The opening line of a cold email decides whether the rest gets read. It is also the line that takes longest to write well, because it has to be about the reader. That makes it the obvious place to bring in AI, and the place where AI is misused most.
This post is about where a generated opener belongs, where a person has to write, and how to run both without embarrassing yourself.
What an opener is for
An opener answers the question every recipient asks in the first second: why me.
A good one proves the email was meant for this person. It points to their role, their company, or a problem that comes with their position. A bad one is either generic ("I hope you are well") or falsely specific ("I loved your recent post" when no post was read).
The second kind does more damage. A generic opener gets ignored. A fake one gets noticed, and it tells the reader the sender is willing to pretend.
What the model has to work with
An AI opener can only be as specific as its input. This is the most important fact about the whole subject, and the easiest to lose sight of.
Fuzegenai's AI openers work from three fields on the lead record: the person's name, their company and their title. They do not browse the prospect's website, read their posts or look up company news.
That sets the limits. From a name, a company and a title, a model can write a line that is relevant to the role and mentions the company naturally. It can vary the phrasing, so a thousand emails do not share one first sentence. It cannot know what the prospect announced last week, which tools they use, or that you met their colleague at an event.
If a generated opener seems to know those things, it made them up. This is the central risk of AI openers. Language models produce plausible detail easily, and a plausible invention about a real company, sent under your name to someone who works there, is the quickest way to lose that account.
Where AI openers belong
Generated openers earn their place in the broad middle of a list: many accounts, similar fit, none large enough to justify research time.
Scale. A rep can research a handful of accounts properly in a day. A list of a few thousand will never get that attention. For those contacts the real choice is not between an AI opener and a researched one. It is between an AI opener and a merge-field template.
Consistency. Human openers drift. The first twenty are sharp, the next two hundred are tired. A model holds the same tone and length from the first lead to the last.
Role relevance. The title carries real information. A head of support, a finance director and a founder care about different things, and an opener that frames the problem for the role is an improvement on one line for everybody.
Variation. Each email gets its own first line instead of an identical sentence repeated across the send.
Where a human must write
Some emails should never open with a generated line.
Accounts that deserve research. If winning the account would matter to your quarter, it is worth fifteen minutes of reading. The opener should show that, with a detail only someone who looked would know.
Anything that references a fact. A funding round, a new hire, a product launch, a mutual contact, a quote from an interview. These have to be true, and a model working from three fields cannot verify them. A person has to.
Senior recipients. Executives receive a great deal of cold email and spot a pattern quickly. A short, plain line written by a person is the safer choice here.
Sensitive contexts. Regulated industries, companies in the news for bad reasons, anyone with whom a clumsy line carries real cost.
A tiered approach
A simple way to organise this is three tiers.
- Top accounts: fully human. Researched opener, tailored body, sent individually or in a very small campaign.
- Core list: human body, AI opener. A person writes the template for the segment, including the problem, the proof and the ask. The model supplies a first line per lead from name, company and title. A person reviews a sample before launch.
- Long tail: consider not sending. If a contact is not worth either a generated or a written opener, ask whether they belong on the list. Low-fit contacts are where complaints tend to come from.
Where a lead sits depends on what the account is worth, not on how much time you happen to have that week.
Rules for running AI openers
Clean the input first. A model given "VP, Sales & Mktg (EMEA) - Interim" will try to use it. Normalise titles and company names before generating. Strip legal suffixes such as "Inc." and "GmbH" so the line reads the way a person would say it. Importing leads from a CSV covers how to map the fields.
Do not let it claim anything it cannot know. No "I saw", "I read", "I noticed", "congratulations on". With three fields of input, every such line is an invention.
Avoid flattery. Compliments from strangers read as a technique. A plain statement of relevance is better than praise.
Keep it to one short sentence. The opener is a bridge into your message, not the message itself.
Read a sample before every launch. Twenty or thirty lines, picked from across the list and not just the top. You are looking for invented facts, awkward use of the title, and lines that would make you wince if the recipient read them aloud.
Plan for missing data. Some leads will have no title, or a company name that is really a domain. Decide in advance what those leads get: a neutral fallback line or a place in a different campaign. Never let a gap produce a broken sentence.
The body is still your job
An opener gets the second sentence read, and that is the end of its job. Whether the email gets a reply depends on what comes next: a problem the reader has, a reason to believe you can help, and an ask that is easy to answer. Those are decisions about your market and your offer, and no model working from three fields can make them for you.
The same applies to the list. A relevant opener on an email to the wrong person is still an email to the wrong person. The anatomy of a cold email sequence covers the rest of the message.
The takeaway
Use AI openers for what they are good at: a role-aware, varied, consistent first line across a list too large to research by hand. Write by hand for accounts that deserve it, and for every sentence that states a fact. Whoever writes the line, one rule covers both cases: never claim to know something about the reader that you do not know.