What we tested
Over 12 months (June 2025 – May 2026), we tracked campaigns across our customer base comparing:
- AI-generated emails with full prospect research (company context, recent news, role analysis, tech stack, signals) - AI-generated emails with minimal context (name, company, job title — the typical merge-tag approach) - Human-written emails by experienced SDRs
All campaigns targeted the same ICP segments, used the same sending infrastructure, and ran for at least 4 weeks. We controlled for deliverability by using the same domain health standards across all groups.
The results (and why they surprised us)
AI-generated emails with full prospect research outperformed human-written emails by 19%. We expected parity at best.
The reason: consistency. Human SDRs produce excellent emails when they're focused and motivated — and mediocre emails when they're tired, rushing, or on their 40th prospect of the day. AI with research produces consistently good emails on prospect 1 and prospect 400.
But AI without research was dramatically worse — 1.4% vs 3.2% for humans. The 'AI minimal' emails read like templates with company names inserted. Recipients can smell this, and they ignore it.
The takeaway: the variable that matters isn't who writes the email — it's whether the writer (human or AI) has genuine context about the recipient. AI just makes per-prospect research economically viable at scale. A human can't research 400 prospects per day. An AI can.
What AI gets right
When given good research context, AI excels at:
1. Connecting the dots: 'Your company just opened a Phoenix office and you're hiring QC engineers — that usually means injection molding volumes are outpacing manual inspection.' A human could make this connection, but it takes 15 minutes of research. AI does it in seconds. 2. Tone matching: given examples of your voice, AI produces consistent tone across hundreds of emails. No variance between your best SDR and your newest hire. 3. Avoiding AI-tells: modern LLMs, properly prompted, avoid the 'I hope this email finds you well' clichés. The AI-tells that remain are about specificity (or lack of it), not grammar. 4. Variant generation: need 5 angles for the same prospect? AI generates them in seconds. An SDR takes 30 minutes.
What AI still gets wrong
Even with research context, AI struggles with:
1. Emotional judgment: knowing when NOT to reach out. A prospect's company just had layoffs — is now the right time? AI doesn't have the social radar for this. 2. Relationship context: if you met this prospect at a conference last month, the email should reference that. AI doesn't know your relationship history unless you tell it. 3. Industry nuance: AI can learn that 'SPD' means 'sterile processing department' in healthcare, but it takes explicit training. Without it, the email reads like an outsider. 4. Knowing when to break the rules: sometimes the right email is a 2-word subject line and a single sentence. AI tends to over-explain.
This is why the best model is humans-in-the-loop: AI drafts with research, humans review with judgment. The human catches the cases where AI's recommendation is technically correct but socially wrong.
Frequently asked questions
Can recipients tell if a cold email is written by AI?
Yes — if the AI is writing from a template without research context. No — if the AI references specific details about the recipient's company, role, or recent activity. The tell isn't the grammar; it's the specificity.
Should I use AI or write cold emails manually?
Use AI with research. The question isn't AI vs manual — it's 'how much context does the writer (human or AI) have about the recipient?' An AI with deep prospect research outperforms a human writing from a template. A human with 20 minutes of research outperforms an AI with no context.
What AI tools are best for cold email?
General-purpose LLMs (Claude, GPT-4) produce good email copy when given good context. The bottleneck isn't the writing model — it's the research pipeline that feeds it. Tools like RocketSDR integrate research and writing so the AI has prospect context automatically.