Personal Dictionary: Terms PiiBlocker Always Masks

The Personal dictionary is a list of terms that PiiBlocker always treats as personal data. It covers what the built-in detectors miss: an uncommon surname, an internal patient reference format, a project codename that should not reach an AI service. Each entry gets a category, so it is masked with the right placeholder, and entries live only in your browser. This guide covers what belongs in the dictionary, how categories work, how it differs from the Allowlist, and the honest alternatives. Short version: the Personal dictionary entry on our support page.

What built-in detection misses

PiiBlocker's detectors handle the common cases well: names in ordinary sentences, email addresses, phone numbers, card numbers, government IDs. No detector is complete, and three kinds of term get past most of them.

Uncommon names. A surname the model has rarely seen, or a name that is also an everyday word, can be read as plain text.

Internal references. A patient reference or a matter number identifies a person inside your organisation and looks like noise to anyone outside it, including a detector that has never seen your format.

Codenames. "Project Heron" is not personal data in the usual sense, but it may be exactly what your employer does not want sitting in a third party's logs.

The dictionary closes these gaps. Entries are checked before the main detection runs, so a term on the list is caught every time, whatever the sentence around it looks like. That reliability is the reason to use it for the few terms that matter most to you.

Adding entries and choosing a category

There are two ways to add a term. The quick way is in the chat box itself. Select the text, right-click, choose "Mask with PiiBlocker", and tick "Always detect this". The term is masked now and saved for next time. The other way is to open the Personal dictionary in the extension and type entries in, which is easier when you have several to add at once.

Each entry gets a category, and the category decides the placeholder. A name in the dictionary shows as [PERSON_NAME] in your prompt, so the AI still knows a person is being discussed and can write a sensible reply around the gap. An entry filed under the wrong category is still masked. The reply may just read oddly, because the AI was told a codename was a person.

The dictionary list has search, filtering and sorting by newest, along with an entry count. Those start to matter once you pass twenty or thirty entries.

Where entries are stored

Dictionary entries only live in your browser. They are never sent anywhere. The free version of PiiBlocker makes no network calls at all, and you can confirm that in the Network tab of DevTools while you add an entry.

Entries are stored locally with the same encryption PiiBlocker uses for its other sensitive data. There is no account and no sync, which has two consequences worth knowing. Your dictionary does not follow you to another computer or browser profile, so a term you added at work is not protected at home until you add it there too. And there is no export or import yet, so a long list has to be rebuilt by hand on a new machine. Export is planned.

To remove an entry, delete it from the dictionary list. Each delete asks you to confirm, so a stray click does not lose a term you rely on. Uninstalling the extension removes the whole dictionary, because Chrome deletes an extension's local storage along with it.

Dictionary or Allowlist?

The two lists do opposite jobs, and mixing them up is the most common mistake.

The Personal dictionary says: always mask this. Use it when PiiBlocker let something through that it should have caught.

The Allowlist says: never mask this. Use it when PiiBlocker keeps flagging something that is not personal data, such as a ward, a department or a product named after a person.

A quick way to choose is to ask what went wrong. If a real name reached the chatbot unflagged, that is a dictionary entry. If the dialog keeps asking about your own organisation's name, that is an Allowlist entry.

One caution applies to both lists. Short, common words make poor entries. A dictionary entry for a common first name is masked wherever it appears, including where it is not a person. That is the safe direction to be wrong in, since over-masking leaks nothing, but it makes replies harder to read.

Alternatives to the dictionary

A dictionary is the right tool for terms that recur. For other situations there are simpler options.

Mask it once. For a name you will never type again, right-click and mask it without ticking "Always detect this". Nothing is saved.

Use a dictionary-only extension. Tools such as Privacy Protector for ChatGPT work entirely from a word list you maintain, with no automatic detection. That gives full control and no false positives, and it catches nothing you did not list. PiiBlocker detects automatically and uses the dictionary for the gaps. Our comparison of masking extensions sets the approaches side by side.

Substitute by hand. Typing "Client A" in place of a name needs no tool at all. It works until you paste a document that contains the real name, which is the case automatic detection exists for.

For a list that must apply to every seat in an organisation, see DLP versus browser extensions.

Frequently asked questions

Are dictionary entries sent to PiiBlocker's servers? No. Dictionary entries only live in your browser. They are never sent anywhere.

Why does each dictionary entry need a category? So it is masked with the right placeholder. A name in the dictionary shows as [PERSON_NAME] in your prompt, which tells the AI that a person is being discussed.

Can I export my Personal dictionary? Not yet. Export and import are planned for a future update, so for now a dictionary has to be re-entered on a new computer or browser profile.

What is the difference between the Personal dictionary and the Allowlist? They do opposite jobs. The Personal dictionary lists terms PiiBlocker must always mask. The Allowlist lists terms PiiBlocker must leave alone. If a real name got through, add it to the dictionary. If a harmless term keeps being flagged, add it to the Allowlist.