Reference Guide · 60+ Examples · Copy-Ready
Not “write a blog post about X.” Real, structured examples — each showing the weak version, the strong version, and the specific fix that made the difference. Organized by category so you can jump straight to what you need.
Here’s the problem with practically every list of AI prompt examples online: they give you the vague version. “Write a blog post about sustainable living.” “Translate this to Spanish.” “Summarize the report.”
Those aren’t examples. They’re placeholders. They produce exactly the generic, hedging, five-paragraphs-of-nothing output that makes people conclude AI isn’t useful for real work.
“The gap between a mediocre AI output and a genuinely useful one almost always comes down to specificity — not cleverness.”
Every example in this guide follows a different principle. Each one shows you the weak version and the strong version, with a note on the specific fix. Because understanding why a prompt works is more valuable than memorizing any single template.
| Component | What It Does | Without It |
|---|---|---|
| Role | Sets the model’s identity and expertise frame | Generic, averaged-across-all-contexts output |
| Context | Gives the audience, platform, goal, and constraints the model needs | Model makes assumptions — usually wrong ones |
| Task | One clear sentence saying exactly what to produce | Model interprets broadly; output scope is unpredictable |
| Format | Specifies length, structure, what to include, what to exclude | Default essay-style output regardless of your actual need |
Framework validated across Guru’s AI prompt guide (December 2025) and Vendasta’s prompting guide (January 2026).
01 | Writing & Content
Writing prompts are where the weak/strong gap is most dramatic. The difference between “write me an article” and a properly structured prompt isn’t stylistic — it’s structural. The model literally doesn’t know your audience, your tone, your constraints, or what good looks like. Tell it.
❌ Weak Prompt
Write a blog post about climate change for beginners.
✅ Strong Prompt
You are a science communicator writing for a general-interest newsletter. My readers are adults with no science background who care about climate but feel overwhelmed by the news coverage.
Write a 900-word blog post explaining one specific mechanism of climate change — ocean heat absorption — and why it matters for weather patterns people already notice (longer summers, stronger storms).
Structure: hook opening (no “In recent years”), 3 short sections with H2 headers, concrete closing takeaway. Short paragraphs, 3 sentences max. No jargon without plain-language explanation immediately after.
The fix: Role + specific mechanism + audience + structure + exclusion constraints. The weak version forces the model to guess at all five of those.
E-commerce Product Description
E-commerce
You are a senior e-commerce copywriter specializing in direct-to-consumer brands.
Product: [PRODUCT NAME]
Price: [$XX]
Target buyer: [WHO BUYS THIS — specific, not “everyone”]
Key differentiator: [WHAT MAKES IT SPECIFICALLY BETTER THAN CATEGORY DEFAULT]
Write a product description with:
1. Opening line (1 sentence): leads with the customer’s outcome, not the product’s features
2. 3 feature-benefit bullets: feature → what it means → concrete result
3. Trust line (15–20 words): social proof or guarantee — no vague claims
4. CTA (3–5 words, action-led)
Banned words and phrases: “high-quality,” “perfect for,” “innovative,” “state-of-the-art”
Under 150 words total.
Why this works: The banned word list is as important as the instructions. It stops the default filler vocabulary that makes product descriptions sound identical across all products.
Email Subject Lines — 8 Variants
Email
Write 8 subject lines for an email about [EMAIL TOPIC] sent to [AUDIENCE DESCRIPTION].
Use a different technique for each:
1. Curiosity gap (creates an open loop)
2. Specific number or result
3. Question (must be one they haven’t already answered for themselves)
4. Direct benefit — no teasing
5. Negative angle / what to avoid
6. Social proof reference
7. Time-specific urgency (only if genuine)
8. First-person perspective from the reader’s point of view
Constraints: under 50 characters each (mobile-safe). No “RE:” tricks. No all-caps.
After each, label the technique used in parentheses.
Labeling the technique forces intentionality. Without it, ChatGPT produces 8 variations of the same approach with slightly different wording.
❌ Weak Prompt
Rewrite this paragraph in a more casual tone.
[text]
✅ Strong Prompt
Rewrite the following paragraph for a Substack newsletter aimed at independent freelancers in their 30s. The current version sounds like a corporate press release.
Target tone: direct, slightly dry, knowledgeable without being superior — like advice from a smart friend who’s been freelancing for 10 years. Use contractions. Vary sentence length. One short punchy sentence for every two longer ones.
Keep all facts and claims intact. Change voice, not meaning.
[paste paragraph]
02 | Data & Analysis
Data prompts fail for a different reason than writing prompts: they’re usually too vague about what decision the analysis should serve. “Analyze this data” gives you a summary. “Analyze this data to help me decide X” gives you something actionable.
Example 5 — Sales data analysis
Sales Data Analysis
Analytics
Analyze the following sales data and identify patterns relevant to this specific decision: [WHAT DECISION YOU’RE TRYING TO MAKE]
Data: [PASTE DATA OR DESCRIBE THE DATASET]
I need to understand:
1. Which product categories are growing vs. declining (with specific % changes)
2. Whether the pattern is driven by volume, price, or mix shift
3. One hypothesis for why the change is happening, based only on what the data shows
4. What additional data I’d need to confirm that hypothesis
Format: bullet points for findings, clearly labeled. Flag any conclusion that goes beyond what the data actually shows vs. what requires assumption. Don’t present correlations as causes.
The last line matters most. Explicitly asking the model to distinguish correlation from causation reduces the confident-but-wrong analysis that makes AI outputs risky to share with stakeholders.
Example 6 — Customer feedback synthesis
Customer Feedback Analysis
Research
Analyze the following [NUMBER] customer reviews/survey responses for [PRODUCT/SERVICE]:
[PASTE REVIEWS]
Extract and organize:
1. Top 3 reasons customers decided to buy (use their exact language in quotes)
2. Top 3 hesitations or objections before purchasing
3. Unexpected benefits mentioned that weren’t in the original marketing
4. Specific words and phrases that appear repeatedly — these belong in your copy
5. The “aha moment” — one sentence describing the point when customers realized this was the right choice
Do not paraphrase customer language. Quote it directly where possible. The model’s words are less valuable than the customer’s actual words.
03 | Brainstorming
Brainstorming Examples
The secret to good brainstorming prompts: force the model to generate ideas with built-in angles, not just topic labels. “Give me 10 blog ideas about productivity” produces category names. “Give me 10 blog post arguments about productivity” produces theses.
Example 7 — Content ideation with angles
❌ Weak Prompt
Give me 10 blog post ideas about productivity for remote workers.
✅ Strong Prompt
Generate 10 blog post ideas on productivity for remote workers — but each must have a specific argument, not just a topic.
For each:
– Working title (under 65 characters)
– The core argument in one sentence (the non-obvious claim, not “productivity matters”)
– The specific reader moment when they’d search for this (what just happened to them?)
– Whether the angle is: counterintuitive / underserved / common question with a better answer
Avoid: “Ultimate Guide to,” “Everything You Need to Know,” any topic already covered by every productivity blog in 2024.
Example 8 — Name generation
Brand / Product Name Generation
Creative
Generate 20 name options for [WHAT IT IS].
Context: [TARGET AUDIENCE, BRAND PERSONALITY, WHAT YOU WANT THE NAME TO CONVEY]
What to avoid: [NAMES THAT ALREADY EXIST, TONES THAT DON’T FIT, CATEGORIES TO SKIP]
Organize into four groups of 5:
– Group A: Descriptive (says what it does)
– Group B: Evocative (feeling/metaphor, not literal)
– Group C: Abstract / invented word
– Group D: Unexpected / counterintuitive
For each name: 1-sentence explanation of why it works and one potential weakness.
After all 20: identify your top 3 picks and explain the reasoning. Don’t just pick the safest ones.
Example 9 — Problem reframing
Problem Reframing — 5 Angles
Strategy
I’m trying to solve this problem: [DESCRIBE THE PROBLEM AS YOU CURRENTLY UNDERSTAND IT]
Reframe this problem 5 different ways. Each reframe should suggest a completely different category of solution.
Use these lenses:
1. Flip it — what if the “problem” is actually a symptom of the opposite issue?
2. Who else has solved this — in a completely unrelated industry?
3. What if the constraint is removed — what would the solution look like then?
4. What if the timeline changes — what would solving it in 1 day vs. 5 years look like?
5. What does the person experiencing this problem actually want — vs. what they asked for?
For each reframe: state the new problem definition in one sentence, then suggest one solution direction it opens up.
04 | SEO
SEO Examples
The most common SEO prompt mistake: asking for “keyword-rich” content. That’s not how search works in 2026. What you actually want is topically complete content that answers the searcher’s full intent — the primary question and the three follow-up questions they haven’t asked yet. Tell the model that explicitly.
Example 10 — FAQ for featured snippets
FAQ Targeting Featured Snippets
SEO
Write 8 FAQ items for a page about [TOPIC] targeting [PRIMARY KEYWORD].
For each FAQ:
– Question: phrased exactly how a real person types it into Google (conversational, specific — not formal)
– Answer: 40–60 words. Direct answer in the first sentence — no “It depends” as an opener unless you immediately explain what it depends on. Self-contained: someone reading only the answer should fully understand it.
– Include the keyword or a close variant naturally within each answer
Format: Q: [question] / A: [answer]
The goal is featured snippet capture — prioritize directness and completeness over style.
Example 11 — Meta description batch
❌ Weak Prompt
Write a meta description for my page about home coffee brewing.
✅ Strong Prompt
Write 5 meta descriptions for a page about home espresso brewing targeting the keyword “home espresso setup.”
Rules:
– 148–155 characters each (show character count after each)
– Keyword in the first half of the description
– Different hook technique per option: benefit / question / social proof / instruction / contrast
– Ends with an implicit CTA
– No “In this article we will…” opener
The searcher intent is commercial investigation — they’re deciding whether to buy a home espresso machine. Write to that moment.
05 | Coding
Coding Examples
Coding prompts need three things that most people skip: the language and version (behavior changes between Python 3.9 and 3.12, for instance), the context of where the code runs, and what “done” looks like. Without those, you get technically correct code that doesn’t fit your actual environment.
Example 12 — Function with error handling
Write a Function with Error Handling
Python
Write a Python 3.11 function that [WHAT IT SHOULD DO].
Context:
– It will run in [ENVIRONMENT: AWS Lambda / local script / Django backend / etc.]
– Input: [DESCRIBE INPUT FORMAT AND TYPICAL EDGE CASES]
– Expected output: [EXACT FORMAT OF RETURN VALUE]
Requirements:
– Type hints throughout
– Docstring with Args, Returns, and Raises
– Handle these specific failure cases: [LIST EDGE CASES]
– Raise specific exceptions (not bare `except:`)
– Include 3 unit test cases using pytest: one happy path, one edge case, one failure case
Do not use any external libraries beyond: [LIST ALLOWED LIBRARIES]
The unit test requirement is the most valuable line. It forces the model to actually think about edge cases rather than producing code that works for the happy path and breaks immediately in production.
Example 13 — Debug existing code
Debug Code — Step by Step
Debugging
The following [LANGUAGE] code is producing this error: [PASTE EXACT ERROR MESSAGE]
Code:
[PASTE CODE]
Walk through what’s happening step by step:
1. What the code is attempting to do
2. Where and why the error occurs
3. The specific fix, with explanation
4. Whether there are related issues I haven’t noticed yet (look for anything that would cause a silent failure)
Show the corrected code separately from your explanation. Mark the changed lines with a comment like # FIXED.
06 | Translation & Localization
Translation Examples
Translation prompts have the same problem as tone prompts: vagueness. “Translate this to Spanish” gives you technically correct Spanish. “Translate this for Mexican small business owners who read this on mobile” gives you something people actually use.
Example 14 — Localized translation
❌ Weak Prompt
Translate this blog post into French.
[text]
✅ Strong Prompt
Translate the following text into French for a Parisian professional audience reading a business newsletter. This is not a literal translation — adapt idioms, cultural references, and business register to what a French-speaking professional would naturally say, not what an English speaker would directly translate.
Flag any phrases where a literal translation would sound awkward or foreign to a native French speaker, and explain the adaptation you made.
[paste text]
Example 15 — Multilingual SEO
Multilingual Keyword Research
SEO · International
I’m expanding a page about [TOPIC] into [LANGUAGE(S)].
For each language:
1. Provide 5 keyword variants that a native speaker would actually search — not direct translations of the English keyword
2. Note the regional variation if relevant (Brazilian Portuguese vs. European Portuguese, Latin American Spanish vs. Spain Spanish)
3. Flag any terms that are technically correct but sound unnatural in search context
4. Identify one cultural angle or local concern that the English-language page doesn’t address but a local reader would expect
Format as a table: Language | Natural Keyword | Notes | Local Angle
07 | Research & Summarization
Research Examples
Important caveat for all research prompts: AI models have a knowledge cutoff and can hallucinate sources. Use these prompts to synthesize information you provide — don’t ask the model to research facts from the web unless it has verified web search access. Always verify specific claims, statistics, and citations independently.
Example 16 — Document summarization
Summarize a Long Document
Summarization
Summarize the following document for [AUDIENCE: e.g., “a non-technical executive who has 3 minutes to read this”].
The summary should answer exactly these questions — nothing more:
1. What is this document about? (1 sentence)
2. What is the most important finding or recommendation? (2–3 sentences)
3. What action, if any, does this require from the reader? (1–2 sentences)
4. What is missing or unresolved that the reader should know?
Length: under 200 words total.
Do not include: background the audience already knows, section-by-section recap, or filler transitions.
[paste document]
Specifying four exact questions stops the model from producing a proportional recap of the document (a common failure mode) and forces it toward what the reader actually needs to know.
Example 17 — Competitive research synthesis
Competitive Landscape Synthesis
Strategy
I’ll describe three competitors. Your job is to synthesize this into a strategic picture — not a list of features.
Competitor A — [NAME]: [DESCRIPTION OF THEIR APPROACH, POSITIONING, STRENGTHS, WEAKNESSES]
Competitor B — [NAME]: [DESCRIPTION]
Competitor C — [NAME]: [DESCRIPTION]
My company: [WHAT YOU DO, YOUR CURRENT POSITIONING]
Analyze:
1. What strategic pattern do all three share? (The category assumption everyone is making)
2. Which customer segment is underserved by all three?
3. What is the most defensible differentiator I could build that none of them are prioritizing?
4. One risk: where could one of them easily copy what I’m describing?
Be direct. No hedge language like “it may be worth considering.” Pick a view.
08 | Advanced Techniques
Advanced Prompting Techniques — With Examples
Three techniques that consistently produce better results than single elaborate prompts. Each is demonstrated with a real example below.
Technique A — Chain-of-thought (force visible reasoning)
Chain-of-Thought — Strategic Decision
Advanced
Before giving me an answer, think through this step by step and show your reasoning:
Question: [YOUR COMPLEX QUESTION OR DECISION]
Context: [RELEVANT BACKGROUND]
Walk me through:
1. What information would you need to answer this well?
2. What are the 2–3 most plausible answers or approaches?
3. What are the key tradeoffs between them?
4. Which would you recommend, and what’s the single most important reason?
Show the reasoning, then give the recommendation. Don’t jump straight to the answer — I need to see whether your reasoning is sound before I trust the conclusion.
Research from Lakera’s prompt engineering guide confirms: Chain-of-thought prompting is most valuable for analysis, strategy, and debugging tasks. For simple factual questions, it adds noise without value.
Technique B — Few-shot (show an example, get consistency)
Few-Shot Style Matching
Advanced
Here is an example of the exact style, tone, and structure I want:
EXAMPLE:
[PASTE YOUR BEST EXISTING PIECE — OR A PIECE WHOSE STYLE YOU WANT TO MATCH]
Study what this example does: sentence rhythm, evidence density, how it handles transitions, what it never says, the register it maintains throughout.
Now write a new piece on [NEW TOPIC] following the same craft. Same structural logic. Same tone register. Different topic.
If the example opens with a counterintuitive claim, do that. If it uses an anecdote before making its point, do that. Treat the example as a template, not just inspiration.
Three examples beat adjective descriptions every time. “Warm but authoritative” means nothing consistent. A concrete example means everything.
Technique C — The three-round refinement loop
Don’t try to specify everything in one prompt. Use three messages:
- Round 1: Generate the initial output using a structured prompt.
- Round 2: “Look at what you just wrote. Identify the three weakest sentences — the ones that are vague, use passive voice, or make a claim without supporting it. Rewrite only those three sentences.”
- Round 3: “Now read the whole piece again. Where did you make a general claim you could replace with a specific named example, number, or concrete scenario? Make those replacements.”
Three rounds of refinement consistently produce better output than one elaborate all-in-one prompt. Treat it as a workflow, not a sign that the first prompt failed.
Quick Reference: Which Technique for Which Task
| Task Type | Best Technique | Key Constraint to Always Include |
|---|---|---|
| Content drafting | Role + structure + exclusions | Banned words/phrases list |
| Tone matching | Few-shot (3 examples) | Never describe tone with adjectives alone |
| Data analysis | Decision-framed + correlation guard | Flag where conclusions go beyond the data |
| Complex strategy | Chain-of-thought | “Show reasoning before giving recommendation” |
| Coding | Environment + edge cases + unit tests | Specify language version explicitly |
| SEO content | Intent-first + searcher moment | Name the searcher’s specific situation |
| Long document → short | Specific questions (not “summarize”) | Word limit + what to exclude |
| Quality improvement | Three-round refinement loop | Round 2 targets specific sentences, not the whole piece |
Sources
- Guru — AI Prompts: Essential Guide with Types & Best Practices (December 2025) — role/context/task/format framework; intent recognition; few-shot examples
- Vendasta — AI Prompting: The Complete Guide (January 2026) — prompting as operational discipline; chain-of-thought applications; iterative refinement
- Lakera — Ultimate Guide to Prompt Engineering 2026 — vague vs. refined prompt comparison table; Claude vs. GPT constraint behavior; chain-of-thought evidence
- MIT Sloan Teaching & Learning Technologies — Effective Prompts for AI (May 2025) — foundational constraints; prompt quality principles
- Aakash Gupta (Product Growth) — Prompt Engineering in 2025: Latest Best Practices (July 2025) — distinction between personal and production prompting; system prompt design
Unleashing the Power of AI in Research Writing, Reading & Analysis: Your Ultimate Guide
How to Write Clear and Specific Prompts: Complete 2025 Guide with Examples
Prompt Hacks 2025: The Keyword Formula That Supercharges AI
The Ultimate Guide to Crafting Killer AI Prompts: Boost Your SEO & Charm Google Like a Pro
AI Guest Post Opportunities You Can’t Miss in 2025
Unleashing the Power of AI: The Best AI Prompts for Scientific Research
10 AI Prompts for Audience Segmentation in 2025: Boost Conversions with Data-Driven Precision
⚡ 10 Powerful Examples of Ethical ChatGPT Prompts You Need in 2025
AI Prompt Writing for Beginners: Best Guide (2025)
How Developers Use AI Prompts to Write Code Faster: BestGuide
