Treat publisher-written ALT text, lettering you can read in the image itself, and OCR output as three separate records with different origins and limits. ALT text is the uploader's accessibility wording, so it describes what the publisher chose to say; visible lettering is whatever is painted, printed, or rendered in the image; OCR is a mechanical transcription that may misread, miss, or reorder characters. Log each in its own column and only let a column support a claim it can actually carry. None of the three identifies a person on its own.
Find public social profile clues with our TraceSeek app.
Start by naming which source you are reading
Before you write anything down, decide which of the three you are looking at. ALT text is entered by the person who uploaded the image through the image-description interface, and viewers read it by opening the ALT badge; it is accessibility context, not a camera record or identity check. Visible lettering is part of the pixels you can see. OCR is a derived mechanical transcription of those marks. Keeping the labels apart from the first moment prevents a later reader from assuming one source confirmed another. [X: user-entered image descriptions]
ALT text is a publisher statement, not a transcript
When the uploader writes an image description, that wording is their choice. It may be generous, terse, atmospheric, or simply wrong about a detail, and that is still the ALT text as published. It should be cited as the publisher's accessibility description, never quoted as if the image itself proved it. A descriptive name in ALT text — a name, a role, a relationship — is a claim the publisher made, not an independently identified person. Separately note that the badge is opened by the viewer, so what an external observer sees matches what the uploader typed, not an external verification. [X: user-entered image descriptions]
Visible lettering and OCR have different failure modes
Lettering in the image can be read by a person and may be crisp, stylised, cropped, or partly obscured. OCR turns those marks into searchable characters; it can misread similar glyphs, drop punctuation, or lose the line breaks that carry meaning. Neither route tells you who wrote the lettering, whether the account that posted it owns the words, or whether the depicted person is the author. Record OCR output as a transcription with its spelling quirks preserved, and note visible lettering separately when you can read it directly. [X: user-entered image descriptions]
A three-column source record for each public artifact
Use one row per artifact and keep the columns strictly parallel. The purpose is to make the gap between description and evidence visible to a later reader, including you. If a column is empty because the image has no ALT text or no legible lettering, say so rather than backfilling from another column. Do not merge columns into a single narrative sentence later.
| observation | limit | next step |
|---|---|---|
| Uploader's ALT text quoting a name and a setting | Publisher wording; not OCR, camera record, or identity check | Quote it as an accessibility description; look for a separate public statement before linking a person |
| OCR transcription of a sign in the frame | Mechanical reading; may misread characters or lose layout | Keep the raw transcription; verify wording against the visible lettering |
| Visible lettering read directly from the image | Shows marks, not authorship or intent | Note who published the page; do not attribute the words to a depicted person |
| No ALT text present | Description absent; cannot be inferred from pixels | Record the absence; stop the description-based inference |
Why descriptive names must not become identities
A name in ALT text, in lettering, or in an OCR line is a string, not a verified person. It can match a username, a business, a character, or nothing at all. Matching a photo to a name is exactly the kind of inference that the separate-source record is designed to block. If you want to link a name to an account, look for an explicit public statement on a public page tying the two together, and if none exists, write the connection as unresolved rather than implied. [X: user-entered image descriptions]
When the source is absent, private, or ambiguous
If the image has no ALT text, the lettering is unreadable, the OCR is garbled, the page is private, or the wording is genuinely ambiguous, write that exact limit and stop. Do not fill the gap with a plausible guess, a similar-looking image, or a request for private access. Record the empty column, the reason, and the pending question, then close the note. A precise limit with no inference is a complete finding; an invented bridge is not. A hypothetical example: an ALT line that says only 'a person outdoors' supports nothing about who that person is, and should be logged that way.
A couple more questions
If the ALT text names a person, can I treat that as identification?
No. ALT text is wording the uploader chose to enter through the image-description interface, so it is a publisher claim, not an independent check. Record it as the publisher's description, and only link it to a person if a separate public page explicitly does so.
Should I merge OCR output with ALT text to make one cleaner description?
No. Merging hides which source carried which claim, and OCR can misread characters while ALT can be vague or absent. Keep them in separate columns so a reader can see exactly what each one supports.
What do I do when the ALT text, the visible lettering, and the OCR disagree?
Record all three as they stand, name the point of disagreement, and stop before drawing any conclusion about a person, an account, or an event. A disagreement is a finding about the sources, not a licence to pick the most convenient one.
Does a clear image mean I can skip the source labels?
No. Clarity of the image does not change who wrote the ALT text or what OCR may have missed, so the labels stay attached. Clarity reduces one kind of doubt without converting description into evidence.
Sources & editorial notes
Official references checked on October 24, 2026. Checklists, comparison tables and example wording are DateClues editorial tools, not official tests or validated risk scores.
General information. No guide or tool can certify a person’s identity, intentions or safety. Read our editorial standards.