The first line of a cold email does one job: prove you are not a machine that found a domain. Everything else — the offer, the proof, the ask — is downstream of that proof. If line one fails, the merchant has already decided, and the rest of your message is being skimmed for a reason to delete rather than a reason to reply.
This matters more than it used to because the first line is also visible in the inbox preview on most clients. It is doing double duty as an open-rate lever and a credibility lever. Which means the standard opener — the one you have written a thousand times — is quietly costing you twice.
The openers that mark you as a stranger
Delete these from your templates permanently. “Hi, my name is X and I run Y” — it is in your signature, and nobody asked. “I hope this finds you well.” “I know you’re busy.” “I’m reaching out because…” “Following up on my previous email.” Each of these is a sentence that transfers zero information and announces that a script is running.
The subtler one, and the one most people still believe is personalization: “I saw you’re the founder of {{company}} and that you sell skincare.” That is not personalization. That is a merge tag, it is what every other sender is doing this week, and the merchant recognizes the shape of it instantly. A variable in a sentence does not make the sentence specific. Specificity comes from having noticed something the merchant did not tell you.
The test is simple and brutal: could this sentence be sent unchanged to five hundred other stores? If yes, it is not an opener, it is padding. Cut it and start with the observation.
Relevance beats personalization
There is a distinction worth holding onto. Personalization is inserting facts about the recipient. Relevance is choosing a recipient for whom the message is inherently true. Relevance wins, and it wins by a lot, because a relevant message with one variable outperforms a heavily personalized message aimed at someone with no reason to care.
In practice this means the filter stack does most of the work the first line gets credit for. If you filtered to stores that rank for thousands of organic keywords but whose organic traffic has been flat for eight months, then the sentence “you rank for a lot and it stopped converting into visits about eight months ago” is simultaneously personal and true of everyone on your list. You wrote it once. It reads as though you wrote it for them.
That is the whole trick, and it is why the previous article in this series was about filters and not about writing. Good targeting turns one sentence into three hundred personalized sentences at no marginal cost.
A signal inventory built from store fields
Here is what a store record actually gives you to open with, ranked roughly by how hard it is to fake. The trend series — roughly a hundred and seventy monthly buckets of organic traffic and positions — gives you inflections: a drop that started in a specific month, a recovery that stalled, a slow bleed nobody inside the company has framed as a number yet. This is the strongest opener material available, because it is a fact about their business they may not have articulated themselves.
Then: the apps and tech stack, both what is present and what is conspicuously absent. Payment methods, which tell you which markets they are set up for and which they are not. Shipping carriers, which tell you the same about logistics. Product count against traffic, which tells you whether they have a merchandising problem or an acquisition problem. Theme, currency and language, which tell you how much of the store is default and how much is deliberate. AI mentions against cited pages — models discussing the brand without ever linking to it is a gap almost nobody has thought about yet.
And the Google Business block, on the 44,132 stores that have a matched profile: rating, review count, whether the listing is claimed, photo count, category, city. An unclaimed profile on a real business is one of the most reliably answered openers in this space, because the merchant either did not know or has been meaning to deal with it for two years.
Five lines, one idea, one ask
A structure that holds up across verticals, in five short lines. Line one, context: the specific observation, stated flatly, no adjectives. Line two, question: a question that qualifies them and simultaneously points at the pain your service addresses. Line three, reason: why you are asking, backed by something real — a pattern you have seen across similar stores, a mechanism, a piece of arithmetic. Line four, tease: the existence of a solution, not the solution itself. Line five, a soft ask.
Worked through for an ecommerce target: “Your organic traffic has been flat since March despite ranking for around four thousand keywords.” / “Is that intentional, or has nobody had time to look at which of those rankings sit on page two?” / “I ask because on stores with that ratio it is usually thirty to sixty keywords one position away from real traffic, and it is the cheapest fix available.” / “I pulled the list for your domain, it took about ten minutes.” / “Want it?”
Fifty to a hundred words, total. One idea, one ask. If you have to scroll on a phone to reach the end, cut. The process that produces this: write it, leave it for a day, come back and delete somewhere between a fifth and half of what you wrote, then read it out loud. If you would not reply to it, the merchant will not either.
Subject lines: one word, lowercase, no merge tag
Three words maximum, one is better, and no capital letter. “question”. “traffic”. “audit”. “idea”. The reasoning is that the subject lines which get opened are the ones that look least like prospecting — a colleague or a supplier writing to you does not compose a headline, they type one word. The historically famous example of this is a political campaign whose highest-performing subject line was the word “hi”.
Merge tags in the subject used to help and now hurt. “video for {{company}}” reads as a mail merge to anyone who has received more than three cold emails, which is every ecommerce founder alive. Bare “video” reads as ambiguous and gets opened. The one exception worth testing: naming a specific competitor as the entire subject line. It is the rare case where the merge tag creates genuine urgency rather than pattern recognition.
Whatever you put in the subject must match the body. If the subject is “audit”, there had better be an audit. Curiosity that is not paid off in the first two lines converts an open into a permanent deletion habit.
The ask is not a meeting
Asking a stranger for thirty minutes in the first email is the most common unforced error in outbound. You are asking someone who does not know you to give up the scarcest thing they have, in exchange for the possibility that you are useful. The answer is no, and it is no for structural reasons that no amount of rewriting will fix.
Distinguish three kinds of ask. A direct ask requests an action — “open to a quick call?”. An interest ask tests appetite without cost — “want me to send it over?”. A conversation ask opens a dialogue — “how are you handling that at the moment?”. The first message should almost always use the second or third. Seven words or fewer, on its own line, phrased so that answering is easier than not answering.
When they say yes to a call, do not send a booking link. You have their email and they have told you they are free Thursday afternoon — put it in the calendar and send the invitation. Sending a link converts a warm yes into another task on their list, and tasks on lists do not get done.
Stack two contexts, and use the postscript
A single observation reads as research. Two unrelated observations in the same message read as attention, and attention is what earns the reply. Pair a traffic-trend fact with an operational one: the organic plateau plus the fact they are running a review app but no post-purchase flow. Pair a catalog fact with a local one: nine hundred SKUs and a Google Business listing sitting at 3.9 stars with unanswered reviews.
Put the second one in a postscript. Everyone reads the postscript — it is the most reliably consumed line in the entire message, and it is the natural home for the observation that would have made the body too long. It is also where a soft opt-out belongs if you use one, and where a piece of humour can go without undermining the seriousness of the body.
One caution learned expensively by others: a follow-up that delivers value but contains no ask gets close to zero replies. Generosity without a question is a broadcast. Every message in the sequence, including the one where you hand over something for free, ends with something the merchant can answer in four words.
The verifiability rule
If you are generating openers with a language model — and you should be, at three hundred stores it is the only sane approach — impose one hard constraint: every personalized clause must cite the field it came from. Not in the email, in the generation step. The model outputs the line and the field, you spot-check the pairing, and anything the model could not source gets dropped.
Without this rule you will eventually send a merchant a compliment about a product they discontinued, or a traffic observation that is off by a factor of ten. One hallucinated fact costs you more than fifty generic emails, because it converts “this person did not do their homework” into “this person is lying to me”.
The corollary is that your best openers will feel almost rude in their plainness. No adjectives, no flattery, just the number and what it implies. Merchants get complimented all day by people who want something. Almost nobody tells them a specific, checkable, slightly uncomfortable fact about their own store. Be the one who does.