AI tools come with a lot of jargon. Most of it is not as complicated as it sounds, but the terminology creates a real barrier to evaluating tools confidently. This glossary defines the terms you will actually encounter when reading AI tool reviews, comparing products, or talking to vendors — in plain English, without unnecessary technical complexity.
Core AI Concepts
- Artificial Intelligence (AI)
- Software that performs tasks that typically require human intelligence — understanding language, recognizing patterns, making decisions. In the professional tool context, "AI" almost always refers to software that can understand and generate natural language (text), though it also covers image recognition, prediction models, and other capabilities.
- Large Language Model (LLM)
- The underlying AI technology behind most text-based AI tools in 2026. An LLM is trained on enormous amounts of text and learns to predict what text should come next in any given context — which, at sufficient scale and sophistication, produces outputs that read as genuinely intelligent. GPT-4 (OpenAI), Claude (Anthropic), Gemini (Google), and Llama (Meta) are the most prominent LLMs underlying current professional tools.
- Generative AI
- AI that generates new content — text, images, audio, video, code — rather than simply classifying or analyzing existing content. When professionals talk about "AI tools" in 2026, they are almost always talking about generative AI. Tools like ChatGPT, Jasper, Harvey AI, and Suki AI are all generative AI applications.
- Hallucination
- When an AI tool generates information that sounds plausible and confident but is factually incorrect or fabricated. The term is most critical in legal and medical contexts: AI legal tools can hallucinate case citations (inventing cases that do not exist), and AI clinical tools can generate inaccurate dosing or diagnostic information. Understanding hallucination risk is why human review of AI output is non-negotiable in professional contexts, not optional.
- Fine-Tuning
- Training a general AI model on additional, domain-specific data to improve its performance in a particular field. When a vendor says their tool is "fine-tuned on legal data" or "trained on medical documentation," they mean they took a general LLM and trained it further on specialized professional content, which typically reduces hallucination rates and improves accuracy on domain-specific tasks.
- RAG (Retrieval-Augmented Generation)
- A technique where the AI searches a specific, curated database for relevant information before generating a response, rather than relying solely on what it learned during training. This is why tools like Casetext CoCounsel and Lexis+ AI cite real cases rather than making ones up — they are retrieving from a verified legal database rather than generating from memory. RAG significantly reduces hallucination risk for factual queries.
- Prompt / Prompt Engineering
- A prompt is the instruction or question you give an AI tool. Prompt engineering is the practice of crafting effective prompts — being specific enough about context, format, constraints, and goal to get reliably useful output. The difference between a vague prompt ("write a lesson plan") and a specific one ("write a 50-minute lesson plan for 9th grade students on photosynthesis, including a 10-minute pair activity and an exit ticket") is typically the difference between a generic and a genuinely useful output.
- Ambient AI
- AI that operates in the background listening to or observing a workflow rather than requiring explicit activation. AI medical scribes like Suki AI and Nuance DAX are ambient AI tools — they listen passively during a patient encounter and generate documentation afterward, rather than requiring the physician to speak commands or fill in forms.
- Tokens
- The units AI language models process — roughly equivalent to pieces of words. "Token" appears in AI pricing (you may pay per thousand tokens) and in context window limits (how much text the model can consider at once). In practice: a 1,000-word document is approximately 1,300-1,500 tokens. For most professional AI tool users, tokens are a pricing and technical detail managed by the vendor platform rather than something you actively manage.
- HIPAA (Health Insurance Portability and Accountability Act)
- US federal law governing the privacy and security of patient health information. For healthcare AI tools, HIPAA compliance is a baseline requirement, not a differentiator. Specifically relevant: any AI tool you use with patient data should sign a Business Associate Agreement (BAA) with your practice. The BAA is a legal contract establishing that the vendor handles PHI (Protected Health Information) in compliance with HIPAA. Do not use any healthcare AI tool with real patient data without a signed BAA.
- FERPA (Family Educational Rights and Privacy Act)
- US federal law governing student educational records. For education AI tools, FERPA compliance means the vendor handles student data with appropriate privacy protections. AI tools used in K-12 or higher education settings that process student information should be FERPA compliant. This is why purpose-built education tools like MagicSchool AI specifically advertise FERPA compliance, while general-purpose consumer AI tools typically are not appropriate for student data processing.
- SOC 2
- A security audit standard (Service Organization Control 2) that verifies a technology vendor's data handling practices meet established security, availability, and confidentiality standards. When an AI tool vendor claims "SOC 2 compliant" or "SOC 2 Type II certified," they are saying an independent auditor has verified their security controls. For professional AI tool procurement, SOC 2 Type II is the standard level to look for — it involves ongoing auditing rather than a one-time assessment.
- Business Associate Agreement (BAA)
- A contract required under HIPAA between a healthcare provider (or other covered entity) and any vendor that handles Protected Health Information on their behalf. For any AI tool used with real patient data, a signed BAA must be in place before use. HIPAA-compliant AI medical scribes and healthcare AI tools will sign a BAA as part of onboarding. If a vendor refuses to sign a BAA or does not have one available, do not use their tool with patient data.