What is an LLM? A plain-English guide for business owners

Large language models, or LLMs, are the technology behind tools like Claude, ChatGPT and Gemini. You don’t need to understand the maths to use them well, but knowing roughly how they work helps you decide where they’ll save time and where they’ll cause problems.

What an LLM actually does#

An LLM is a program trained on a huge amount of text: books, websites, code, articles. From all that text it learned patterns in language, and it uses them to do one thing: given some text, predict what text should come next.

That sounds simple, but at a large enough scale it becomes surprisingly capable. To continue a question with a good answer, the model has to pick up grammar, facts, reasoning and tone along the way. So when you ask it to summarise an email, it isn’t looking up a stored answer; it’s writing a new reply, piece by piece, based on everything it learned.

Words you’ll hear#

Prompt: the text you send the model, such as your question, instructions and any documents.
Tokens: the pieces models read and write, usually a word or part of a word. Pricing and limits are counted in tokens.
Context window: how much text the model can consider at once, including your prompt and its answer. Modern models can read entire long documents in one go.
System prompt: standing instructions a developer gives the model, like “You are the assistant for Example Studio. Answer only questions about our services.”
Hallucination: when the model states something false with confidence.

What LLMs are good at#

Writing and rewriting: drafts, product descriptions, emails, translations, changing the tone of a text.
Summarising: turning long documents, meeting notes or reviews into a few clear points.
Extracting information: pulling names, dates, prices or categories out of messy text such as emails or forms.
Answering questions about documents you provide.
Classifying: sorting support messages by topic or urgency, or flagging spam.
Writing and explaining code.

What they’re bad at#

Facts they weren’t given. A model’s general knowledge stops at its training date, and it can confidently fill gaps with made-up details. If accuracy matters, give it the facts in the prompt.
Exact maths and counting over large amounts of data. A normal program or spreadsheet is more reliable.
Knowing your business. It knows nothing about your prices, policies or customers unless you tell it.
Being right every time. The same question can get slightly different answers. For anything important, a person should check the output, or the system should check it automatically.

What it costs#

Most businesses use LLMs through an API: your website or app sends text to the provider and pays per token, for both the text sent and the text returned. For short tasks, like answering a visitor’s question or tagging a support message, the cost per request is usually a fraction of a cent. Costs grow with long documents, long conversations and the most capable models.

Ways to keep costs low:

Use a smaller, cheaper model for simple tasks.
Send only the information the task needs.
Set a monthly spending limit in the provider’s console.

Where an LLM can help a small business#

Good first projects are tasks that are repetitive, text-heavy, and easy to check:

An assistant on your website that answers common questions from your own information
Drafting replies to enquiries for a person to review and send
Summarising form submissions or reviews into a daily email
Writing first drafts of product or listing descriptions
Sorting incoming messages and sending them to the right person

Start with one task, measure the time it saves, and expand from there.

Privacy#

Before sending customer data to an AI provider, check its data policy: whether it stores your data, for how long, and whether it’s used for training. Business API plans usually have stricter terms than free consumer apps. Send only what the task needs, and avoid sensitive personal data unless your agreement covers it.

Next steps#

To make an LLM answer from your own content, see RAG explained. For better results from any AI tool, see how to write prompts that get better AI answers.

If you’d like AI added to your website or workflow, I do AI integration.