How to automate prospect research without losing personalization
Automate the collection and summarising of research (company website, recent news, your own notes) and the first draft of the email. Keep the judgement human: approving the hook, the fit and the send. A good setup reads a lead list, gathers 2–3 sources per prospect, asks an AI to pick one verifiable hook and score ICP fit, and hands you a draft to review.
What "prospect research" actually involves
For a personalised first email you usually need four things: what the company does, something that changed recently, whether they match your ideal customer profile, and one specific detail to open with. Done by hand, that means opening several tabs per prospect. Time yourself on a few to see what it costs you; our ROI calculator turns that number into hours per month.
Step 1: Decide what to research (and what to skip)
- Research: homepage and about page, careers page (hiring signals), blog or newsroom (launches, expansion), and anything you noted when you found the lead.
- Skip: personal trivia (schools, hobbies) and anything you can't verify. It rarely helps and can feel intrusive.
Step 2: Pick your sources
| Source | Good for | How to automate |
|---|---|---|
| Company website | Offer, customers, positioning | HTTP fetch + HTML-to-text |
| News / web search | Recent triggers (launches, funding, hiring) | A search API such as Tavily |
| Your CRM or sheet notes | Context you already have | Pass the notes column into the prompt |
| Role, recent posts | Read it manually; automated scraping breaks LinkedIn's terms |
Step 3: Let AI summarise and choose a hook
Give the model the raw text and ask for three things: a 2–3 sentence summary, one hook with its source, and an ICP fit score with a reason. Tell it explicitly not to invent facts and to lower the score when the research is thin. That rule is the difference between useful personalisation and confident fiction.
Step 4: Draft, don't send
Have the system write the first line and a short sequence, then save them as drafts. Your review is quick because the research summary sits next to the draft, so you can confirm the hook is real.
Step 5: Track and improve
- Log status per lead (drafted, low fit, error) so nothing gets lost.
- Review the low-fit list weekly. It often shows your ICP definition needs work.
- Rewrite the prompt when you see repeated weak hooks.
Tools to do this
- No-code: a Custom GPT or Claude Project with web search, using a research-and-write instruction set (our free starter includes one).
- Automated: n8n, Make or Zapier connected to your sheet, an LLM API and Gmail. See our n8n walkthrough.
FAQ
Is automated prospect research accurate?
Only as accurate as its sources. Ground the AI in fetched text, require it to cite where each hook came from, and review drafts before sending.
Is it legal to research prospects automatically?
Reading public company websites is generally fine. Respect each site's terms (e.g. no LinkedIn scraping) and data-protection law such as GDPR when you store personal data. This is not legal advice.
The AI SDR is this workflow, tested and ready to import, with a setup guide (about 15 minutes on n8n Cloud). $19/month on Starter, 7-day free trial.