Fixing a name or field that was read wrong

OCR misread a surname, swapped the address for the name, or left a field as "unknown". Support can't edit it for you, but you have three ways to fix it - the user data review step, the update-data API, and a resubmission of the ID step.

Written by Didit Support · Updated 15 September 2026

Short answer

Didit support cannot edit the values on your session. You can, three ways: let the person confirm and correct the data in the flow (User data review), patch the fields yourself with the update-data API, or request a resubmission of the ID step after tightening the image quality bar. Nearly every misread comes from a poor capture.

A name with one letter wrong, a surname that "did not load", a session whose name shows as unknown, or an address that landed in the name field are all the same problem: the extraction ran on an image that was not good enough, or on a document side the model does not read. The data in the session is what the model read, and Didit does not change customer data by hand: your team owns the record, and any change has to be traceable to you.

#First, understand why it happened

Open the session and look at the document images and the warnings:

  • Blurry, dark, glared or partially cropped image. The most common cause by far. The session usually carries a low image quality score or an IMAGE_TOO_BLURRY warning. See fixing document upload and reading problems.
  • Uploaded file instead of a live capture. Uploads are often screenshots or compressed photos and misread far more often than camera scans.
  • A non-Latin script. Names in Cyrillic, Arabic, Greek, Chinese or Korean can come back scrambled or empty if the workflow's Character format setting doesn't match what you expect (original script versus Latin transliteration).
  • A document side the model doesn't read yet. Some national IDs carry the data on the back in a layout Didit has not been trained on. In that case the front is read and the back is accepted but not extracted; send a specimen to support so the mapping can be added.

#Three ways to fix it

  1. Let the person correct it in the flow (best for new sessions)

    In the workflow's ID verification step, under Advanced, turn on User data review: after the capture, the person sees the extracted fields and can correct reading mistakes before the result is decided. Edits to the Critical fields you list (name, date of birth, document number) take the action you choose - review or decline - so a correction is never silent. This fixes the problem at the source and needs no work from your team.

  2. Patch the session yourself (for an existing session)

    Call PATCH /v3/session/{session_id}/update-data/ with only the fields you want to change: first_name, last_name, date_of_birth, document_number, expiration_date, issuing_state and the other extracted fields. full_name is recomputed for you. The session must be Approved, Declined, In Review or KYC Expired, the change fires a data.updated webhook, and the audit trail records which fields changed and who changed them. It is a paid call. There are matching endpoints for proof-of-address fields and for KYB registry data. Reference: update session data.

  3. Ask for a new capture (when the image is the problem)

    If the image is too poor to trust any of the fields, request a resubmission of the ID verification step: the person receives the same link and repeats only that step. Before you do, raise the Minimum image quality slider in the step and consider enabling the Image capture review screen, so the retake can't be as bad as the first attempt. See letting someone try a verification again.

Note

Resubmitting the ID step replaces the document images and re-runs the checks that depend on them (face match, AML name screening, database validation) at their usual price. Patching a field with the API does not re-run anything: if the wrong name already produced an AML or registry result, run that check again with the corrected data - the standalone APIs exist for exactly this.

#What patching does not do

  • It does not change the status of the session. A declined session with a corrected name is still declined until your team decides otherwise.
  • It does not re-verify the document. The correction is your assertion, recorded as such in the audit log.
  • It does not apply retroactively to other sessions of the same person.

#Stopping it from recurring

  • Raise Minimum image quality in the ID verification step and turn on the Image capture review screen so people confirm the photo before it is uploaded.
  • Prefer camera scan over file upload for production traffic.
  • Set Character format deliberately for markets with non-Latin scripts.
  • If one document type misreads consistently, send two or three anonymised specimens to support so the model can be updated.