Cold Email Personalization That Isn't Fake

The fastest way to ruin a decent list is a first line that is almost true. In comments under the "dirty secret" AI SDR thread, people describe models inventing lines like "I saw you recently expanded to Europe." Adugbo Victory has gone after the wider habit of wiring a prompt to a list and calling it a motion. thatGTMdude's public notes keep coming back to a simpler habit: open with the signal you actually saw.

Pramod Kumar, co-founder of Kingdom of Kumar. On X

·

That is the standard I want for your domain. Personalization is a fact a person would recognise, placed in the first line, or it is nothing — a short, plain email that respects their time. Fake personalization is worse than none, because it trains the reader to distrust you before the offer arrives.

The approval path is in An AI SDR approval workflow. The copilot build is in How to build an AI SDR workflow. Signals are in Types of buying signals. The surrounding work is on GTM engineer and GTM engineering. If the product is not ready to sell, take the App Launch Score and do not decorate a pitch that should not go out.

"I saw you recently expanded to Europe"

I have seen versions of this line, and close cousins: a city they do not work in, a product they do not use, a hire they did not make, a podcast they were not on. Sometimes the model guessed from a stale page. Sometimes it combined two other companies. Sometimes a scraper put the wrong firmographic in the field, and the prompt treated the field as scripture.

The repair is not a better adjective. The repair is evidence, then approval.

  • If you cannot point at a source, the first line may not mention the fact.
  • If the source is old, the first line should not speak as if it happened this week.
  • If you would not bet a small amount of your own money that the fact is true, cut it.

A plain alternative: "I work with [role] on [problem]. If that is not you, tell me and I will close this." That line is not clever. It is hard to call a liar.

"A prompt wired to a list and called a motion"

A motion has a fit rule, a reason a person is in the queue, a first line policy, an approval step, a reply path, and a review. A prompt on a list has a hope that unique sentences will excuse a vague list. They will not.

If you want personalization at all, do the unglamorous work first: fewer accounts, better fit, one evidence field you actually fill, then a draft. The prompt comes last, and it is allowed to write only from that field plus the offer. It is not allowed to "research freely" on the live send path.

"Open with the signal you actually saw"

This is the whole craft, in one instruction.

Write the first line as a restatement of a fact you would be comfortable reading back to them, plus, if needed, a short clause about why you are writing. Do not interpret their motives. Do not compliment their "impressive growth." Do not mention a signal that is outside the window you wrote on your signals page.

Examples of the shape, not of magic copy:

  • "You recently posted that you are hiring a [role] to work on [problem]. We help teams in that seat with [job]. If this is badly timed, I can stop."
  • "We spoke last spring about [topic], and I have not assumed you are still in the same search. If you are, I can send a short update; if not, I will leave you alone."

I am not giving you a swipe file of guaranteed winners. I am showing the constraint: visible fact, modest offer, easy out. If a draft cannot live inside that constraint, reject it.

A checklist before any first line leaves

  • [ ] I can point at the evidence in one click
  • [ ] The person would recognise the fact
  • [ ] The fact is inside the timing window
  • [ ] I did not invent a motive
  • [ ] The offer in the rest of the mail is true
  • [ ] A generic version of this mail would still be acceptable if the fact were removed
  • [ ] I would send this to a person I respect

That last tick is not scientific. It is how you catch the ones that are technically accurate and still slimy.

How to use a model without letting it invent

Give it only the evidence field, the offer, the role, and the constraint: "Use only this evidence. If it is thin, write a generic first line. Do not add geographies, funding, headcount, or tools I did not provide." Ask it to list the claims it made. Compare the list to the evidence before you approve.

If a vendor hides that loop, do the loop in a sheet and treat the vendor as a sending path, or do not use it.

Volume will drop. Good. Fake first lines at volume are how domains and reputations get expensive.

Two kinds of "custom" I would retire

The first is scrape-and-compliment: a line about their funding, their stack, or their news, lifted from a database, dressed as if you had been paying attention. Even when the fact is true, the pose is false. If the fact matters, say it plainly as the reason you are writing, not as proof you are a fan.

The second is story-personalization: a first paragraph about their journey, their leadership, their "chapter." Almost none of that belongs in a first cold email. It is hard to verify, easy for a model to generate, and uncomfortable to receive from a stranger. Cut the story. Keep the fact, the problem, and the out.

If a freelancer or a tool is selling "hyper-personalized sequences" and the samples look like those two kinds, ask them to show the evidence field for each sample. If they cannot, you are buying prose, not a process.

When you should not personalize at all

If the evidence is a profile view, a vague intent score, or "someone at the company downloaded a whitepaper" you cannot tie to this person, do not pretend you have a signal. Send a short relevant note, or do not send. Personalization is not a requirement for courtesy. It is a privilege of having seen something real.

FAQ

Is a generic first line a failure?

No. A generic, relevant, short email to a fitting person is honest. Save the custom line for when you have a fact.

What about complimenting a blog post or a talk?

Only if you read or watched it, and only if the compliment leads to a relevant reason for writing. "Loved your post" from a model that did not read it is the Europe line in a friendlier coat.

Can you do this at $500 a month?

Kingdom of Kumar builds and runs a scoped system on Grok and Claude for $500 a month, three-month minimum. You own it, and you approve what goes out, including first lines. We will not sell you personalization as a substitute for fit. This is not autopilot, and it is not a $135,000 employee.

Contact

hello@work.kingdomofkumar.com