How to automatically fill Word documents from a photo of an ID

A practical guide to turning a phone photo of an identity document into a correctly filled Word document: what recognition can read, how to review the result, and how to handle a client’s ID responsibly.

A client hands you an ID card. Somewhere in your office there is a Word document waiting for that person’s name, number, address and date of birth. In most offices the bridge between the two is a person reading the card and typing. This post covers a better bridge: photograph the ID, let software read it, and review the result before anything leaves your desk.

Why retyping IDs causes errors

Identity documents are dense with values that look almost alike. A national ID number is a long string of digits with no internal logic you can check by eye. Names carry diacritics and unusual spellings. Dates appear in more than one format on the same card. When a person retypes these values, the common failures are predictable.

  • Transposed digits in long numbers: two neighbouring digits swap and nothing looks wrong.
  • Dropped or doubled characters in names, especially names the typist has never seen before.
  • Date format confusion: 04.08 read as April 8 in one field and August 4 in another.
  • Copying the right value into the wrong field, such as the document number where the personal number belongs.

None of these errors announce themselves. A misspelled name in a declaration is found by the client, at signing, or later. The cost is not the retyping time. The cost is the correction round.

What data recognition can read from a phone photo

Modern text and data recognition reads a normal phone photo of an ID, front or back, without a scanner. From a reasonably sharp photo it can extract the values an office actually needs.

  • Full name, exactly as printed, including diacritics and non-Latin scripts such as Cyrillic.
  • Identity numbers: the personal or national number and the document number, as separate values.
  • Address, where the document carries one.
  • Dates: birth date, issue date, expiry date, kept apart by their labels on the card.
  • The issuing authority, where printed.

A few practical rules improve the read. Photograph the card flat, in even light, filling most of the frame. Avoid glare across the laminate. If the card has data on both sides, photograph both. A blurry photo does not produce wrong values so much as missing ones, and missing is the safer failure.

Filling a Word document with Run Done, step by step

Run Done is built for exactly this handoff. The flow takes a few minutes the first time and less after that.

  • Upload the Word document you already use. It opens as it is: your fonts, your headers, your layout.
  • Add the photo of the ID, and any other materials you have: scans, other documents, pasted text. You can also dictate details instead of typing them.
  • Run Done reads the materials and enters the data in the right places in the document. If a value cannot be found in the materials, the field stays empty. Nothing is invented.
  • Review each entered value, correct anything you want, and download a finished Word file.

A document you produce often can be saved and filled again next time in seconds. There is also a task pane inside Microsoft Word, so the same flow works in the document already open in front of you.

What a good review flow looks like

Automatic filling does not remove the review. It changes the review from retyping plus proofreading into proofreading alone, which is faster and more reliable. A good review flow has three properties.

  • Every entered value is listed, so you check a short list instead of re-reading the whole document.
  • Each value can be traced to where it landed in the document, so you confirm it in context.
  • Missing values are visible as gaps, not silently filled with a guess.

Check the long numbers digit by digit against the card once. That single check replaces the typing, the first proofread and the second proofread.

Privacy: this is your client’s ID, not yours

An ID photo is personal data of the most sensitive everyday kind. Before you send it to any tool, three questions matter: where is it stored, under what agreement, and is it used for anything else.

  • Storage: prefer servers in the European Union, with encrypted communication and access limited to your office. That keeps the data inside the GDPR’s home jurisdiction.
  • Agreement: for client personal data you should have a data processing agreement, in which the tool is the processor and your office remains the controller.
  • Purpose: the data should be used to fill your document and for nothing else.
Be careful with consumer chat tools. OpenAI’s published data-usage policy states that for personal Free, Plus and Pro accounts, conversations may be used to train models by default unless the user opts out; business tiers are not used for training by default. A client’s ID should never end up in a tool under settings you have not checked. Run Done stores data on protected servers in the EU, complies with GDPR, signs a data processing agreement, and uses your data for nothing else.

The short version

Photograph the ID instead of retyping it. Let recognition extract the names, numbers, addresses and dates. Review a list of entered values against the card once. Keep the client’s data in the EU, under a processing agreement, used for one purpose. The document is finished sooner and the correction round mostly disappears.

See it on your own documents

Your first 50 documents are free. Upload a document you already use and the materials you have, and download the finished file.