in coverage mode switched on, which is why the honest default-install figure is 191 and not 213.
The full coverage breakdownOf 34 entity types benchmarked, Google offers no text detector for 13, so 21 are genuinely comparable and no other exclusion is needed. On those, Omit matched or beat Google's defaults on 21 of 21 and its best configuration on 21 of 21. Three exclusions we argued for when this was first published have since stopped being true: the Fast tier now carries passport and driving licence detectors, and the harness bug that pinned our region config to Europe plus India is fixed.
Results, method, and where we are still weakThe question a buyer actually asks is whether the PII comes out redacted, and for that, a vendor shipping no detector is not an excuse: the passport number is still in the document. Counting every one of the 34 types, Omit caught 30 and Google caught 16 configured at its best, 4 on its defaults. In fairness to Google, 13 of its misses are types it does not offer, not detectors that failed. In fairness to you, all 4 of our misses are ours: two are open-ended catch-all types the Accuracy tier carries and the Fast tier does not, and two are ties at zero where no engine scored.
Both ways of counting, in fullAn earlier version of this claim said 30 of 34 against Google's best configuration. A scoring bug had counted the 13 types Google does not offer as Google failures instead of excluding them, turning races Google was never entered in into wins for us. Re-scoring the identical predictions moved our aggregate down and Google's up, roughly halving the measured lead. We would rather you heard that from us than found it yourself.
The correction in fullThat is inherent to the model and not a flaw in any particular product. The content has to arrive before it can be classified. For material that must not leave a device or a jurisdiction, being scanned and being shared are the same event, and the assessment you have to write says so.
Where our detection runsCloud DLP watches data across many SaaS applications and stores from one console, with dashboards, workflows and lineage. Omit protects the desktop it runs on and writes a local report. If your requirement is org-wide monitoring, we are not a substitute and we will not pretend to be.
How the pieces fitChoose Omit when the exposure is at the desktop, when the file must not be uploaded even to be inspected, or when you want a control people apply to their own documents before sending them rather than a report on what already went out.
Choose a cloud DLP platform when you need to find sensitive data across many cloud services centrally, with dashboards, case management and lineage. That is what those products are built for and an endpoint tool cannot replace it. And if passports or US bank routing numbers are the core of your use case, test both before you decide, because those are types we measurably lose.
Omit runs fully offline, and the Omit Redact beta is a free download. Read the Trust Center and verify the offline promise yourself.