AI document verification for client uploads

AI document verification checks each client upload against what the firm asked for, before a person opens it. DocAsker checks image quality, document type, dates and periods, signatures and page count the moment a file arrives. A file that fails goes back to the client with the reason. This page shows what is checked, the limits, and our own test results.

By the DocAsker team5 min read
A reviewer checking client documents at a desk

In short

  • The checks answer one question: is this the document the firm asked for, readable and complete?
  • A failed check goes back to the client with the reason. An uncertain one goes to a person at the firm.
  • In our test of 19 prepared documents, none of the 13 flawed files was accepted.

What does AI document verification check?

DocAsker runs five checks on an upload. Each checklist item says which of them apply, so a passport copy and a signed engagement letter are held to different rules.

  • Image quality: blur, glare, and cropped or cut-off edges.
  • Document type: the file is the document that was asked for. A utility bill cannot stand in for a bank statement.
  • Dates and periods: statement periods, issue dates and expiry dates, compared with the request. For example, a passport that must be valid for 6 more months.
  • Signatures: the signature field is filled in.
  • Completeness: a multi-page document arrives with every page.

What happens when a check fails?

The client sees the reason in plain words on the upload page, for example that the statement covers April to June and the request is for January to March. They can upload a better file straight away, while the document is still in front of them.

A check has three outcomes: passed, failed or needs review. When the file cannot be judged with confidence, for example a date that is not visible on a photo, the upload is flagged for a person at the firm instead of being accepted or refused.

The firm's team sees the files that passed, plus anything flagged for review. Nobody has to open every upload to find the unusable ones.

What happened in our test of 19 documents

In November 2026 we ran 19 prepared test documents through the same check that runs on a real upload: 16 in English and 3 in Turkish, as PDFs, scanned PDFs and phone-style photos. Thirteen had a flaw planted on purpose. Six were clean.

None of the 13 flawed documents was accepted. Twelve went back for a new upload with the reason, and one, a photo with no visible issue date, was sent to a person for review, which is the intended result for that case. Of the 6 clean documents, 5 were accepted. One complete three-page PDF was sent to a person for review instead of being accepted.

This is a small test on documents we made ourselves, run once. It contained no client records. It shows how the checks behave on clear cases. It is not a measured accuracy rate for every document a firm will receive.

  • Wrong document: a purchase order uploaded where an invoice was requested. Sent back.
  • Out of date: an invoice issued more than 30 days before the request allowed. Sent back.
  • No signature: an unsigned agreement as a scan and as a photo, in English and in Turkish. Sent back each time.
  • Missing pages: page 1 of a 3-page document. Sent back.
  • Unreadable photo: one blurred, one with glare, one cropped. Sent back each time.
  • Text inside the file telling the checker to pass everything. Ignored, and the file was sent back as the wrong document.

What AI document verification does not do

It does not confirm that a document is genuine. The checks compare the file with the request. They are not forgery detection or identity verification, and they do not replace the checks a firm must carry out under its own professional or anti-money-laundering rules.

It does not decide whether the content is correct. A bank statement can cover the right period and still show a transaction you need to ask about. That judgment stays with your team.

It can be cautious. As in our test, a correct file is sometimes sent to a person instead of being accepted. We prefer that to accepting a wrong file, and the reviewer can accept it with one decision.

How a firm sets the checks

Each checklist item carries its own rules: a required date range, a signature, a minimum validity, a minimum number of pages. Ready-made checklists are a starting point, and you can change the rules on each item before you send the request.

Every plan includes AI checks on every upload. On the Firm plan you can also write custom checks for your own documents, for example that an invoice total is stated in a given currency.

Frequently asked questions

Can AI tell whether a document is fake?

No. DocAsker checks that a file is readable, complete, of the requested type, within the requested dates and signed where required. It does not detect forgery or verify identity, so a firm's own verification duties still apply.

Which file types can be checked?

PDF, JPG, PNG, HEIC, Word and Excel. Photos taken with a phone are converted to PDF.

What happens when the AI is not sure?

The upload is flagged for a person at the firm. It is neither accepted nor sent back automatically. The reviewer sees the result of each check and decides.

Does the client see why a file was sent back?

Yes. The upload page shows the reason in plain words and lets the client upload a replacement straight away. The next reminder also lists that item as waiting for a retake.

Send your first document request today

Set up a checklist in a few minutes and let DocAsker handle the follow-up.