Email replies to clients: the "context + options + soft close" template
I write twenty client emails a week. Proposals, updates, polite "no"s, requests for feedback. The ones that take longest are the careful ones: saying no without burning a bridge, asking for a decision without sounding pushy, explaining a delay without excuses.
Here's the template I use for almost all of them:
You're replying to [client name], a [relationship: new lead / longtime client / former client]. They asked [specific question or request]. Your tone is [warm and professional / apologetic but firm / enthusiastic].
Context (1–2 sentences): Acknowledge what they said and show you understood the situation.
Options or answer (2–3 sentences): Give them a clear path forward—either your recommendation or two choices they can pick between.
Soft close (1 sentence): Make it easy for them to reply or take the next step.
Keep it under 100 words. No jargon. Sound like a human who's done this before.
Why this works: Most email prompts fail because they're too vague ("write a professional email") or too rigid ("use this exact template"). This one forces you to name the relationship and tone up front, so the model adjusts naturally. The three-part structure keeps replies short and actionable.
Kryotta tip: I run this through Claude Haiku 4.5 first—it's fast and handles tone really well. If the reply feels too formal or too casual, I paste it into Gemini Flash and ask, "Make this warmer but still professional." Two models, fifteen seconds, done.
Product listings: the "benefit-first, then proof" formula
I've written hundreds of product descriptions for Flipkart, Amazon, Meesho. The ones that convert follow the same pattern: lead with the benefit the buyer cares about, then give the specs that prove it.
Product: [name and category]
Target buyer: [who needs this and why—be specific: "a mother buying her first mixer grinder" or "a freelancer who works from cafés"]
Write a 60–80 word product description:
- First sentence: the main benefit in plain language (what changes for the buyer).
- Second sentence: one or two specs that prove the benefit.
- Third sentence: a secondary benefit or use case.
Tone: helpful, not salesy. No superlatives like "best" or "premium" unless they're true and provable.
Why this works: Most listings bury the benefit under a pile of specs. "500-watt motor, stainless steel blades, 3-speed control." Cool, but what does that do for me? This template forces the model to start with the outcome—"Grinds a week's worth of masala in under two minutes"—then backs it up with the 500-watt motor.
Kryotta tip: Use Gemini Flash Lite for the first draft—it's cheap and fast. Then compare it side-by-side with GPT-OSS 20B in the Compare arena. GPT-OSS often nails the casual, benefit-first language that feels less robotic.
Code review comments: the "what, why, suggest" structure
I review pull requests for three clients. The worst reviews are the ones that just say "this is wrong" or paste a wall of refactored code with no explanation. The best ones teach.
You're reviewing a pull request. The code [describe what the code does and what's wrong or unclear: "loops through an array twice when one pass would work" or "uses var instead of let in a modern codebase"].
Write a code review comment:
1. What (1 sentence): Point out the specific issue.
2. Why (1 sentence): Explain why it matters (performance, readability, future bugs).
3. Suggest (1–2 sentences): Give a concrete alternative or ask a clarifying question.
Tone: collegial, not condescending. Assume the author is smart and just missed something.
Why this works: Junior developers often write vague reviews ("this could be better"). Senior developers sometimes write harsh ones ("why would you do this?"). This template hits the middle: specific, educational, kind.
Kryotta tip: I use DeepSeek V3 for this. It's strong on code reasoning and explains why something matters without sounding like a textbook. If the review feels too long, I ask Claude Haiku 4.5 to "cut this to two sentences, keep the suggestion."
WhatsApp replies to customers: the "acknowledge + answer + next step" template
If you run a small business in India, you live in WhatsApp. Customer asks if you deliver to Raipur. Another wants to know if the kurta comes in size XXL. A third is angry because the courier is late.
Customer message: [paste the exact message]
Your business: [what you sell]
Situation: [new inquiry / repeat customer / complaint]
Reply in 2–3 sentences:
1. Acknowledge what they said (use their words if possible).
2. Answer the question or explain the situation.
3. Tell them what happens next or ask what they need.
Tone: friendly, helpful, no corporate speak. Write like you're texting a neighbor.
Why this works: WhatsApp isn't email. People expect fast, human replies. This template keeps the model from writing a paragraph when two lines will do. The "acknowledge" step is key—it shows you read the message, not just pasted a canned response.
Kryotta tip: Gemini Flash is my go-to here. It's fast enough that I can generate replies on my phone, and it handles Hinglish really well. If a customer writes "Bhaiya, ye kurta XL mein milega kya?", Gemini will match the tone without making it weird.
Meeting follow-ups: the "decisions + action items + no fluff" prompt
I sit through a lot of client calls. The follow-up email is always the same: remind everyone what we decided, who's doing what, when we're talking next. No recap of the entire conversation, no "it was great to connect."
Meeting: [topic and participants]
Decisions made: [list 1–3 key decisions]
Action items: [who's doing what, with deadlines if relevant]
Write a follow-up email:
- One sentence greeting.
- Bullet list of decisions (max 3).
- Bullet list of action items with owners.
- One sentence about the next meeting or check-in.
Under 120 words. No pleasantries beyond the greeting.
Why this works: Most meeting recaps are too long. People skim them, miss the action items, then ask you three days later what they're supposed to do. This template forces clarity: decisions up top, tasks below, nothing else.
Kryotta tip: I use Claude Sonnet 4.5 for this. It's excellent at pulling structure out of messy notes. I'll paste my raw notes from the call—half-sentences, typos, tangents—and Sonnet turns them into a clean list. If I need it faster, Llama 3.3 70B does a solid job too.
Social media captions: the "hook + body + call-to-action" formula
I write LinkedIn and Instagram captions for two clients. The ones that get comments follow the same pattern: start with a hook that makes someone stop scrolling, say one thing in the body, end with a question or soft CTA.
Post topic: [what you're sharing: a tip, a story, a product launch]
Platform: [LinkedIn / Instagram]
Audience: [be specific: "freelance designers" or "parents shopping for kids' clothes"]
Write a caption:
- First line: a hook (a surprising fact, a question, or a bold statement).
- Body (2–4 sentences): one clear idea, no tangents.
- Last line: a question that invites a reply, or a soft CTA ("link in bio" / "tag someone who needs this").
Tone: [conversational / inspiring / helpful]. Max 100 words for Instagram, 150 for LinkedIn.
Why this works: Most AI-generated captions sound like captions. They start with "Ever wondered..." or end with three emoji and "Double-tap if you agree!" This template keeps the model focused on one idea and one action, so it reads like something a human would actually post.
Kryotta tip: I test two models here. Mistral Large often nails the hook—it's good at surprising angles. Gemini Pro writes cleaner body copy. I'll generate both in the Compare arena, steal the hook from Mistral and the body from Gemini, then tweak the last line myself.
Content briefs for freelancers: the "audience + goal + constraints" template
I hire writers and designers. The best briefs I send them aren't long—they're specific. Who's the audience, what's the goal, what are the constraints.
Project: [blog post / landing page / email sequence / Instagram carousel]
Audience: [specific persona: "a Shopify store owner in tier-2 India with under 100 orders a month"]
Goal: [what should the reader do or understand after reading this?]
Constraints: [word count, tone, things to avoid, required keywords]
Write a creative brief in bullet points:
- Audience (2 sentences: who they are, what they care about)
- Goal (1 sentence)
- Key points to cover (3–4 bullets)
- Tone and style (1 sentence)
- Constraints (word count, deadline, anything else)
Why this works: Vague briefs waste everyone's time. "Write a blog post about email marketing" gets you generic SEO slop. This template forces you to name the audience and goal before you hand it off, so the freelancer knows exactly what success looks like.
Kryotta tip: I use Claude Sonnet 4.5 to turn my rough notes into a clean brief. I'll dump a paragraph of ideas, and Sonnet organizes it into bullets. Then I paste the brief into the project workspace so the writer and I can both see it—no lost emails, no "which version did we agree on?"
Questions people ask
Can I use these templates with any AI model, or do they only work in Kryotta?
They work anywhere—ChatGPT, Claude, Gemini, whatever you're using. The model-specific tips I mentioned (Haiku for emails, DeepSeek for code) are just what I've found works best. You'll get decent results from any halfway-capable model if the prompt is clear.
Do I need to customize these templates every time, or can I save them as-is?
Customize the bracketed parts—audience, tone, specifics—every time. The structure stays the same. I keep them in a note on my phone and fill in the blanks as I go. Takes fifteen seconds.
What if the output still sounds too robotic or formal?
Add one line to the prompt: "Write like a human, not a content bot. Use contractions, short sentences, and plain language." Or paste the output into a second model and ask it to rewrite in a more casual tone. I do that all the time.
Which of these templates saves the most time?
Email replies, by far. I used to spend ten minutes agonizing over tone for client emails. Now it's two minutes: fill in the template, generate, tweak one sentence, send. The others save time too, but email is the daily win.
These seven templates handle about 70% of my repetitive writing. I don't think about structure anymore—I fill in the blanks, generate, edit, done. If you're doing similar work and you're tired of staring at a blank text box, try one today. You can test any of these in Kryotta with multiple models side-by-side, see which one matches your voice, and build your own library of prompts that actually work.
