Personalise cold emails with a single LLM call on a scraped page, and use a soft 'reply YES' ask
A custom-agent studio explains why one LLM call with the full scraped website beat a multi-step chain for cold email personalisation, and why changing the call to action mattered more than the personalisation.
Evidence: The author reports this. We have not checked it beyond reading the source.
The business problem
Founders doing outbound need personalised first lines but cannot afford slow, expensive pipelines that invent details about prospects.
What was tried
Each prospect's website is scraped with a self-hosted Firecrawl and the cleaned text goes to a local model in a single call that extracts three or four facts worth mentioning and drafts the email. A multi-step chain (scrape, summarise, plan an angle, draft, self-review) was slower, cost more in tokens and drifted more from the source. The recommended setup is a hybrid: buy a platform for sending, warm-up and sequencing, and build only the personalisation step. Replacing a 'book a call' link with a soft 'reply YES' ask raised reply rates more than better personalisation did.
What was reported (mixed)
The post reports qualitative findings only: the single-call design gave tighter and more accurate emails for less money, and the soft call to action consistently lifted replies. No reply rates, sample sizes or test lengths are given.
Limitations
The authors sell custom AI agent work and the advice leans toward building. There are no figures for the reply-rate lift, the test method or the model used, and the page does not show a year. Scraping can fail when sites block bots or have little content, token cost grows at volume, and US CAN-SPAM rules apply equally to AI-drafted email. Costs are given only as a few thousand dollars of build time and cents per lead.
What you need
A scraper (a self-hosted Firecrawl is used), access to a language model, prospect data, and a sending platform such as Instantly or Smartlead. Pricing is not stated.
Sources
- DEV Community (Pykero) ↗ Firsthand write-up, publication date unknown
Source published: unknown. Last reviewed here: October 11, 2026. Spot a mistake? Tell us.
Tools in this workflow
- Firecrawl: Scrapes prospect websitesTry Firecrawl
- Apollo: Prospect dataTry Apollo
- Instantly: Sending and sequencing platformTry Instantly
Related workflows
Research sponsors, build a sponsorship CRM and schedule checks with the Manus agent
A content creator used Manus to find companies advertising in AI newsletters, build a web-app CRM for YouTube sponsorships, and schedule recurring lead and trend checks.
Find content gaps with a Claude Code agent and then cross-check against your live archive
A creator ran a published content gap agent with Perplexity and Firecrawl, got a scored gap report, then discovered her local files missed 37 published posts and the real problem was distribution.
Build a Claude Code subagent that writes SEO and answer-engine research briefs with Perplexity and Firecrawl
A step-by-step build of a Claude Code subagent that, given a post or topic, gathers search trends and search results and writes a research brief with keywords, snippet opportunities, competitor gaps and a suggested title and slug.
Send 500 AI-written cold emails and iterate the prompt with 40-prospect tests, for sponsorship sales
A paid newsletter post says a prompt reverse-engineered from replied-to emails produced 167 replies and 43 meetings from 500 sponsorship cold emails, with each prompt version tested on 40 prospects.