{"course":{"slug":"choosing-your-tools","title":"Choosing and Using the Tools","url":"https://addaly.com/learn/choosing-your-tools","level":"beginner","track":"work"},"lesson":{"slug":"redaction-that-works","title":"Redaction that actually works","url":"https://addaly.com/learn/choosing-your-tools/redaction-that-works","position":53,"of":76,"minutes":8,"answer":"Replacing identities with roles preserves everything the model needs and removes most of the risk, but files carry authorship metadata, hidden rows, extractable text under black rectangles and location data.","keyIdea":"Replacing identities with roles preserves everything the model needs and removes most of the risk, but files carry authorship metadata, hidden rows, extractable text under black rectangles and location data, and removing a name column is not anonymisation."},"body":{"format":"markdown","text":"## Replace the role, not the letters\n\nThe technique that does most of the work costs nothing and takes no time: write the role instead of the identity.\n\n- \"Mrs Adeyemi has missed three payments\" becomes \"the tenant has missed three payments\".\n- \"Acme Ltd's contract with Northern Supplies\" becomes \"our client's contract with their supplier\".\n- \"Patient DOB 12/03/1978, hypertension\" becomes \"a patient in their late forties with hypertension\".\n\nThe answer you get back is the same. The model was never using the name for anything; it needs the relationship, the sequence and the numbers. Once this becomes a habit rather than a step, most of the confidentiality problem in most people's work disappears, because the sensitive material is not being sent at all.\n\n## When you need consistency\n\nFor longer documents where several people appear, replace consistently. Person A, Person B, Company X. Keep a small mapping in a file you do not send, and substitute back afterwards. This preserves the structure the model needs while removing the identities.\n\nA find-and-replace in any editor does this in a minute. Free tools that automate it exist: **Microsoft Presidio** is an open-source library that detects and replaces names, numbers and identifiers in text, and spaCy's named-entity recognition does a rougher version in three lines of Python. For occasional work, doing it by hand is faster than installing anything.\n\n## The parts of a file you did not mean to send\n\nThis is where careful people still get caught, because the sensitive material is not in the text.\n\n**Document metadata.** Word and PDF files carry author names, organisation, editing history and sometimes tracked changes and comments that are not visible in the document. Use the built-in inspector (*File, Info, Check for Issues, Inspect Document* in Word) before sending anything outward.\n\n**Spreadsheet hidden content.** Hidden rows, hidden sheets, filtered-out data, and the source data behind a chart all travel with the file. Deleting the visible rows is not enough.\n\n**PDF layers.** A black rectangle drawn over text hides it from the eye and not from a text extractor. This has produced published court filings where the redacted names could be copied out with a mouse. Proper redaction removes the text; the free **qpdf** and most PDF editors can flatten, and checking with `pdftotext` afterwards takes ten seconds.\n\n**Image metadata.** Photographs carry EXIF data including GPS coordinates and the device. Free tools such as **ExifTool** strip it; most phones can be set not to record location in the first place.\n\n**Screenshots.** The most common leak by far. A screenshot of one message includes the browser tabs, the notification that arrived, the file names in the sidebar and the account you are logged into. Crop before sending, always.\n\n## Why \"anonymised\" is harder than it sounds\n\nRemoving names does not always make data anonymous. A well-known line of research has shown that a surprisingly small number of ordinary attributes (a postcode, a date of birth, a sex, or a handful of location points) is often enough to identify a specific person within a large population. Combining a de-identified dataset with a public one is a documented re-identification technique.\n\nFor a single document with one person in it, role substitution is effective. For a dataset of thousands of records, removing the name column is not anonymisation, and treating it as such is a mistake with regulatory consequences. If you are handling data at that scale, the question needs somebody who does this properly.\n\n## What redaction costs you\n\nBe honest that it is not free. Removing detail sometimes removes information the model would have used: a name can carry a gender, a nationality or a professional register that changes how a letter should be written, and a company name lets a model apply what it knows about that industry.\n\nWhere that matters, put the useful attribute back without the identity (\"a supplier in the garment trade\", \"a female patient in her sixties\") which supplies the context and not the person. And where the identity is the question, such as research about a named public figure, redaction is not applicable and the decision is whether to send it at all.\n\n> Figure: Hidden from the eye, or actually removed. Looks redacted: A black rectangle drawn over text in a PDF; Rows hidden rather than deleted in a spreadsheet; A screenshot showing tabs, filenames and a notification; A photograph with the location still in its metadata; A document carrying author name and tracked changes. Actually removed: Text deleted from the file, then checked with pdftotext; A new file holding only the rows you meant to send; Cropped before it leaves the screen; Metadata stripped with ExifTool, or never recorded; Inspected with the office suite's own document inspector. Role substitution does most of the work and costs nothing, because the model never needed the name: only the relationship, the sequence and the numbers. At the scale of a dataset rather than a document, removing the name column is not anonymisation and the question needs somebody who does it properly.\n\n## The rule that survives everything\n\nBefore pasting, ask one question: **could I email this to a stranger?**\n\nIf the honest answer is no, either redact it or use a local model. That single question, asked reliably, prevents more problems than any technical measure in this course, because it is applied at the only moment when prevention is still possible."},"check":{"question":"A lawyer covers names in a PDF with black rectangles before uploading it. Why is this insufficient?","options":[{"text":"The rectangles are stored as a separate annotation layer that a viewer can toggle off with a keyboard shortcut.","correct":false,"tempting":"Describes one specific mechanism rather than the general problem, and the text remains extractable whether or not annotations can be hidden."},{"text":"Compression when the file is uploaded can degrade the rectangles enough for the text beneath to show through.","correct":false,"tempting":"Imagines a visual artefact, when the failure is that the characters themselves were never removed from the file."},{"text":"Drawing over text does not remove it, so the characters are still extractable from the file.","correct":true},{"text":"The rectangles remove the text from view but leave it in the document's metadata, where it can be recovered.","correct":false,"tempting":"Relocates the text to metadata; it stays in the page's own content stream, which is why plain extraction recovers it."}]},"prev":{"slug":"who-owns-the-output","title":"Who owns what it produces","url":"https://addaly.com/learn/choosing-your-tools/who-owns-the-output"},"next":{"slug":"secrets-and-what-not-to-paste","title":"Secrets, and what not to paste","url":"https://addaly.com/learn/choosing-your-tools/secrets-and-what-not-to-paste"},"authorship":"Founder and series editor: Dr Mustafa Mun. Written by AI models to a written standard. Exam keys checked blind by a second model: 3,665 of 3,666 agreed. Academic review invited.","review":{"status":"Draft, open for academic review"},"cite":"Addaly (2026) 'Redaction that actually works', in Choosing and Using the Tools. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/choosing-your-tools/redaction-that-works (Accessed: 13 October 2026).","licence":{"name":"CC BY-NC-SA 4.0","url":"https://creativecommons.org/licenses/by-nc-sa/4.0/"},"reader":{"schema":1,"title":"Redaction that actually works","courseTitle":"Choosing and Using the Tools","module":{"index":6,"slug":"data-terms-and-law","title":"Data, terms and the law"},"answer":null,"keyIdea":"Replacing identities with roles preserves everything the model needs and removes most of the risk, but files carry authorship metadata, hidden rows, extractable text under black rectangles and location data, and removing a name column is not anonymisation.","hero":{"kind":"figure","figure":{"t":"figure","label":"Figure","figure":{"left":{"title":"Looks redacted","items":["A black rectangle drawn over text in a PDF","Rows hidden rather than deleted in a spreadsheet","A screenshot showing tabs, filenames and a notification","A photograph with the location still in its metadata","A document carrying author name and tracked changes"]},"type":"compare","right":{"title":"Actually removed","items":["Text deleted from the file, then checked with pdftotext","A new file holding only the rows you meant to send","Cropped before it leaves the screen","Metadata stripped with ExifTool, or never recorded","Inspected with the office suite’s own document inspector"]},"title":"Hidden from the eye, or actually removed","caption":"Role substitution does most of the work and costs nothing, because the model never needed the name: only the relationship, the sequence and the numbers. At the scale of a dataset rather than a document, removing the name column is not anonymisation and the question needs somebody who does it properly."},"alt":"Figure: Hidden from the eye, or actually removed. Looks redacted: A black rectangle drawn over text in a PDF; Rows hidden rather than deleted in a spreadsheet; A screenshot showing tabs, filenames and a notification; A photograph with the location still in its metadata; A document carrying author name and tracked changes. Actually removed: Text deleted from the file, then checked with pdftotext; A new file holding only the rows you meant to send; Cropped before it leaves the screen; Metadata stripped with ExifTool, or never recorded; Inspected with the office suite’s own document inspector. Role substitution does most of the work and costs nothing, because the model never needed the name: only the relationship, the sequence and the numbers. At the scale of a dataset rather than a document, removing the name column is not anonymisation and the question needs somebody who does it properly."}},"blocks":[{"t":"h2","id":"replace-the-role-not-the-letters","runs":[{"text":"Replace the role, not the letters"}]},{"t":"p","runs":[{"text":"The technique that does most of the work costs nothing and takes no time: write the role instead of the identity."}]},{"t":"ul","items":[[{"text":"“Mrs Adeyemi has missed three payments” becomes “the tenant has missed three payments”."}],[{"text":"“Acme Ltd’s contract with Northern Supplies” becomes “our client’s contract with their supplier”."}],[{"text":"“Patient DOB 12/03/1978, hypertension” becomes “a patient in their late forties with hypertension”."}]]},{"t":"p","runs":[{"text":"The answer you get back is the same. The model was never using the name for anything; it needs the relationship, the sequence and the numbers. Once this becomes a habit rather than a step, most of the confidentiality problem in most people’s work disappears, because the sensitive material is not being sent at all."}]},{"t":"h2","id":"when-you-need-consistency","runs":[{"text":"When you need consistency"}]},{"t":"p","runs":[{"text":"For longer documents where several people appear, replace consistently. Person A, Person B, Company X. Keep a small mapping in a file you do not send, and substitute back afterwards. This preserves the structure the model needs while removing the identities."}]},{"t":"p","runs":[{"text":"A find-and-replace in any editor does this in a minute. Free tools that automate it exist: "},{"text":"Microsoft Presidio","b":true},{"text":" is an open-source library that detects and replaces names, numbers and identifiers in text, and spaCy’s named-entity recognition does a rougher version in three lines of Python. For occasional work, doing it by hand is faster than installing anything."}]},{"t":"h2","id":"the-parts-of-a-file-you-did-not-mean-to-send","runs":[{"text":"The parts of a file you did not mean to send"}]},{"t":"p","runs":[{"text":"This is where careful people still get caught, because the sensitive material is not in the text."}]},{"t":"p","runs":[{"text":"Document metadata.","b":true},{"text":" Word and PDF files carry author names, organisation, editing history and sometimes tracked changes and comments that are not visible in the document. Use the built-in inspector ("},{"text":"File, Info, Check for Issues, Inspect Document","i":true},{"text":" in Word) before sending anything outward."}]},{"t":"p","runs":[{"text":"Spreadsheet hidden content.","b":true},{"text":" Hidden rows, hidden sheets, filtered-out data, and the source data behind a chart all travel with the file. Deleting the visible rows is not enough."}]},{"t":"p","runs":[{"text":"PDF layers.","b":true},{"text":" A black rectangle drawn over text hides it from the eye and not from a text extractor. This has produced published court filings where the redacted names could be copied out with a mouse. Proper redaction removes the text; the free "},{"text":"qpdf","b":true},{"text":" and most PDF editors can flatten, and checking with "},{"text":"pdftotext","code":true},{"text":" afterwards takes ten seconds."}]},{"t":"p","runs":[{"text":"Image metadata.","b":true},{"text":" Photographs carry EXIF data including GPS coordinates and the device. Free tools such as "},{"text":"ExifTool","b":true},{"text":" strip it; most phones can be set not to record location in the first place."}]},{"t":"p","runs":[{"text":"Screenshots.","b":true},{"text":" The most common leak by far. A screenshot of one message includes the browser tabs, the notification that arrived, the file names in the sidebar and the account you are logged into. Crop before sending, always."}]},{"t":"h2","id":"why-anonymised-is-harder-than-it-sounds","runs":[{"text":"Why “anonymised” is harder than it sounds"}]},{"t":"p","runs":[{"text":"Removing names does not always make data anonymous. A well-known line of research has shown that a surprisingly small number of ordinary attributes (a postcode, a date of birth, a sex, or a handful of location points) is often enough to identify a specific person within a large population. Combining a de-identified dataset with a public one is a documented re-identification technique."}]},{"t":"p","runs":[{"text":"For a single document with one person in it, role substitution is effective. For a dataset of thousands of records, removing the name column is not anonymisation, and treating it as such is a mistake with regulatory consequences. If you are handling data at that scale, the question needs somebody who does this properly."}]},{"t":"h2","id":"what-redaction-costs-you","runs":[{"text":"What redaction costs you"}]},{"t":"p","runs":[{"text":"Be honest that it is not free. Removing detail sometimes removes information the model would have used: a name can carry a gender, a nationality or a professional register that changes how a letter should be written, and a company name lets a model apply what it knows about that industry."}]},{"t":"p","runs":[{"text":"Where that matters, put the useful attribute back without the identity (“a supplier in the garment trade”, “a female patient in her sixties”) which supplies the context and not the person. And where the identity is the question, such as research about a named public figure, redaction is not applicable and the decision is whether to send it at all."}]},{"t":"h2","id":"the-rule-that-survives-everything","runs":[{"text":"The rule that survives everything"}]},{"t":"p","runs":[{"text":"Before pasting, ask one question: "},{"text":"could I email this to a stranger?","b":true}]},{"t":"p","runs":[{"text":"If the honest answer is no, either redact it or use a local model. That single question, asked reliably, prevents more problems than any technical measure in this course, because it is applied at the only moment when prevention is still possible."}]}],"check":{"question":[{"text":"A lawyer covers names in a PDF with black rectangles before uploading it. Why is this insufficient?"}],"options":[{"text":[{"text":"The rectangles are stored as a separate annotation layer that a viewer can toggle off with a keyboard shortcut."}],"correct":false,"tempting":[{"text":"Describes one specific mechanism rather than the general problem, and the text remains extractable whether or not annotations can be hidden."}]},{"text":[{"text":"Compression when the file is uploaded can degrade the rectangles enough for the text beneath to show through."}],"correct":false,"tempting":[{"text":"Imagines a visual artefact, when the failure is that the characters themselves were never removed from the file."}]},{"text":[{"text":"Drawing over text does not remove it, so the characters are still extractable from the file."}],"correct":true,"tempting":null},{"text":[{"text":"The rectangles remove the text from view but leave it in the document’s metadata, where it can be recovered."}],"correct":false,"tempting":[{"text":"Relocates the text to metadata; it stays in the page’s own content stream, which is why plain extraction recovers it."}]}],"order":[3,0,2,1],"why":null},"prev":{"slug":"who-owns-the-output","title":"Who owns what it produces","minutes":8},"next":{"slug":"secrets-and-what-not-to-paste","title":"Secrets, and what not to paste","minutes":7}}}