15 Prompt Engineering Examples You Can Copy Today

Prompt engineering examples are only useful if you can see why one version beats another. So each of the fifteen below shows the weak version people actually write, the strong version, and the one principle that made the difference.

Copy the structure. Do not copy my facts — the strong versions work because they carry context that is specific, and yours will be different.

Content and copy#

1. Blog outline. Weak: "Write a blog outline about AI automation." Strong: "You are writing for operations managers at 50-to-200-person logistics firms who have never automated anything. Produce an outline for a 1,400-word article on AI automation: six H2s, each with one sentence on the reader's objection at that point. No introduction section." Principle: name the reader and their objection.

2. Rewrite in brand voice. Weak: "Make this sound more like us." Strong: "Rewrite the draft below to match the voice in the two examples. Keep every fact. Shorter sentences. No rhetorical questions, no 'in today's landscape', no em dashes." Principle: show examples, ban patterns.

3. Headline variants. Weak: "Give me 10 headlines." Strong: "Ten headlines under 60 characters for this page. Five lead with the outcome, five with the objection. No colons. No questions. Include the phrase 'AI training' in at least three." Principle: split the set by angle so you get range, not ten of the same.

4. Repurpose to social. Weak: "Turn this into a LinkedIn post." Strong: "Turn this article into a LinkedIn post: hook in the first 12 words, one concrete number from the article, one thing I learned the hard way, a single-line closing question. Maximum 150 words. No hashtags." Principle: specify the shape, including what to cut.

5. Email sequence. Weak: "Write a 5-email sequence for our course." Strong: "Five emails for people who downloaded the checklist but did not buy. One objection per email, in this order: price, time, 'I can learn this free', 'my team won't adopt it', deadline. Each under 180 words with one call to action." Principle: one job per message, ordered by real objections.

Research and analysis#

6. Competitor teardown. Weak: "Analyse this competitor." Strong: "From the pasted page, extract: their promise in one sentence, who it excludes, pricing model, three claims they make without evidence, and the one thing they do better than us. Table format. Do not speculate about anything not on the page." Principle: define the fields and forbid speculation.

7. Data summary. Weak: "Summarise this spreadsheet." Strong: "From this CSV: the three biggest movers month over month with percentages, any row where the sample is under 30, and the one number I should be worried about. Show your calculation for each." Principle: ask for the calculation so you can check it.

8. Decision support. Weak: "Should we hire a freelancer or an agency?" Strong: "Compare freelancer versus agency for our situation against: cost per deliverable, ramp-up time, continuity risk, quality variance, management overhead. Work through each before concluding, list your assumptions, then argue the strongest case against your own recommendation." Principle: reasoning plus counter-argument. See chain-of-thought prompting.

9. Claim checking. Weak: "Is this true?" Strong: "For each claim in the text below, mark SUPPORTED, UNSUPPORTED or NEEDS SOURCE, and quote the exact sentence you are judging. Do not use outside knowledge; judge only against the document provided." Principle: constrain the evidence base.

10. Meeting to actions. Weak: "Summarise this meeting." Strong: "From this transcript produce: decisions made, open questions, and actions as owner plus task plus deadline. If an owner was not named, write UNASSIGNED rather than guessing." Principle: force an explicit gap marker instead of invention.

Operations and automation#

11. Ticket classification. Weak: "Categorise these support tickets." Strong: a few-shot prompt with two labelled examples per category and an explicit OTHER class. Principle: examples beat category definitions. See few-shot prompting.

12. SOP from a recording. Weak: "Write an SOP." Strong: "From this screen recording transcript, write a numbered SOP a new hire could follow. Each step: action, where to click, what confirms success. Flag any step where the transcript is ambiguous with TODO rather than filling the gap." Principle: define what each step must contain.

13. Structured extraction. Weak: "Pull out the key info." Strong: "Return JSON with exactly these keys: company, contact_name, role, budget_eur, timeline, next_step. Use null for anything absent. No prose before or after the JSON." Principle: exact schema, explicit nulls, no chatter.

14. Weekly report. Weak: "Write our weekly report." Strong: "Using the attached metrics, write the weekly report in our standard format: three wins with numbers, one problem with its owner, next week's single priority. Maximum 200 words. Use only numbers present in the data; write TK for anything missing." Principle: fixed format plus a rule against invented numbers.

15. QA against brand rules. Weak: "Check this is on brand." Strong: "Check the draft against the brand rules document attached. Return a table: rule, pass or fail, the offending line quoted. Do not rewrite anything; only report." Principle: separate reporting from fixing so you keep control.

The pattern behind these prompt engineering examples#

If a prompt of yours is underperforming, it is almost certainly failing one of those six rows.

Make the good ones permanent#

A prompt that works twice should never be retyped. Put the context and rules in a Claude Project so they apply automatically, and package the full procedure as a Claude Skill when more than one person needs the same output quality. Best Claude Skills for marketing shows the finished versions.

For the method behind these examples, work through my free prompt engineering course, and for more marketing-specific starting points see AI prompts for marketing. Anthropic's prompt engineering documentation is the best free reference when you want the model-side detail.

If you would rather have these built around your own briefs, with your team in the room, that is exactly what my AI training packages deliver.