Draft personalized outbound that doesn't read like a template

Per-prospect outreach grounded in one specific, true, recent fact about each person, drafted by AI in seconds and approved by a human before it sends, so your cold emails read like you actually looked.

What you'll have when you're done

A workflow that takes a list of named prospects and drafts a short, genuinely personalized opener for each, built on a real detail (a funding round, a new hire, a podcast quote) rather than "I came across your company." You review each for accuracy, then send through your normal sequencer. The result is outbound that earns replies because it reads like a human did the homework, with a hard human-approval gate so the AI never sends on its own.

Generic outreach is dead, and "Hi {FirstName}" is generic

Cold outreach reply rates have been falling for years, and the reason is simple: everyone mail-merges, so inboxes are full of obvious templates that get deleted on sight. As one operator put it, a personalized cold email to someone who does not care is still spam. Personalization alone is not the edge. Relevance is, reaching the right person with a real reason. I have been on both ends of this. I have sent the "Hi {FirstName}, I came across your company" template and watched it return nothing, and I have received the version where someone clearly read a thing I wrote and responded to it, those I answer, every time, even when I am not in market. The asymmetry is the whole opportunity: almost nobody does the homework, so the homework is the edge.

This is where AI genuinely helps, and where it genuinely hurts if you point it wrong. Pointed right, it does the per-prospect research that used to be too slow to do at scale: it finds the one true, recent fact that makes an email feel earned. Pointed wrong, it hallucinates "facts" about prospects and blasts volume that torches your sending domain. So this workflow has two hard rules: verify every fact, and a human approves every send.

What you need first

• A list of named prospects worth real effort (this is for considered outreach, not spray-and-pray).
• AI with live web access: Claude with a web-search connector or an MCP tool, or a research tool like Perplexity.
• Your sequencer/CRM (HubSpot, Salesforce, Instantly) for sending, and a healthy sending setup.
• A Claude Project loaded with your offer and your voice.

Step-by-step

Step 1Give it the list and the research job

Hand the model your prospect list and ask for one specific, recent, true fact per person:

For each prospect below, find ONE specific, recent, verifiable fact (funding,
a new hire, a product launch, a podcast quote, a post they wrote). Return the
fact with its source link. If you can't find something specific and real,
say "no strong hook" and do not invent anything.

The "no strong hook" escape hatch is what stops it from making things up to fill the row.

Step 2Verify the facts before anything is written

This is the step that protects you. Skim the facts and their source links. A hallucinated detail in a cold email ("congrats on the Series B!" when there was no Series B) is worse than no email, it brands you as careless or fake. Cut anything you cannot confirm. Treat the "no strong hook" prospects as a separate, lower-effort batch.

Here is the trap, illustrative, because it is rarely a clean hallucination:

AI returned: "Sarah Chen, VP Product at Acme, just closed a $40M Series B (source: a real, linked article)." You click the link: the article is real and the round is real, but it is for a different Acme (Acme Robotics, not the Acme Sarah works at). The model pattern-matched on the name.

Send that as drafted and your opener congratulates a VP on a round her company never raised. That is not a typo; it is the exact tell that screams "a bot wrote this and nobody read it." The whole point of the source link is that you can catch this in one click. The check costs you a second. Skipping it costs the prospect and dents the reputation of every email leaving your domain that week.

Step 3Draft the opener on the verified fact

Now ask for the email, built on the confirmed hook:

Using only the verified fact for each prospect, draft a cold email under 80 words:
one line referencing the specific fact, one line connecting it to [my offer],
one clear low-friction ask. My voice: direct, no fluff, no "I hope this finds you well."

Short and specific beats long and templated. Under 80 words forces it. Here is the shape, illustrative, built on a verified hook:

Subject: your take on usage-based pricing

Saw your post arguing seat-based pricing punishes the customers who adopt fastest. We hit that exact wall last year and moved our analytics tool to usage-based; churn on power users dropped noticeably.

We help teams make that switch without the billing chaos. Worth 15 minutes to compare notes?

Notice it never says "I hope this finds you well," never name-drops a feature list, and the personalization is load-bearing rather than decorative: cut the first line and the email collapses, which is the test of whether a hook was real.

Step 4A human approves every send

Read each draft before it goes. You are checking two things: is the fact right, and does it sound like a person? The AI never sends on its own, and you never bulk-blast unreviewed drafts. Beyond the quality reason, auto-blasting AI-generated volume is how you trip spam filters and burn the domain you send from. In practice this is a fast batch review, not a slog: put the verified fact and the draft side by side, and expect to kill or rewrite roughly one in five, the ones where the hook is technically true but boring, or the draft drifted into stiffness. That 20% you cut is exactly the 20% that would have made the whole batch look automated.

Step 5Follow up like a human

Reply rates climb with a real follow-up cadence, not a single shot. Have the AI draft a short sequence (a few touches over a couple of weeks), each adding something, not just "bumping this." A good second touch shares a relevant resource ("we wrote up how we ran that migration, here's the link"); a good third offers a graceful out ("if this isn't a priority, no worries, I'll stop here") which, counterintuitively, often gets the reply. And respond fast when someone replies, speed to a warm reply is one of the highest-leverage habits in outbound.

How you'll know it's working

Reply rates go up, not because you sent more, but because the ones you sent were relevant and real. You also stop getting the "this is obviously automated" non-responses. The tell that you have the balance right: prospects reply to the specific thing you referenced, which only happens when the hook was genuine.

When it breaks

• It invented a fact. Tighten the "say no strong hook, do not invent" rule, and never skip Step 2's verification. This is the cardinal risk.
• Emails read templated anyway. The hook is too generic ("saw your company is growing"). Push for a specific fact, and keep it under 80 words.
• Your deliverability tanks. You are sending too much, too fast, unreviewed. Slow down, warm the domain, keep the human gate. AI volume without deliverability discipline is self-defeating.
• Replies go cold. No follow-up cadence, or you replied slowly. Draft the sequence up front and respond to warm replies fast.
• The hook is real but stale. A six-month-old funding round read as "just raised" is worse than no hook; it shows you scraped, not read. Add a recency window to Step 1 ("nothing older than 60 days") and prefer the freshest true detail.
• Every opener has the same sentence shape. The model templated the personalization itself ("Saw that you..." every time). Tell it to vary the opener form, a question, a shared observation, a direct reference, so the batch does not read as one mail-merge with the nouns swapped.

Make it yours. Match the effort to the tier. For a short list of dream accounts, spend the full research-and-verify cycle and write each by hand off the AI's draft. For a broader mid-tier list, a verified single-fact hook per prospect is the right level. The discipline that never changes regardless of tier: the human approval gate and the verification step. The moment you drop those to "save time at scale" is the moment you start torching your domain.

Where this fits in your harness

Outbound shares its engine with the rest of your go-to-market stack: the same Claude Project that holds your voice for content holds it for outreach. The Monday market scan often surfaces the very triggers (a competitor's customer churned, a prospect raised) that make good outbound hooks. And when a prospect becomes a happy customer, that relationship becomes a case study.