AI, automation & document glossary

All terms

• OCR (Optical Character Recognition) — OCR is the technology that converts text inside an image or scanned PDF into machine-readable, searchable, editable text.
• ICR (Intelligent Character Recognition) — ICR is an advanced form of OCR that recognizes handwritten or cursive characters using machine learning.
• OCR Confidence Score — A 0–100 score an OCR engine assigns to each recognized character or word, indicating how certain it is.
• Searchable PDF — A PDF that contains an invisible text layer behind the scanned image, making the document searchable and selectable.
• Image-Based PDF — A PDF whose pages are scanned images with no underlying text layer — meaning text cannot be searched or selected.
• PDF/A — PDF/A is an ISO-standardized PDF variant designed for long-term archiving — all fonts and resources are embedded so the file renders identically forever.
• PDF/X — PDF/X is a PDF variant standardized for print production — color profiles, fonts, and bleeds are constrained so the file prints reliably.
• Tagged PDF — A tagged PDF carries a hidden structure tree (headings, paragraphs, lists, tables) that screen readers and automation tools use to understand reading order.
• Reading Order — The logical sequence in which page content should be consumed — critical for screen readers, OCR output, and PDF-to-Word conversion.
• Table Extraction — The process of detecting tabular regions in a PDF and exporting them as structured rows and columns — usually into Excel or CSV.
• PDF Redaction — Permanently removing sensitive text or images from a PDF — not just covering them with a black box.
• PDF Encryption — Cryptographic protection that requires a password to open or modify a PDF — typically AES-128 or AES-256.
• PDF Permissions — Flags inside a PDF that limit actions like printing, copying, or editing — enforced by viewers, not cryptography.
• PDF Compression — Reducing PDF file size by re-encoding images, subsetting fonts, removing duplicates, and optimizing object streams.
• Linearized PDF — A PDF reorganized so the first page can render before the whole file finishes downloading — also called 'Fast Web View'.
• Bates Numbering — Sequential identifiers (e.g., ACME000001) stamped on every page of a PDF — standard practice in legal e-discovery.
• PDF Form (AcroForm / XFA) — Interactive fields inside a PDF that users can type into and submit — built on AcroForm or the legacy XFA standard.
• PDF Digital Signature — A cryptographic signature embedded in a PDF that proves who signed it and that the file has not been altered since.
• Document AI — AI systems that combine OCR, layout analysis, and language models to extract structured data from invoices, receipts, contracts, and forms.
• PDF Metadata — Hidden descriptive fields inside a PDF — title, author, creation date, software used, XMP packets, and custom keys.
• Large Language Model (LLM) — An LLM is an AI model trained on huge amounts of text that predicts language well enough to write, summarise, translate, and answer questions.
• Prompt Engineering — Prompt engineering is the practice of writing instructions that reliably get the output you want from an AI tool.
• Fine-tuning — Fine-tuning is retraining an existing AI model on your own examples so it adopts your style, format, or domain knowledge.
• RAG (Retrieval-Augmented Generation) — RAG is a technique where an AI tool searches your documents first, then writes an answer grounded in what it found.
• API (Application Programming Interface) — An API is a defined way for one piece of software to request data or actions from another, and it is how AI tools plug into your stack.
• No-Code — No-code describes software that lets you build apps, sites, and automations visually, without writing programming code.
• Webhook — A webhook is an automatic message one app sends to another the moment an event happens, so workflows start instantly instead of on a schedule.
• Token (AI) — A token is the small chunk of text — roughly three-quarters of a word — that AI models read and generate, and that most AI pricing is based on.
• Hallucination (AI) — A hallucination is when an AI tool produces confident, fluent output that is factually wrong or entirely invented.
• Context Window — The context window is the maximum amount of text, measured in tokens, an AI model can consider at one time.
• Zero-shot vs Few-shot — Zero-shot means asking an AI to do a task with no examples; few-shot means including a handful of examples so it copies the pattern.
• Multimodal AI — Multimodal AI describes models that understand or produce more than one type of content — text, images, audio, and video together.
• AI Agent — An AI agent is a system that plans multi-step work and uses tools or APIs to complete a goal, instead of just answering one question.
• Workflow Automation — Workflow automation is connecting apps so a trigger in one automatically runs steps in others, with no manual copying between tools.
• Freemium — Freemium is a pricing model offering a permanently free tier with limits, with paid plans unlocking capacity or advanced features.
• Open Source (AI Tools) — Open-source AI means the code, and sometimes the model weights, are published so anyone can inspect, self-host, and modify the software.
• SaaS (Software as a Service) — SaaS is software delivered over the internet on a subscription, hosted and maintained by the vendor instead of installed by you.
• Vector Database — A vector database stores text or images as numerical embeddings so software can find results by meaning rather than exact keywords.
• Natural Language Processing (NLP) — NLP is the field of AI focused on making software understand and work with human language — classifying, extracting, translating, and summarising it.
• Generative AI — Generative AI is any AI that creates new content — text, images, audio, video, or code — rather than only analysing existing data.