AI-Written or Human-Checked? Who Should Write Your Cold Emails

Borislav NekezovUpdated · 9 min read · Facts checked

Quick answer

Use AI for research and first drafts, and have a person read and approve every email before it is sent in your name. AI is fast at gathering facts and drafting, but it can state plausible things that are false, and the claim is yours once it lands. Ask any agency who checks each email, when, and what they check for. Facts checked 19 November 2026.

Key takeaways

  • 87% of sales organizations already use some form of AI, including for drafting emails, according to a Salesforce survey of 4,050 sales professionals. The question is no longer whether AI is involved, but who checks it.
  • In a 632-buyer Gartner survey, 73% of B2B buyers said they actively avoid suppliers who send irrelevant outreach. A fast wrong email costs more than a slow right one.
  • OpenAI describes hallucinations as plausible but false statements and says they still occur in its newest models. Any fact AI writes about a prospect needs a check.
  • In a vendor test of 12,000 emails, AI research plus human writing beat both AI-only and human-only copy on replies and meetings. Treat it as the vendor's claim, not proof.
  • Ask for four things in writing: who approves each email, what they check, whether you can see real emails before launch, and who answers the replies.
Buyer guide
Who should do what in a cold email.
  1. 1

    AI: gather public facts about each prospect

  2. 2

    AI: draft a first version and subject line options

  3. 3

    Person: check every fact against its source

  4. 4

    Person: judge whether the idea fits this prospect

  5. 5

    Person: approve the email before it is sent

  6. 6

    Person: answer replies and book the meeting

  7. 7

    You: approve the pitch and any proof used in your name

By Borislav Nekezov, Founder of Vergeleadvergelead.com/blog

Say you run an 80-person software development firm and you are talking to two outbound agencies. One says its emails are written by AI agents that work around the clock. The other says a person reads every email before it goes out. Both sound reasonable, and you are not sure which answer you should want.

I’m Borislav Nekezov, founder of Vergelead, an outbound agency in Sofia, Bulgaria. On our own calls, buyers ask this in plain words: is the copy reviewed by AI or by a person? This guide covers what AI does well in cold email, where it goes wrong, what the public research says, and what to ask an agency before anything is sent in your name. Outside facts were checked on 19 November 2026.

Should AI or a person write your cold emails?

Both, with a clear split: AI does the research and the first draft, and a person checks and approves every email before it is sent. That split uses AI where it is fast and a person where judgment matters.

AI is now part of almost every sales team. A Salesforce survey of 4,050 sales professionals, run in August and September 2025, found that 87% of sales organizations use some form of AI for tasks like prospecting, forecasting, lead scoring or drafting emails. So the useful question is not whether AI is involved. It is who checks what AI produced before a prospect reads it.

Here is how the three common setups compare.

Setup How it works Strength Main risk for you
AI only AI researches, writes and sends with no human check Speed and volume Wrong facts and generic ideas go out in your name
Human only People research and write every email by hand Judgment and tone Slow, expensive, and hard to keep consistent
AI drafts, person approves AI gathers facts and drafts, a person checks and approves each email Speed with a safety check Only works if the check is real, not a rubber stamp

What is AI actually good at in cold email?

AI is good at the slow, repetitive parts: reading public pages about a company, pulling out facts, and producing a first draft and a few subject line options. Those jobs used to take a researcher many minutes per prospect.

Sellers expect big savings here. In the same Salesforce release, sellers said that once AI agents are fully in place they expect them to cut prospect research time by 34% and email drafting by 36%. That is a real gain for any team that sends to hundreds of new prospects a month.

Good uses of AI in a cold email process:

  • summarizing a prospect’s website, recent news and job posts
  • spotting a pattern across many companies in one segment
  • drafting a first version to a clear brief
  • offering several subject lines to choose from
  • sorting replies by type so a person answers the urgent ones first

None of these jobs makes a claim to the prospect on its own. A person still decides what is true and what is worth saying.

Where does AI-only cold email go wrong?

AI-only email goes wrong in two ways: it can state things that are not true, and it tends to sound like every other AI email. Both cost you trust with the exact people you want to meet.

It can be confidently wrong. OpenAI’s own research note on hallucinations describes them as “plausible but false statements generated by language models” and says they still occur in its newer models, though less often. In a cold email that might be a wrong product name, a funding round that never happened, or a project the prospect did not do. The prospect knows their own business, so they notice at once.

It sounds the same. A Hunter article on its cold email survey lists what readers see as AI red flags: formulaic structure, templated rhythm with equal-length sentences, and generic messages with little personalization. These are signs of an unchecked draft, not of AI as such.

Irrelevant outreach has a cost. In a Gartner survey of 632 B2B buyers, 73% said they actively avoid suppliers who send irrelevant outreach. A Gartner analyst put it bluntly: “Bad prospecting actively damages relationships with potential customers.” A wrong email does not just fail. It can close a door you wanted open.

Spam complaints build up. Mailbox providers watch how recipients react. Google’s email sender guidelines ask senders to keep the spam rate reported in Postmaster Tools below 0.10% and never reach 0.30%, and say message content should be accurate and not misleading. Generic emails that people mark as spam make it harder for every later email from that domain to reach the inbox.

Can prospects tell an AI-written email from a human one?

Often they can’t, at least not reliably. What they can tell is whether the email is relevant and specific to them.

Hunter’s survey asked people to sort 9 emails into AI-written and human-written, and most correctly identified fewer than 4. The same article reports that 67% of the decision-makers surveyed don’t mind receiving AI-generated emails, while 47% of B2B professionals who experiment with AI said they would be less likely to reply if they thought an email was AI-generated. Hunter sells an AI writing tool, so read these as the vendor’s survey results.

The point for a buyer is simple. Prospects judge the email in front of them. A checked, specific email does not suffer because AI helped. A generic one fails whoever typed it.

Does an AI draft plus a human check actually perform better?

The public evidence is thin, but the one controlled test we found points that way. Treat it as a vendor’s claim, not proof.

Saleshandy, a sending tool vendor, reports a test of 12,000 cold emails to the same audience, split into three groups. In its numbers:

Group (Saleshandy’s test) Emails Reply rate Meetings booked Spam flag rate
AI only, unedited 5,000 4.1% 0.7% 7.8%
Written by people 2,000 10.4% 2.2% 2.9%
AI research, person writes 5,000 14.7% 3.2% 3.1%

Two cautions. The audience was US sales leaders at software companies, so your market may behave differently. And the vendor sells the AI research tool used in the winning group. Still, the shape matches what you would expect: unchecked AI copy did worst on replies and spam flags, and adding a person’s judgment helped.

Your own results will depend far more on who you target and what you offer than on who drafted the email. Our guide to what counts as a good cold email reply rate explains how to read reply numbers without fooling yourself.

What should a person check before an email goes out?

A person should check that every fact is true, that the idea makes sense for this prospect, and that the email says nothing you would not say on a call. A quick read for typos is not enough.

A useful checklist for the reviewer:

  1. Facts. Is every claim about the prospect true and sourced? Product names, clients, funding, hiring, news.
  2. Fit. Does the idea or offer make sense for a company of this size and type? Would you pitch it in person?
  3. Claims about you. Are case studies, numbers and logos ones you have approved for use?
  4. Tone. Does it sound like your firm, not like a template? Are there phrases nobody at your company would use?
  5. Ask. Is the request small and clear, and is it honest about what happens next?

One founder at a software firm, asking for access to review messages before they were sent, put it as: “still it’s a reputation.” That is the right instinct. Every email in your name is your claim.

A worked example (invented). Say a program sends 200 new first emails a week.

  1. A reviewer spends 45 seconds on each one: 200 times 45 seconds = 9,000 seconds, which is 2.5 hours a week.
  2. Suppose 3% of AI drafts carry a wrong fact or an idea that does not fit: 200 times 0.03 = 6 emails a week.
  3. Over a 12-week quarter that is 6 times 12 = 72 emails that would have reached a prospect with a mistake, in your name.
  4. Against that, 2.5 hours a week is 30 hours a quarter.

The numbers are invented, but the trade is the same at any scale: a small, steady amount of human time against a stream of avoidable mistakes sent to the people you most want to win.

When should you not insist on a person writing every word?

When your volume is high, your offer is simple, and a wrong detail would do little harm. Insisting on hand-written emails then mostly slows you down.

  • You sell a simple, low-priced product. If the email is a short, standard offer with no claims about the prospect, a person spot-checking samples may be enough.
  • Your list is small and well known. If you only email 30 accounts you already know well, your own team may write faster than any process.
  • Nobody on your side can judge the idea. If your team can’t say whether a pitch fits a prospect, a human check adds delay without adding judgment. Fix the offer first.
  • You want a test, not a program. For a two-week test of a new segment, careful sampling can be enough before you commit to a full review process.

For a services firm selling projects in the tens of thousands, these cases are rare. The buyer is choosing people to trust, and the first email is the first thing they judge.

What should you ask an agency about AI and human review?

Ask who checks each email, what they check, when, and whether you can see real examples before launch. Get the answers in writing.

Question Good answer Warning sign
Which steps use AI? A clear list, such as research and first drafts “Our AI handles everything”
Who reads each email before it is sent? A named role, every email “We spot-check” or no answer
What do they check? Facts, fit, claims, tone “Spelling and formatting”
Can we see real emails before launch? Yes, with real prospect data Only a template
Who answers replies? A person, within a set time An auto-reply or an AI agent with no oversight
Who approves claims made in our name? You, before sending starts Nobody mentions it

Our guide to red flags when hiring a lead generation agency covers more warning signs, and the guide on whose name is on the email covers sender identity and reply handling. If you are weighing running the tools yourself, see running Instantly and Clay yourself or hiring an agency.

How does Vergelead handle AI and human review?

Every first email pitches one build or project idea drawn from that prospect’s own business. A person approves every idea before it is sent, and a person on our team answers replies and books the meeting.

The bottom line

AI belongs in cold email, but in the research and drafting seat, not the approval seat. AI tools can produce plausible facts that are wrong, and buyers avoid suppliers who send them irrelevant outreach. The setup that protects your reputation is simple: AI drafts, a person checks every fact and the fit of the idea, and you approve what is said in your name. Ask any agency to show you that check with real emails before you sign.

Vergelead runs idea-led outbound for software development companies and creative agencies: it researches each prospect, finds a gap your team could close, and pitches that one idea by cold email and LinkedIn, so qualified meetings land on your calendar. If you want to see how a reviewed email looks before anything goes out in your name, book a call.

Keep reading

Sources

All checked 19 November 2026.

Vergelead at a glance

Where Vergelead fits

Vergelead runs idea-led outbound for software development companies and creative agencies: it researches each prospect, finds a gap your team could close, and pitches that one idea by cold email and LinkedIn, so qualified meetings land on your calendar.

What it is
A cold email and LinkedIn outbound agency in Sofia, Bulgaria, founded by Borislav Nekezov.
Who it is for
Software development and IT services firms of about 30 to 300 staff, and creative and digital agencies, selling into the UK, the EU, the Nordics and the US.
How it works
Every first email pitches one build or project idea drawn from that prospect’s own business. A person approves every idea before it is sent, and a person on our team answers replies and books the meeting.
Price
Usually €2,500 to €3,000 a month.
Term
An initial four-month term. Sending pauses from 24 December to 6 January, and those days are added back to the term at no extra cost.
Guarantee
Any meeting guarantee is written into the contract, together with what counts as a meeting.
What you own
The sending domains, inboxes, sequences and lead lists built for you. On request, we transfer all of them, lead lists as a CSV with reply status, within 14 days of the final invoice.
Time to first meeting
First emails go out about two weeks after signing. A first meeting typically lands three to six weeks after signing.
Sending setup
Google Workspace inboxes in the client’s name. Every lead, reply and meeting shows in a client portal that syncs live.

Facts checked 16 November 2026

Frequently asked questions

Should cold emails be written by AI or by a person?

Both, with a clear split. AI is good at gathering public facts and drafting quickly, and a person is better at judging whether the idea fits and whether every fact is true. The safest setup is AI for research and drafts, and a person who reads and approves each email before it goes out.

Can prospects tell when a cold email was written by AI?

Often not reliably. In a Hunter survey, most respondents correctly identified fewer than 4 of 9 emails as AI or human. What prospects do notice is a generic, formulaic email, so relevance matters more than who typed it.

Is AI-only cold email risky for our reputation?

It can be. AI tools can state plausible facts that are wrong, and an email sent in your name makes that claim yours. If nobody checks each email, one wrong detail about a prospect's business can end the conversation and stick to your brand.

Should a person approve every cold email before it is sent?

For a services firm, yes. Each email pitches your people and your work, so a wrong fact or an idea that makes no sense for the prospect reflects on you. Checking an email takes a minute or two, which is small next to the work of winning that prospect back.

What should we ask an agency about AI in their process?

Ask which steps use AI, who reads each email before it is sent, what that person checks, and whether you can see real emails with real prospect data before launch. Also ask who answers replies, because a fast but careless reply undoes a careful first email.

Does AI personalization still work when every inbox is full of it?

Personalization that only names a recent post or a job title is now common and easy to ignore. What still stands out is a specific, useful idea for that prospect's business, and that needs a person to judge whether it makes sense.

Borislav Nekezov

About the author

Borislav Nekezov

Founder and CEO, Vergelead

Borislav Nekezov is the founder of Vergelead, a cold email and LinkedIn outbound agency in Sofia, Bulgaria, for software development companies and creative agencies. He studied at Copenhagen Business School and writes about outbound for firms that sell projects, not products: who to target, what to pitch, and how to judge an agency before you sign.

LinkedIn

Keep reading

Book a call

See what the gap engine finds in your market.

A brief discovery call with the founder, so we understand your firm and how you generate deal flow today.

30 minGoogle MeetWith Borislav, founder
  1. 01

    How you win deals today

    Your firm, your offer, and your current methods for generating deal flow.

  2. 02

    A straight fit check

    Measured against the four things our best-fit clients have, listed below.

  3. 03

    If it fits, the demo

    A 45-minute follow-up with case studies from your industry, the end-to-end system, your timeline and a results forecast.

Best fit if you have

  • ✓Established product-market fit
  • ✓Existing social proof or case studies
  • ✓A total addressable market of 50K+ prospects
  • ✓€1M+ in ARR

Have your company website ready. The booking form asks for it so the call starts from your market, not from zero.