Most AI cold emails fail for the same reason: the model has nothing real to say about the recipient, so it invents flattery. The fix is simple. Feed the generator facts from the company's own website and forbid it from making things up.
What makes a cold email feel personal
A good first line shows you looked. Not "I love your website" but a specific, checkable observation: what the company focuses on, which platform it runs on, or a gap you can see from the outside. Specific beats flattering every time.
The data you need
For each prospect, collect a short company description, the top two or three topics from the site, the technologies it uses, and one opportunity signal such as no CRM detected. That is enough for two or three sentences of genuine personalization.
A template that works
Keep it under 110 words. One observation, one relevant pain point, one sentence about what you do, and a low-friction question at the end. No attachments, no long pitch.
Automating the data collection
Disclosure: I built the AI B2B Lead Intelligence Actor for this. You give it a URL and it returns the keywords, tech stack, contact details, opportunity signals and a draft email with a subject line. By default the email comes from a template filled with the extracted signals. You can optionally add your own Anthropic API key to have an LLM rewrite it, with the instruction to use only the extracted facts.
Rules for using generated emails responsibly
• Read every draft before sending. Automation should prepare the email, not replace your judgment.
• Do not invent claims. If the data does not say it, the email should not either.
• Verify addresses and honor opt-outs. Comply with CAN-SPAM, GDPR and local rules.
• Send small batches. Low volume with relevance protects your domain reputation.
Measure what matters
Track reply rate, not open rate. Test two subject lines against each other across the same segment and keep the winner. If replies are low, the problem is usually targeting, not wording.
Bottom line
An automated cold email generator is only as good as the data behind it. Start from real website signals, keep a human in the loop, and your outreach will read like research rather than spam.
Disclosure: I built the AI B2B Lead Intelligence Actor on Apify. It generates these emails from real extracted data. You can try it for free here: https://apify.com/kazaz_organization/my-actor-1
Top comments (0)