Automation and AI

Prompt

Also called Instruction, query

The instruction and context you give a model in a single request, which largely decides how useful the answer is.

Quick facts: Prompt

Category
Automation and AI
Also called
Instruction, query
Level
Beginner
Affects
Output quality, consistency, editing time
Where to see it
ChatGPT, Claude, Gemini, Copilot, any model API
In this article4
  1. How a prompt works
  2. Why prompts matter
  3. Common mistakes with prompts
  4. How to act on it

How a prompt works

Everything the model has to go on for a given request is in front of it: your instruction, whatever material you attached, and the earlier turns of the conversation it can still see. It has no memory of your business, no access to your files unless a tool supplies them, and no way to ask you what you meant. It predicts the response that best fits what it was handed.

That is why the same tool gives one person vague filler and another person something usable. A weak brief leaves the model to fill the gaps with the most average answer available, and average is exactly what a model trained on the whole internet is best at producing. A specific brief narrows the possibilities before it starts.

Why prompts matter

Most disappointment with AI tools is a briefing problem rather than a capability problem. Asked to write about a business, a model writes something that could describe any business, because nothing in the request ruled the alternatives out. Given the audience, the offer, the objection to answer, the tone to avoid and an example of past work, it produces something with a shape.

Prompts also decide how much editing you do afterwards, and editing is where the time actually goes. A brief that specifies format and length up front saves more effort than any amount of rewriting later. Once a brief works, it is worth saving, because a repeatable instruction is a small asset — a template your team can reuse and improve rather than reinventing each time.

Common mistakes with prompts

Asking too broadly is the first. “Write a blog post about digital marketing” has no audience, no purpose and no boundary, so nothing in the answer can be wrong and nothing can be useful. Naming the reader and the job the piece must do fixes most of it.

Piling everything into one request is the second. A single instruction that asks for research, an outline, the draft and the meta description tends to do all four thinly. Splitting the work into steps gives you a point to correct course. The third is describing tone in adjectives. “Professional but friendly” means nothing consistent; showing a piece of your own writing communicates it immediately.

The fourth is trusting anything factual that comes back. A model will produce a statistic, a source or a feature claim to complete a sentence, and it reads exactly like the parts that are true.

How to act on it

Write briefs the way you would for a competent freelancer who has never met you. State who the reader is, what they should be able to do afterwards, what must appear, what must not, the format and roughly how long. Attach your own material — an old page, a price list, a call summary — because supplied facts beat remembered ones every time.

Then work in steps and keep what works. Ask for an outline, correct it, and only then ask for the draft. Save the version that produced good output where colleagues can find it, and check every claim before publishing, whether the output is a support reply or a piece of website copy. If you want to go further, few-shot prompting is the habit of showing examples instead of describing them.

Do and do not

Do

  • Name the reader and the job the piece must do
  • Attach your own facts instead of relying on the model's
  • Save briefs that work as reusable templates

Do not

  • Ask for research, outline and draft in one request
  • Describe tone in adjectives when an example will do
  • Publish a figure or source the model supplied unchecked

Questions people ask about this

What makes a prompt good?

Specificity about things the model cannot guess. Who the reader is, what you want them to do, what must be included, what to leave out, the format, and an example of the style you want. Attaching your own facts matters more than clever wording, because it replaces what the model would otherwise invent. Length is not the point; a short brief with the right constraints beats a long vague one.

Should I use the same prompt every time?

For repeated tasks, yes — build a template with blanks for the parts that change, such as the product, audience or offer. Consistency is the benefit: everyone gets output in the same shape and you can improve one template instead of everybody guessing separately. Review it occasionally, because model updates can change how a familiar instruction is interpreted.

Why does the model ignore part of my instruction?

Usually because the request contains too many competing demands, or because the instruction has scrolled out of what the model can still see in a long conversation. Contradictions also cause it — asking for something detailed and very brief at once. Split the work into steps, put the rules that matter most near the end of the request, and restate them in a fresh conversation.

Related terms

Found this useful?

Share it, or ask an AI to summarise it

Back to the glossary

Knowing the term is the easy part

Applying it to your own site and budget is the work. Book a call and I will tell you what actually applies to you.