Generate cold emails

Runs on Cloudflare Workers AI. Free: 10 generations an hour, 25 a day per visitor, and a shared daily allowance for everyone. Nothing you enter is stored.

How it works

what the generator is told to do
OPENabout the reader, never “my name is” or the companyRULE
SIZE50–110 words · 2–4 short paragraphs · grade 5–8RULE
WORDSno leverage / synergy / streamline / cutting-edge · no exclamation marksRULE
ASKone question they can answer in a few wordsRULE
OUTthree variants, three angles, merge tags intactDRAFTS
A draft to edit, not a campaign to ship.
01

Describe the offer and the reader

What you sell, who it is for, and the pain or trigger that makes now the moment. A proof point helps — a result, a number, a similar customer.

02

Pick tone and ask

Direct, friendly or formal, and whether the close is an interest check, a permission ask or a meeting request. The first two out-reply the third; the option is there anyway.

03

Generate three angles

The model writes three emails that differ in how they open — an observation, a result, a pointed question — so you have a real A/B/C test rather than one email and two rewrites.

04

Edit, check, test

Copy a variant, cut anything that is not true for your reader, then run the readability and spam-word checkers and put the survivors in the A/B test calculator.

Why cold email needs different rules from copywriting

A landing page has a visitor who chose to be there. A cold email has a reader who did not, on a phone, with a thumb hovering over delete. So the things that make marketing copy good — a strong brand voice, a full value proposition, a confident headline — make cold email worse. What works is closer to a good text message from a stranger who clearly knows something about you: short, specific, easy to answer. Every generic AI writer produces the first kind by default because that is most of what it was trained on. This one is instructed against it, line by line, using the rules I use when I edit sequences for clients.

Getting a good draft out of it

The output is only as specific as the input. “We help companies with email” produces the same email every SDR has already sent. “We set up secondary domains and warm 50 inboxes a week for agencies whose reply rate dropped after Google's 2024 rules” produces something a reader recognises. Put the trigger in — a hire, a funding round, a tool change, a new office — and put a real number in the proof field. Use merge tags for anything that will differ per recipient; the model keeps them and does not invent details around them.

How the generator is tuned

The model behind this page is an open-weight model running on Cloudflare's network, and the instructions it follows are mine — the same rules I apply when I edit a client's sequence: short sentences, no marketing words, open with the reader, one question at the end, keep merge tags intact. Those instructions are edited over time as I see what it gets wrong, so the output improves without the page changing. It is a starting draft, not a finished campaign: read it as you would a junior writer's work, cut what does not fit, and run the result through the readability checker and spam word checker before it goes into a sequence.

Limits and privacy

Each visitor gets ten generations an hour and twenty-five a day, and there is a shared daily allowance across the whole site that resets at 00:00 UTC — when it runs out the tool says so rather than failing silently. Inputs are sent to the model and discarded; nothing is stored, logged with your text, or used to train anything. The generated text is yours.

Frequently asked questions

Which variant should I send?

Test them. Split the list randomly, send all three at the same time, and let the A/B test calculator tell you which one is really better once the sample is large enough.

Can I give it my own examples or voice?

Put a sentence or two in the notes field — phrases to use, phrases to avoid, a link you always mention. Full custom voice training is on the roadmap.

Why no meeting request in the first email?

Because a specific calendar ask forces the reader to commit before they know whether they care. An interest check gets a one-word yes, and the meeting follows in the reply. You can still choose the meeting option if your process needs it.

Is my offer text stored or used for training?

No. It goes to the model for the one request and is not logged with your text or kept anywhere.

Why did it hit a limit?

Ten generations an hour and twenty-five a day per visitor, plus a shared daily allowance for the whole site. Limits reset at 00:00 UTC.

Last reviewed

Related tools

What to run next

The checks that most often follow this one.

AI generators

More in this category

Read more

Guides that go deeper

Services

When the tools tell you something is wrong

The diagnostics here are free and always will be. When the fix is bigger than a DNS record, this is the work I do.

Get in touch

Start with a call

Bring a domain and the symptom. I will tell you what is actually wrong and whether you need me at all — plenty of people leave that call able to fix it themselves.

Thirty minutes, no pitch

We will run the checks together on your actual domains, and you will leave knowing what is broken, what it takes to fix, and what it should cost. If that is a job you can do in-house, I will say so.

Based inRangpur, Bangladesh — all time zones
RepliesWithin one business day
LicensingWorkspace below list price
Back to top