Half the reason AI feels inaccessible to business owners is the vocabulary. The technology is often simpler than the words wrapped around it, and some of those words exist mainly to make consultants sound clever.
Here’s the working glossary we wish everyone had. For each term: what it means, and whether you actually need to care.
AI (artificial intelligence)
The umbrella term for software that does things we’d normally expect to need human judgement: writing, reading, deciding, recognising. It covers everything below, which is why the word on its own tells you almost nothing. When someone says their product “uses AI”, your next question should be: to do what, exactly?
Do you need to care? About the umbrella term, no. About the specific things underneath it, yes.
LLM (large language model)
The engine behind tools like ChatGPT, Claude and Gemini. It’s a program trained on enormous amounts of text that predicts what words should come next, which turns out to make it startlingly good at writing, summarising, translating and answering questions. When people say “AI” today, they usually mean an LLM.
Do you need to care? Yes, this is the single most useful piece of technology to reach small businesses in decades. You don’t need to know how it works, just what it’s good at (words, summaries, first drafts) and what it’s bad at (see hallucination).
Prompt
The instruction you give an LLM. “Write a polite payment reminder for an invoice that’s 14 days overdue” is a prompt. Better instructions get better results: more context, clearer output requirements, an example of the tone you want.
Do you need to care? Yes, but don’t overthink it. “Prompt engineering” is mostly just being specific, the same skill as briefing a new employee well.
Hallucination
When an LLM confidently makes something up: a fake statistic, an invented case, a plausible-sounding wrong answer. It’s not lying, the model has no idea it’s wrong. It’s the single most important limitation to understand.
Do you need to care? Absolutely. The rule that follows from it: AI drafts, you check. Never let an unreviewed AI answer reach a customer, a contract or a tax return.
Automation
Software doing a task without you: missed call triggers a text-back, invoice overdue triggers a reminder, form submitted triggers a booking email. Automation is often just rules (when X happens, do Y), and plenty of the highest-value automation in your business involves no AI at all.
Do you need to care? More than any other word on this list. For most businesses, boring rule-based automation pays for itself faster than anything with “AI” in the name.
No-code
Tools you assemble by clicking and connecting rather than programming: Zapier and Make are the big names. They’re the glue between the apps you already use: “when someone fills in my website form, add them to my spreadsheet and send me a text.”
Do you need to care? Yes. No-code tools are how non-technical owners build real automations. They’re most of what we use in our masterclass workshop.
Agent
An AI that doesn’t just answer a question but takes a series of actions to finish a job: reading your emails, deciding which need replies, drafting them, filing the rest. The industry’s current favourite word, and genuinely powerful (we build agent-based systems ourselves), but it’s also slapped onto products that are really just chatbots.
Do you need to care? Increasingly, yes, but be sceptical of the label. Ask what the “agent” actually does, step by step. If the seller can’t tell you, walk away.
Chatbot
A program that holds a conversation, usually on a website or WhatsApp. Old chatbots followed rigid scripts and infuriated everyone; modern LLM-powered ones can genuinely answer customer questions about your prices, availability and services.
Do you need to care? If you get repetitive customer enquiries, yes, it’s one of the easier wins. Give it your real information to draw from, and give people an obvious route to a human.
API
The plug socket that lets one piece of software talk to another. When your booking system texts your customers, an API is doing it. You’ll never build one, but tools you buy will talk about “API access”, which just means “this can connect to your other systems”.
Do you need to care? Only enough to not be intimidated. Sentence to remember: API means it connects.
Token
The chunks of text an LLM reads and writes, roughly three-quarters of a word each. AI tools are often priced by the token, which is why some bills scale with how much you use them.
Do you need to care? Only when comparing prices. Otherwise ignore.
RAG (retrieval-augmented generation)
A setup where the AI looks things up in your documents (your price list, your policies, your past quotes) before answering, instead of relying on its general training. It’s how you get a chatbot that knows your actual cancellation policy rather than inventing one.
Do you need to care? Know the idea, skip the acronym. If a supplier says “we use RAG”, they mean “it reads your documents first”. That’s a good thing.
Machine learning / fine-tuning
Machine learning is the broad technique of training software on examples rather than rules; fine-tuning means additionally training a model on your specific data. Both real, both important to the people building models, and almost never something a small business needs to buy directly.
Do you need to care? No. If someone tells you your ten-person firm needs a fine-tuned model, get a second opinion.
That’s the vocabulary. The pattern behind it: the words are stand-ins for simple ideas: software that writes, software that follows rules, software that connects. Once you can see through the terminology, deciding what your business actually needs gets much easier.
And if you’d rather skip straight to the doing: our AI Automation Masterclass at Heron House in Manchester is a half-day, plain-English, laptops-open session where you build working automations for your own business. No jargon quiz at the door. Tickets are normally £20. This one’s free, a limited-time offer to launch the series.