Guide
How to write alt text for PDF images
The decision about whether an image is informative comes before the writing. Then: what to say, what to leave out, and where AI helps.
About 8 minutes · quick
Decide first, write second
Every image in a document is one of three things, and getting this wrong matters more than the wording.
Informative — carries meaning the text does not. Figure element, /Alt description.
Decorative — a rule, a background wash, a flourish, a stock photo of people at a whiteboard. Artifact, not a Figure. Artifacting removes it from the structure tree so the reader never meets it, which is exactly right.
Text as an image — a drop cap, a wordmark, a scanned signature. /ActualText, so the characters flow into the sentence they belong to.
Describing decoration wastes the reader’s attention. Artifacting a chart deletes it from the document for anyone who cannot see it. Both are worse than clumsy wording.
Writing it
Say what it contributes, not what it depicts. Not “a line chart showing quarterly revenue”. Rather: “Revenue grew steadily from Q1 to Q3, then fell 12% in Q4.”
Do not start with “image of”. The role is already announced by the tag. Those three words are wasted on every image in the document.
Do not repeat the caption. If a <Caption> sits with the figure, the reader gets both. Duplicate text is announced twice.
Match the register of the document. Alt text in a legal document should sound like the legal document.
Doing it free
Acrobat: Accessibility → Set Alternate Text walks every Figure in the document with a Decorative figure checkbox that artifacts properly. This is the fastest free route and it handles the decorative case correctly.
Its limit is that it only finds images already tagged as Figure. An image that was never tagged does not appear, and you will not know it is missing. That condition →
Upstream: Word, PowerPoint and InDesign all carry alt text into the export, and Word’s Mark as decorative checkbox exports as an artifact. Doing it there means it survives every revision.
Where AI helps, and where it does not
It writes a competent first draft of what an image depicts. It does not know what the image contributes to your document, and a description that is fluent and wrong is worse than none, because nobody reviewing can tell.
Taggart will generate drafts in three modes chosen at setup and shown persistently in the interface: a local model where image bytes never leave the machine, your own provider key with explicit per-document consent, or off. In every mode the draft goes into the review queue and your decision — approved, edited, rejected, or “this is decorative” — is recorded in the remediation report with a timestamp.
Whether an image is informative or decorative is a judgement about the document’s purpose. No model has that context, and pretending otherwise is how documents end up with confident descriptions of things that did not need describing.
Frequently asked questions
Can I leave Alt empty for decoration?
No. In HTML alt="" marks an image decorative. In PDF there is no equivalent — an empty /Alt is a Figure that promises a description and gives none, and it is a failure. Decorative images must be artifacts instead.
How long should it be?
Long enough to convey what the image contributes. A photograph usually needs a sentence. A chart needs whatever conveys the finding — and if that runs to a paragraph, the data probably belongs in a table.
Should I describe a chart or give the data?
Give the finding in the alt text and the data in a table. 'Revenue grew steadily to Q3, then fell 12% in Q4' is the finding. Twenty numbers read aloud is not a description, it is a table being announced badly.
Related failure conditions
- 13-004 Figure tag alternative or replacement text missing A Figure element has neither /Alt nor /ActualText. Every Figure must have one or the other; if the image is decorative it should be an artifact instead of a Figure.
- 13-001 Graphics objects are not tagged with a Figure tag An image or vector drawing that carries meaning is not inside a Figure element. Either it is untagged, or it has been swept into a paragraph.
- 13-005 ActualText used for a Figure for which alternative text is more appropriate ActualText replaces the content for text-extraction purposes; Alt describes it. A photograph needs Alt. A drop cap or a glyph-as-image needs ActualText.
- 13-008 ActualText not present when a Figure is intended to be consumed primarily as text A drop cap, a logo that is a word, an image of a signature block — content that is text — needs ActualText so it flows into the text stream.
Check your own file. Taggart's validator is free and unlimited — no page cap, no watermark, no account.
Last updated .