"How do you protect personal data?" sounds like one question, but companies answer it in four different ways. Most of the confusion in this market comes from comparing tools built for different jobs. This is a map of the four: what each does well, where it fails, and the one case promptShield is built for.
The four strategies
| Strategy | What it is | Good at | Where it falls short |
|---|---|---|---|
| Governance platform | Classifies and labels data at rest across the organisation | Company-wide inventory and policy | You need one workflow protected this quarter, not a rollout over several quarters |
| Cloud redaction API | Programmatic detection, billed per call | Volume and automation | The documents must not leave your control, and no person sees the result |
| Managed redaction | A vendor redacts the files you send and returns them | Work you have no staff for | The transfer is itself the risk, and turnaround takes days |
| Manual blackout | A person with a marker or an annotation tool | Full control and clear accountability | Beyond a few pages: no detection, and misses go unnoticed |
A few real products make this concrete. Google's Sensitive Data Protection (formerly Cloud DLP), Amazon Comprehend's PII detection and the open-source Presidio SDK describe the same job in their own documentation.
They discover and classify personal data at rest and in streams, at volume, across an organisation. That is a hard problem, and they do it well. It is a different problem from one document that a named person must vouch for before it goes out. None of the four was designed for what now triggers most of this work: someone pasting a document into a chatbot.
Why they are hard to compare
The four can't be ranked on a single scale from worst to best. A governance platform and a manual blackout are not competitors. Treating them as one market produces comparisons that make no sense.
What matters is the cost of a single miss and the certainty you need. Volume alone is a poor guide. A pipeline scrubbing ten million log lines can accept a few misses, since each miss costs little and other defences catch the rest.
A law firm sending fifty discovery PDFs cannot. There, one leaked Social Security number can mean a sanction, a failed deposition or a malpractice claim. Take a vendor's claim of "99.5% recall", used here only as an example. On fifty documents, that still lets some names through.
What is missing
What is missing is a tool for a common professional case: under a thousand documents, where each file matters on its own. A clinic shares two hundred records with a researcher. An accountant prepares files for an external audit. A lawyer builds a data room, or an HR team releases an investigation file.
Here automation alone is most dangerous, because anonymisation fails silently. A missed name breaks nothing. The document leaves with a real identity in it, and you learn of it when it is too late to recall.
Why a person must stay involved
Responsibility means accepting the cost when something goes wrong, and a machine has nothing to lose. When a document leaves your office marked as anonymised, a person is vouching for it. You cannot vouch for what you never saw, so you must see what will be removed and confirm it.
A fully automatic tool asks you to trust its output without looking. Ninety-nine correct redactions do not make up for one missed name. Low volume is what makes review possible: you can read every page of fifty documents, but nobody reads fifty thousand. We explain more in "A machine can't be responsible. You can."
How promptShield compares
Documents stay on your machine. In the desktop app, pattern matching, entity detection and an optional local model all run locally. By default, no document, and no part of one, leaves the device. With no transfer there is no third party, which removes the main objection to cloud APIs and managed services.
It installs in minutes. There is no integration project and no consultants. It is a desktop application, so you can be redacting the same afternoon.
The cost is flat. There is no metering per call, character or page, so a busy month costs the same as a quiet one. A small practice can budget for it once.
A person reviews every result. The software proposes each detection as a region you review. It links a name across pages and filters obvious noise, so hours of scanning take minutes. You confirm, adjust, add or remove regions, and only then is the coded output produced. By default, a file with protected regions cannot be downloaded until they are verified in review.
Who this is for, and who it isn't
It is for you if you handle sensitive client documents yourself and answer for what leaves your hands. That covers law firms, clinics, accountants, financial advisers and HR teams. It also suits a compliance lead who needs a defensible answer for one workflow, or a team in a large organisation that needs an offline option for one process.
It is not for you if you need to clean millions of records in an automated pipeline with no human review. That is a real job, and cloud APIs and governance platforms are built for it. We would rather say so than sell you the wrong tool.
Common questions
Which strategy do I need?
It depends on volume and on the cost of one miss. For millions of records, use a cloud API or a governance platform. If the document itself is the secret, sending it to a service is the exposure we describe in this post.
Why are these tools so hard to compare?
Each was built for a different problem, so a feature list puts unlike things side by side. Compare them by the job you need done.
Where does promptShield fit?
It fits a professional with a few high-value documents to give an AI today. The work stays on their machine, and the identities are restored afterwards. That is the job we built our own engine for.