FRENES

Glossary · Everyday AI use

OCR

OCR, short for optical character recognition, is a technology that turns text shown in an image into characters software can use. It allows you to search, copy or process words from a scanned page, a photograph or a document that does not already contain selectable text.

OCR: the explanation

An OCR system locates text within an image and estimates the characters and words it contains. It may also try to preserve layout information, including lines, columns and paragraph boundaries. Accuracy depends on factors such as sharpness, contrast, orientation and language support. Depending on the tool, the output may be plain text, a searchable document or structured elements that another program can use. Recognizing the text and reconstructing its layout are related but distinct tasks.

OCR makes printed and photographed material easier to search and reuse. It is used for digitizing archives, processing receipts and making older documents available to software. For an AI agent, it often provides an initial reading step before summarization or information extraction. The agent then works from the recognized text rather than directly from the paper. Any mistake introduced during recognition can therefore affect later answers, even if the subsequent processing behaves as intended.

OCR does not guarantee a perfect copy or a correct interpretation of a document. It can confuse similar characters, mix columns or detach a value from its label. Handwriting and complex tables may require specialized capabilities. Some products combine OCR with layout analysis and image-capable models to handle these challenges. Such combinations can improve usefulness, but important amounts, dates and identifiers still need checking against the original image before they drive an action or become a permanent record.

A concrete example

You photograph an invoice and ask an agent to prepare an expense entry. OCR extracts the words and numbers, then the agent identifies the supplier, date and total. A crease makes one amount difficult to read. Before saving the expense, you compare the proposed fields with the photograph and correct the value that the recognition system interpreted incorrectly during the initial extraction.

Why it matters

OCR quality affects every later task that relies on a scanned document. An agent can process text sensibly even when that text misrepresents the original page. Check language support, table handling and whether you can return easily to the source image. For sensitive files, also review where processing happens and how uploaded documents are retained by the service and any connected components.

Common misconceptions

  • A document with selectable text is not necessarily error-free, because an OCR text layer can contain mistaken characters or an incorrect reading order even when the visible page looks clear and perfectly understandable to a person reviewing the image.
  • OCR does not automatically understand the meaning of an invoice or contract, since recognizing text is different from identifying fields, relationships and obligations, and the additional systems used for those tasks can introduce their own interpretation or extraction errors.

Related terms: everyday ai use

prompt
A prompt is input supplied to a model to guide what it should produce or do.
RAG (retrieval-augmented generation)
RAG, or retrieval-augmented generation, combines information retrieval with a model’s ability to generate an answer.
vector database
A vector database stores and searches numerical representations called vectors.
semantic search
Semantic search retrieves material according to its likely meaning, beyond matching the same words.
coding assistant
A coding assistant is an AI tool that helps you write, understand, change or test computer programs.
no-code
No-code is an approach to building applications or automations through visual interfaces and settings, without directly writing code for the supported functions.
text-to-speech
Text-to-speech is a technology that converts written text into machine-generated speech.
speech recognition
Speech recognition is a technology that turns spoken audio into text a computer can process.

Where this site covers it

The term appears on 20 pages of this site. A few of them:

Same family

Go further

Frequently asked questions

Can OCR read handwriting?

Some systems recognize handwriting, but results depend on the writing style, legibility and model capabilities. A tool that performs well on printed pages may be less suitable for personal notes, handwritten addresses or forms completed by different people.

Does every PDF contain text?

No, a PDF may consist entirely of page images. OCR can add a text layer for searching and copying. If usable text already exists, extracting it directly may avoid an unnecessary recognition step and the errors that step could introduce.

How can you improve accuracy?

Use a sharp, correctly oriented image with good contrast and minimal shadows, and select the right language when possible. Compare important fields against the original, especially numbers, names and information arranged across several columns or inside complex tables.

Neutrality. No affiliate links. No vendor pays to be reviewed, rated or ranked. Reviews are directed by François Chollet, neutral and factual, and every data point is dated and sourced. How we score