A prompt is the text you give an AI tool to tell it what you want: the question, the instruction and any material it should work from. Whatever a large language model writes back is shaped almost entirely by what the prompt contains and what it leaves out.
How a prompt works
The model reads your prompt as a sequence of words and predicts, piece by piece, the most fitting continuation. It has no knowledge of your business, your customers or your last conversation unless that information is in front of it. So a prompt is not just a request. It is the whole briefing the tool gets.
A useful prompt usually covers five things:
- The task: what you want produced, stated as an instruction (“Write three subject lines” rather than “Can you think about subject lines?”).
- Context: who the business is, who the reader is and what they already know.
- Source material: the facts the output must rest on, such as a product sheet, a price list or notes from a client call.
- Format: length, structure, headings, a table, plain text for pasting into an email tool.
- Constraints: what to avoid, which claims are off limits, the spelling and currency to use.
Many tools also carry a system prompt behind the scenes, a standing set of instructions that applies before your own prompt is read. What you type sits on top of that.
Why it matters
For a small marketing team, the prompt is where most of the quality is won or lost. Two people using the same tool on the same day can get a generic, slightly American blog intro or a usable first draft in the right voice, and the difference is nearly always the briefing.
UK businesses have a specific reason to be careful. Most AI tools default to US conventions: American spelling, dollar prices, US holidays, US consumer law and US date order. If the prompt does not ask for British English, prices in GBP and UK context, you will spend the time you saved correcting spelling and swapping “zip code” for “postcode”. A product page that quotes the wrong returns rights, or a Black Friday email dated 11/29, looks careless to a UK buyer.
Prompts also affect risk. If you ask a tool for “some customer statistics to back this up”, it may supply plausible numbers that come from nowhere, which is a form of hallucination. If those figures end up in an advert, the claim still has to meet the CAP Code, whoever wrote it.
Common mistakes
- One-line prompts for work that needs a brief. “Write a page about our boiler servicing” gets a page about boiler servicing in general, not about yours.
- No source material. Without your facts, the tool fills the gaps with likely-sounding ones.
- Asking for evidence instead of supplying it. Give the tool the figure and its source; never ask it to find one.
- Forgetting the locale. No mention of British English, GBP or the UK market.
- Pasting personal data. Customer names, emails and complaint details pasted into a consumer AI tool may be a disclosure your privacy notice does not cover.
- Judging the tool on a first attempt. A weak result is usually a weak brief; adjust the prompt before giving up.
How to act on it
Write a short reusable prompt for each recurring job: product descriptions, social posts, email subject lines, meta descriptions. Start each with the same block of context, for example: “You are writing for a family-run garden centre in Kent. Use British English and prices in pounds sterling. Readers are local homeowners, not trade buyers. Do not invent statistics, prices or customer quotes.” Then add the task and the material for that job.
Keep the prompts that work in a shared document so everyone on the team uses them, and note what you changed when a result improved. Over time that file becomes a small house style guide. If you want to get more structured about it, prompt engineering covers testing and refining prompts properly, and few-shot prompting shows the tool examples of what good looks like.
Finally, treat every output as a draft that a person checks. If AI is helping you produce web pages, the plan behind those pages matters more than the wording of any single prompt; that planning is the core of content SEO and strategy work.