Prompt Engineering: The Six Patterns That Matter
Share
Prompt engineering means writing instructions that get useful, reliable output from an AI tool. You do not need forty prompt techniques. You need six, and you can learn all six this afternoon.
Direct answer
Six patterns cover most work tasks. Set the role and audience, give a worked example, state constraints before the task, supply the source material, ask for the reasoning on anything numerical, and change one thing at a time when iterating.
Based on the AI and Automation Practitioner standard (ST1512), a Level 4 apprenticeship for people already in work. Funding rules described here are the England position and are set by the funding body.
Potentially free for you
The learner never pays. Employers potentially pay nothing either, and where the apprentice is aged 16 to 24 the government funds the whole cost even if the employer does not pay the levy.
Closing any gap. A levy transfer from another employer covers 100% of the cost. Unused transfer allowance is not carried forward at the end of the tax year, and it must be agreed before the apprentice starts.
From 1 August 2026 the government stopped adding the 10% top-up to new funds entering levy accounts, so there is less in the system than there was.
Part of our guide to Prompt Engineering: What It Is and How to Learn It.
Pattern one and two: role and example
Say who is writing and who is reading. A complaint response for a regulator reads nothing like one for a customer, and the request alone does not reveal which you need.
Then give one worked example of the output you want. Paste in the best version you ever produced by hand.
This single move fixes more bad output than anything else, because it replaces your description of the format with the format itself.
If you have no example, write three sentences of the opening you want. Even a partial example outperforms a paragraph of description.
Pattern three and four: constraints and source
State constraints before the task, not after. Length, tone, things to avoid, house style. Constraints added later get lost in the rewrite.
Then supply the source material. Paste the document, the policy, the figures. Do not ask the model to recall them.
Anything it has to remember, it can invent. Anything you gave it, it can quote and you can check.
This is the difference between a tool that drafts from your material and one that produces confident fiction.
Pattern five and six: reasoning and iteration
For anything with a number in it, ask how the answer was reached. You are not being polite, you are creating something checkable.
When the output is wrong, change one instruction and try again. Changing three at once and getting a better answer teaches you nothing.
Keep the versions. The prompt that worked is the asset, and most people rebuild it monthly because they never saved it.
Two rounds of single-variable iteration usually beats ten rounds of rewriting everything.
There is more on this in our guide to Prompt engineering explained.
The four mistakes beginners make
Being polite instead of specific. Please write something good about our service is a wish, not a brief.
Asking for too much in one go. A five-part request produces five mediocre parts. Split it.
Accepting the first output. The first draft is a starting position, and the second round is where the value is.
Trusting anything factual without checking. These tools produce wrong figures in exactly the same confident tone as right ones.
Prompting for UK output
Add UK English to your constraints and give one paragraph of your own writing as the example. The example does more than the instruction.
Specify date format explicitly. Ambiguous dates interpreted the American way are the most damaging silent error in workplace output.
For anything about rules, entitlements or funding, supply the gov.uk page rather than asking. UK regulation is where these tools are least reliable.
Ask for sterling explicitly where money is involved, because dollar amounts appear in generated UK documents more often than you would expect.
Building your own library
Pick three recurring tasks. Write and refine a prompt for each until the output needs only light editing.
Save them somewhere the team can reach, with a one-line note on what each is for.
Review them quarterly. Models change and a prompt that carried heavy scaffolding may now work without it.
That library, in your organisation's voice, is worth more than any certificate you could buy.
Key takeaways
- Role and audience first, then one worked example.
- Constraints go before the task, never after.
- Paste the source rather than asking it to recall.
- Change one instruction at a time when iterating.
- Save the prompt that worked. That library is the real asset.
Frequently asked questions
What is the most important prompting technique?
Giving one worked example of the output you want. It replaces your description of the format with the format itself, and it fixes more bad output than anything else.
How long does it take to learn prompting?
The six core patterns take an afternoon. Getting reliably good on your own work takes about a fortnight of deliberate practice.
Why does my prompt produce generic output?
Usually because you described the format instead of showing it, or because you did not say who the audience is. Both are quick to fix.
Should I be polite to AI tools?
It makes no measurable difference to output quality. Being specific does. Spend the words on constraints rather than courtesy.
Why does it make things up?
Because it predicts plausible text rather than looking things up. Anything it has to recall, it can invent. Paste the source in instead.
How do I improve a prompt that is not working?
Change one instruction and try again. Changing several at once means you learn nothing about which change helped.
Should I save my prompts?
Yes. Most people rebuild the same prompt monthly because nobody saved it. A shared library in your own voice is the real deliverable.
Do I need a course to learn this?
No. The vendor documentation is free and current. A course buys structure and a deadline, which is worth something only if you need those.
The funded route, in figures
Figures are the England position. Levy-paying employers usually draw the whole cost from the levy they already contribute, so they pay nothing extra. Employers below the £3 million pay bill threshold pay nothing for an apprentice aged 16 to 24. For an older apprentice the government funds 95 per cent, and a levy transfer from a larger employer can cover the rest. Eligibility and funding rules are set by the funding body and can change.
Where to go next
If you want the prompt library you build assessed as part of a recognised qualification, the funded route does exactly that.
We help UK employers and employees access government-funded AI training. Training is delivered by third-party providers. We do not guarantee a place, a qualification, a job or any particular outcome. Eligibility and funding rules are set by the funding body and can change. Always check the current position before you commit. See our disclaimer.