Redacting Charts, Maps, and Diagrams in a PDF

Most redaction advice is written for text. Find the words, cover the words, make sure the covering is permanent and that nothing survives in a hidden layer, metadata field, or comment. That model works because text is discrete: a social security number is in one place, it is nine digits long, and once the pixels and the underlying characters are gone, it is gone.

Figures break that model. A chart, a map, a floor plan, or an org chart is a drawing whose positions carry the information. The printed numbers on a chart are a courtesy — a label repeating what the geometry already shows. Remove the courtesy and the geometry remains, sitting against an axis or a scale that still tells a reader exactly what to measure it against. Nothing is broken about the file and nothing is recoverable from underneath the box. The document simply still contains the answer, drawn rather than written.

This is a different failure from the one on most of this site's pages, and it is also different from recomputing a covered number from a visible total. Arithmetic recovery needs the document to do sums; graphical recovery needs nothing but a ruler and a labelled axis. It shows up in exactly the documents people are most careful with: financial reports with trend charts, HR packs with headcount graphs, medical records with growth curves and lab trend plots, property documents with surveys and site plans, and safety filings with a map of where an incident happened.

Why a figure is harder to redact than a sentence

A chart is a coordinate system with marks plotted in it. That gives a reader three ingredients, and they only need two of them to recover a value: a reference frame (the axis, its range, its tick marks, the gridlines, a scale bar), a plotted mark whose position they can measure, and a label. Redaction almost always removes only the third.

How precisely a value can be recovered is a property of the chart, not of the reader's determination. A tall bar chart with a narrow axis range, labelled ticks, and fine gridlines can be read closely. A thumbnail-sized sparkline on a compressed axis gives a rough reading at best. The useful question is not "could someone get the exact number?" but "is the range they could get still sensitive?" For a salary, a headcount in a small team, a settlement amount, a clinical value, or a home location, a narrow range is frequently just as disclosing as the exact figure — which is the same reason a tightly fitted redaction box is itself a hint about what sits underneath it.

Two more properties make figures leakier than they look. First, charts are usually repeated: the same series appears in the body, in a summary chart on the cover page, and sometimes as a small multiple in an appendix. Redacting the figure you were worried about while leaving its twin three pages earlier is a very common outcome. Second, figures are frequently generated from the same data as the table you redacted carefully, so a chart can quietly restore a value you already dealt with properly elsewhere.

What actually carries the value, by figure type

Before you draw a single box, work out which part of the drawing is the carrier. It is rarely the part that looks like data.

• Bar and column charts. Bar length against the axis. Also the axis range itself: a chart whose axis starts at a non-zero value and is labelled at two points lets a reader reconstruct the scale completely. Stacked bars leak each segment separately, and a covered segment can be recovered from the full bar height minus the visible segments.
• Line charts and trend plots. Vertex positions, and the shape between them. Covering a single point on a line leaves the line segments entering and leaving it, which pin the covered vertex down to a small region. Growth curves and lab trend plots in medical records behave this way.
• Pie and donut charts. Slice angles. Angles are percentages, so covering the percentage labels changes nothing material. A covered slice is also recoverable as the remainder of the circle.
• Scatter plots and dot plots. Every individual point is a record. Covering the outlier you were worried about leaves the rest of the distribution, and point counts can disclose how many people or cases are in a group — a serious problem when a group is small.
• Maps and location graphics. The pin, the highlighted parcel, the hatched area, the route line. These are coordinates drawn rather than written, and covering the printed address beneath the map does not move the pin. Watch also for the inset locator map, the scale bar, street and landmark labels, the north arrow, and the grid reference in the margin. For shelter addresses, custody arrangements, or protective-order documents, this is the single highest-stakes case on this page.
• Floor plans and surveys. Dimension lines, room labels, the title block, the legal description, the surveyor's seal, and the revision table — the drawing frame often holds more identifying detail than the drawing.
• Org charts and network diagrams. Box positions and connecting lines encode reporting structure and group sizes. Cover every name and the shape still shows who reports to whom and how many people sit in each team; in a small organisation the shape alone can identify individuals.
• Heat maps, hatching, and shaded categories. The fill pattern is the value and the legend merely names it. Covering the legend leaves the pattern, and categories are usually inferable from the surrounding text.
• Gauges, progress bars, and status icons. Small graphics that encode a single number or state, easy to miss precisely because they do not look like charts.

A workflow for redacting a figure

1. Decide whether to redact the figure at all, or regenerate it. If you can still reach whatever produced the chart or map, rebuilding it from bucketed, rounded, or aggregated data — or redrawing the map at a coarser zoom with no pin — gives you a figure that is both readable and non-disclosing. A redacted figure is often neither. Redaction in the PDF is the right answer when the source is gone, the document is signed or certified, the figure came from a third party, or the page must stay otherwise identical.
2. Name the carrier. For each figure, write down which element carries the sensitive value using the list above. If you cannot say which element it is, you are not ready to draw a box.
3. Cover the carrier, not its caption. In practice this usually means covering the whole plot area, or the whole figure including its frame. Partial redaction inside a plot area is only safe when the remaining reference marks genuinely do not locate the covered mark, which is rarer than it feels.
4. Sweep the frame. Axis labels and ranges, the legend, the caption, the figure title, data labels, footnotes about the source or sample size, and the page header. Captions and source notes are a frequent leak: "Monthly totals for the Riverside branch, n = 4" can be more identifying than the chart.
5. Find the figure's duplicates. Check the cover page, the executive summary, any dashboard page, the appendix, and any small-multiple panel. Search the document for the same shape, not the same words.
6. Reconcile with the tables. If you also redacted numbers in a table, confirm the chart does not restore them, and that the chart's own values are not recoverable from a visible total. The two checks are separate and both are needed.
7. Read the result as a stranger would. Open the exported file, look only at what is visible, and try to estimate the value you covered. If you can bracket it, so can your recipient.

What this tool does and does not do here

HidePDF draws manual rectangles and nothing else. It does not parse, interpret, or recognise anything in the document, and it has no concept of a figure, an axis, a series, a pin, or a legend. It cannot find the charts for you, it cannot tell you that a covered bar is still measurable against a visible axis, and it will not stop you exporting a figure that still discloses the value. Every judgement above is a human step taken before you click download.

What the tool handles is the mechanical part, and for figures specifically there is a useful side effect. On export it rebuilds every page of the document as an image at a fixed higher resolution and burns your boxes into those images. That is what makes the covered pixels genuinely unrecoverable rather than hidden beneath a shape that can be moved or deleted — and because the rebuild applies to the whole page, it also discards the vector drawing instructions behind a chart, including whatever exact coordinates were in them, rather than leaving a precise vector figure sitting underneath an opaque rectangle. The processing all happens on your own device in the browser, with no upload required. The trade-off is that the exported document is an image throughout, so no part of it stays selectable or searchable text — decide before exporting whether your recipient needs to search the file or copy from it.

Common mistakes and misconceptions

Covering the data labels and the legend. This is the single most common version of the mistake, and it is seductive because labels are the part that looks like data. The geometry is the data; labels are a reading aid for people who do not want to measure.

Blurring or pixelating a figure instead of covering it. A blur preserves large-scale structure by design — that is what makes a blurred face still recognisable as a face and a blurred chart still legible as a trend. For a bar chart, the bar tops usually survive a blur well enough to measure.

Assuming a small or low-resolution figure is unreadable. Small reduces precision; it does not remove information. A figure that is too coarse to read exactly may still place a value in a range you did not want to disclose.

Treating a map as a picture rather than as coordinates. People who would never leave an address in text will leave the pin that marks it, because a map does not read like personal data on the page.

Forgetting the figure's duplicate on the cover page. Summary pages and dashboards are generated from the same data and sit before the reader reaches the page you were careful with.

Assuming the chart's source data must have been removed with it. Covering a chart on the page says nothing about what else the file carries. A generator may have attached the source workbook, written field names into the document's structure, or left a descriptive title in the metadata, and those are separate problems from the drawing.

Over-redacting the entire figure out of caution. The opposite failure is real too. If the recipient asked for the report to see the trend, blacking out every chart usually means the document comes back — and the second request is rarely redacted as carefully as the first. Where a rule, an order, or a records procedure governs what must be disclosed, that rule decides what stays, not convenience.

Checking the redaction technically and calling it verified. Confirming that nothing can be copied from under the box is a worthwhile check, and a figure leak will pass it cleanly. There is nothing technically wrong with a document that merely draws its secret.

Related guides

Explore more ways to redact PDFs privately, or use the redaction tool above:

• Redacting Numbers in a PDF: When the Totals Give Them Away
• How to Redact a Photo or Image Embedded Inside a PDF

Frequently asked questions