Free prompt library · no sign-up

Prompt Templates: 18 reusable prompts for real work

A prompt template is a reusable instruction with placeholders. The 18 templates below cover marketing, sales, engineering, support, research and HR. Copy one, replace the brackets with your facts, and run it on any Plugsky model. The structure is the value; the placeholders are what make it specific.

P
Plugsky Editorial Team
Reviewed 19 November 2026

How to use a template: copy the prompt, replace every [bracket] with a specific fact, add the output format you need, and run it. A template is a starting structure, not a guarantee. The model still needs your context, constraints and definition of “good”.

What makes a prompt template work

Most weak prompts fail for one of five reasons: no role, no task boundary, no context, no output format, or no constraint. A good template makes all five explicit.

  • Role. “You are a senior code reviewer” narrows the vocabulary and the standard the model applies.
  • Task. One verb and one deliverable. “Review this function for edge cases and return a numbered list” is testable; “help with my code” is not.
  • Context. The product, audience, stack, ticket or document the model cannot infer.
  • Format. JSON, a table, three bullets, a 150-word email. If the output feeds another system, specify the schema.
  • Constraints. Word limit, tone, sources, what not to invent, and what to do when a fact is missing.

The templates below include role, task and format. You add the context and constraints that are specific to your work. If a placeholder is unknown, do not let the model guess; tell it to ask or mark the gap.

Marketing templates

Ad copy

You are a direct-response copywriter. Write a high-converting ad for [product]. Target audience: [audience]. Key benefit: [benefit]. Proof point: [proof]. Call to action: [cta]. Return: 3 headline options (max 40 characters), 2 body variants (max 90 words), and one CTA line. Do not invent claims or statistics.

Email subject lines

Generate 10 email subject lines for a campaign about [topic]. Audience: [audience]. Tone: [tone]. Goal: [open/reply/click]. Keep each under 55 characters. Avoid clickbait and emoji. Return a numbered list, then mark the three you expect to perform best and why in one sentence each.

Landing page outline

You are a B2B landing-page strategist. Create an outline for a page selling [product] to [audience]. Include: hero promise, 3 pain points, 3 benefit blocks, proof section, objection handling, FAQ (5 questions), and final CTA. For each section, write the headline and one sentence of body copy. Flag any claim that needs evidence.

Sales templates

Cold email

Write a cold email to [prospect_name] at [company]. Context: [trigger/observation]. Value proposition: [value]. Goal: book a 20-minute call. Constraints: under 120 words, no generic praise, one specific observation, one clear ask. Return the subject line and body only.

Follow-up sequence

Write a 3-email follow-up sequence after a demo of [product]. Recipient: [role] at [company]. Main objection: [objection]. Each email: under 100 words, one new piece of value, one clear CTA. Do not repeat the same ask three times. Return subject + body for each email.

Discovery questions

List 8 discovery questions for a prospect interested in [product/service]. Cover: current process, cost of the problem, decision criteria, stakeholders, timeline, and cost of inaction. Return a table with the question and what a good answer tells the seller. Avoid leading questions.

Engineering templates

Code review

You are a senior [language] engineer. Review this code for correctness, edge cases, performance, security and readability: [code]. Return: (1) blocking issues, (2) non-blocking suggestions, (3) missing tests, (4) one rewritten function if a fix is small. Do not invent APIs or dependencies.

Bug root cause

Analyze this bug report and propose the most likely root cause: [report]. Include: reproduction steps, expected vs actual, the component most likely at fault, three hypotheses ranked by probability, and the cheapest diagnostic for each. Separate evidence from speculation.

Architecture decision

Compare architecture options for [requirement]. Options: [option A], [option B], [option C]. Score each on cost, latency, operational complexity, scalability, security and migration risk. Return a decision table, a recommendation, and the top three risks of the recommendation. State assumptions explicitly.

Support templates

Ticket reply

You are a customer support specialist. Reply to this ticket in a [tone] tone: [ticket]. Acknowledge the issue, explain the next step, give a realistic timeline, and avoid blaming the customer. Under 140 words. If information is missing, ask one specific question.

Escalation note

Write an escalation note for a support issue that needs engineering attention: [issue]. Include: customer impact, reproduction steps, what support has already tried, severity, workaround, and the specific engineering question. Keep it under 200 words and use bullet points.

Knowledge-base article

Write a knowledge-base article on how to [topic] for [audience]. Include: who this is for, prerequisites, numbered steps, expected result, common errors and fixes, and when to contact support. Use plain language. Do not invent menu names or UI labels; use [placeholder] where the exact label is unknown.

Research templates

Paper summary

Summarize this research paper: [paper]. Return: research question, method, sample/data, main result, limitations, and one practical implication. Separate what the authors measured from what they speculate. Use no more than 250 words.

Competitor comparison

Compare [competitor] with [our_product] across: pricing model, core features, integrations, target audience, deployment options and support. Use only the sources I provide: [sources]. Return a table with a “source” column and a short “where they win” section. Do not fill gaps with guesses.

Market sizing

Estimate the market size for [product] in [region] using a top-down approach. Show: total addressable market, serviceable market, target segment, assumptions, and a sensitivity range. Cite the inputs I provide: [inputs]. If a required input is missing, list it rather than inventing a number.

HR templates

Job description

Write a job description for a [role] at [company]. Include: mission, 5 responsibilities, must-have skills, nice-to-have skills, and what success looks like in 90 days. Tone: [tone]. Avoid gendered language and unnecessary degree requirements. Keep it under 500 words.

Interview questions

Create 6 interview questions for a [role] that test [skill]. For each: the question, what a strong answer includes, and one follow-up probe. Include two behavioral and two practical questions. Avoid brainteasers and questions that invite illegal or personal disclosures.

Performance review

Draft a performance review for an employee who excelled at [project]. Include: outcomes, behaviors, one development area, and two specific goals for the next quarter. Use evidence from [examples]. Keep the tone fair and specific; avoid personality judgments.

Choosing the model for the template

A template is model-agnostic, but the model changes the result. A practical split:

  • Fast, low-cost model: short extraction, classification, first drafts and high-volume support replies. Use plugsky-lite or plugsky-plus.
  • Stronger reasoning model: code review, architecture decisions, competitor analysis and multi-step research. Use plugsky-pro or plugsky-max.
  • Long-context model: whole-document or multi-file work. Use a 256K-context model, but retrieve the relevant section before sending everything. The context-window calculator shows what is actually usable.

Run the same template on two models and compare output, latency and cost. The LLM API cost calculator turns a workload into a monthly bill; the pricing page shows the flat-plan caps.

Common prompt-template mistakes

  • Leaving placeholders generic. “[product]” produces generic copy. “A $20/month API for two-person SaaS teams” produces specific copy.
  • No output format. If the answer must be JSON, a table or 120 words, say so. Otherwise the model chooses a format you have to clean up.
  • No missing-fact rule. Add “if a required fact is missing, ask one question” to stop invented numbers.
  • Too many tasks in one prompt. Research, write, critique and format in one call produces mediocre work. Split the steps.
  • No evaluation. A template is not good because it is long. Test it on 20 real inputs and score correctness, format compliance and edit distance.

FAQ

What is a prompt template?

A prompt template is a reusable instruction with placeholders. You keep the structure and variable slots, then fill in the product, audience, code or ticket for the task in front of you. The template reduces blank-page work and makes outputs easier to compare.

Are these prompt templates free?

Yes. The library is free to read and copy. Running a template on Plugsky is free within the free-tier caps; paid plans start at $8 per month. See pricing for the current caps.

Which model should I use for a template?

Use a fast, low-cost model for short extraction, classification and first drafts. Use a stronger reasoning model for code review, architecture, multi-step analysis and long documents. Test the same template on two models before choosing one.

Why do placeholders like [product] matter?

They separate the reusable structure from the facts that change. Filling every placeholder with specific, concrete detail is the difference between a generic answer and a useful one. If a placeholder is unknown, ask for the missing fact instead of letting the model invent it.

Run any template on Plugsky

39 models, an OpenAI-compatible API and a free tier. Copy a template, change the base URL, and test it on your own input.

Start Free → API docs