Runvax
Back to blog
19 July 20266 min read

How to Use AI to Write Cold Emails That Get Replies (2026 Guide)

A practical process for using ChatGPT, Claude, or a purpose-built tool to write cold emails that get replies — including the exact prompts and why generic AI output fails.

AI can write cold emails that get replies — but only if you feed it specific details about the prospect first. Generic prompts ("write a cold email selling web design") produce generic emails, and generic emails are exactly what recipients have trained themselves to ignore. The fix is a two-step process: gather 2-3 real data points about the business, then prompt the AI to write around those specifics.

Here's the exact workflow, the prompts that work, and why "just use ChatGPT" alone isn't a strategy.

Why raw AI output gets ignored

Feed any AI model a bare prompt like "write a cold email for a web design agency" and it will produce something technically correct and completely forgettable — vague opening line, generic value proposition, weak call to action. The AI isn't the problem. The input is.

2026 research on B2B outbound backs this up directly: personalized subject lines lift reply rates by 30.5% compared to generic ones (Martal Group, 2025). The average cold outbound reply rate in 2026 sits at 6-9%, with top performers hitting 14-18% — and the gap between those two numbers is almost entirely explained by how specific and relevant the message is to the recipient, not by clever copywriting tricks.

AI's real value isn't writing skill. It's speed at applying a good structure to specific inputs, repeatedly, without you manually typing 30 variations.

Step 1: Gather the inputs before you prompt

Before opening ChatGPT or Claude, collect these for each prospect:

  • Business name and what they actually do (not just the industry — "family-run Italian restaurant, been open 12 years" beats "restaurant")
  • The specific gap — no website, an outdated site, no online booking, no Google reviews response
  • One local or timely detail — a recent expansion, a seasonal angle, a competitor who does have a strong online presence

This is the step most people skip, and it's the entire difference between a 3% and a 15% reply rate. A tool like Runvax pulls the first two automatically when it flags a business as having no website — which removes the research step entirely and leaves you only the prompt.

Step 2: The prompt structure that works

Use this structure with ChatGPT, Claude, or any capable model:

"Write a cold email to [business name], a [business type] in [city]. They currently have no website / an outdated website [specify]. I'm a [freelancer/agency] offering [service]. Keep it under 90 words. Open with something specific to their business, not a generic greeting. State one clear problem their lack of a website creates for them. End with a low-friction call to action — asking for 10 minutes, not asking them to buy. Do not use the phrase 'I hope this email finds you well' or any generic opener. Write like a real person, not a marketing template."

The instructions matter as much as the input data. Telling the model explicitly to avoid stock phrases and cap the length prevents the two most common AI cold email failures: bloated length and robotic openers.

Step 3: Generate 3 variations, not 1

Don't accept the first draft. Ask the model to produce three versions with different opening lines, then pick the one that reads most naturally for that specific business. This takes 30 extra seconds and consistently produces a better result than editing a single draft.

Step 4: Edit for voice, not for facts

AI-drafted emails need a human pass before sending — not to fact-check (you already gave it accurate inputs) but to make sure it sounds like you. Read it out loud. If a sentence sounds like something you'd never actually say, rewrite it in your own words. This 60-second edit is what separates outreach that reads as AI-written from outreach that reads as personal.

A before/after example

Bad prompt output (generic input, no editing):

"Hi there, I hope this email finds you well. I noticed your business could benefit from a professional website. We specialize in creating stunning websites for local businesses. Would you be interested in learning more?"

Good prompt output (specific input, structured prompt):

"Hi Maria — saw Bella's Kitchen has been serving the same block on 4th Street for over a decade with zero online presence, while three newer spots nearby are pulling in reservations through their websites. That's foot traffic walking past you toward the competition. I build simple, fast sites for restaurants like yours — usually live within a week. Worth a 10-minute call this week?"

The second version is 15 words shorter and infinitely more likely to get a reply, because every sentence is specific to that one business.

AI cold email tools compared

| Tool | Good for | Limitation | |---|---|---| | ChatGPT / Claude (manual prompting) | Full control, works for any use case | You manually gather inputs and copy-paste for each prospect | | Lavender | Email scoring and rewriting inside Gmail/Outlook | Built for volume B2B outbound, not local business targeting | | Runvax's built-in generator | Local business outreach — pulls "no website" data automatically into the prompt | Focused specifically on local/small business prospecting |

Automating this without losing the personal touch

Once you've nailed the prompt structure for your niche, the temptation is to batch-generate 100 emails and blast them out. Resist the urge to skip the specificity step even at volume — a fully automated pipeline that skips real business details is how "AI-written" starts to read as an insult instead of a compliment. We cover how to scale personalization without going robotic in automating cold outreach without sounding robotic.

This is also just one piece of the outbound puzzle — see the best AI tools for lead generation in 2026 for how AI writing fits alongside prospecting and follow-up, and best AI sales prospecting tools in 2026 for the tools that find the prospects you're writing to in the first place. For the deeper mechanics of cold email structure beyond AI, cold outreach in 2026: the complete guide covers subject lines, timing, and follow-up cadence.

Where Runvax fits

Runvax combines the prospecting and writing steps into one action: search a city and industry, see which businesses have no website, and click "Generate Outreach" to get an email already populated with that business's specific details — no manual data-gathering required.

Try Runvax free and generate your first batch of personalized cold emails in minutes, not hours.