OCR extracted the wrong information from my document

OCR is accurate in most cases but can make mistakes, especially on hand-written text, low-resolution scans, or documents with unusual formatting. Here is how to handle incorrect extractions.

What can go wrong?

• A name is extracted with a spelling error due to a blurry scan
• A date is read incorrectly (for example, "2023" read as "2028")
• An amount is misread due to a smudged digit
• The wrong field is extracted as the primary name (for example, an account holder name vs a joint holder)

How to fix it

1. Re-upload a better quality scan

The most reliable fix is to upload a clearer version of the document. Follow the quality tips in the Document quality tips article. A sharper scan gives the OCR engine less room to make mistakes.

2. Mark the Signal as resolved manually

If the OCR error is causing a false Signal and you are confident the underlying document is correct, you can manually mark the Signal as resolved. Go to the Signal in your case, review the description, and click the resolve checkbox. Include a note if needed.

3. Ignore low-severity false Signals

If the false Signal is rated as Low severity and you have confirmed the documents are actually consistent, resolving it manually is the right approach. It will remove the point penalty from your score.

What if the Signal is accurate despite the OCR error?

Sometimes the OCR extracts an error that reveals a real inconsistency in your documents. For example, if your name is spelled differently on your bank account than it is on your passport, the signal is accurate even if the OCR confidence is low. In this case, obtain a corrected document from the issuing institution.

Reporting a persistent OCR issue

If a specific document type consistently fails OCR or produces incorrect results, please email support@migraiq.com with the document type (not the file itself) and a description of the error. We use this feedback to improve extraction accuracy.