In this article
167 asked for, 62 kept, 105 refused
A first message is the first email and network note to a person at an account. Mach writes it after it researches the company and its people.
For a software services firm we work with, Mach wrote for 87 people from 68 research runs. It ran on October 15 and October 19, 2026.
Across those two days it was asked for 167 first messages. It kept 62 and refused 105. A refused message is never sent. Your rep sees why instead.
Count people, not drafts, and the picture is plainer. 46 of the 87 people ended with a ready message. 41 got none.
The two days were not alike. On October 15, 9 of 26 people got a ready message. On October 19, 37 of 61 did.
Most advice on AI-written sales email is about better drafts. This post counts the drafts that should not go out, and why.
Code can stop a message before a word is written. The writer can decline, with its own reason. Or code can refuse a finished draft.
69 drafts were written and then refused. 17 were stopped before writing. The writer declined 19 on its own.
| Reason | Messages | Where it stopped |
|---|---|---|
| Broke the writing rules after one fix | 64 | After writing |
| Same opening fact as an earlier message | 5 | After writing |
| Not the right time for this account | 9 | Before writing |
| The firm already has an open deal there | 4 | Before writing |
| Not a match for what the firm sells | 7 | 4 before writing, 3 by the writer |
| No honest opening | 9 | Declined by the writer |
| Already talking, or the same person twice | 4 | Declined by the writer |
| Several emails lately, no reply | 3 | Declined by the writer |
| All refused | 105 |
Each refusal stores its reason in plain words. That is what made this count possible. A tool that logs only sent mail cannot tell you what it chose not to send.
The rules every draft must pass
Every first message has a written contract. Code checks it after the writer is done. Here it is, in a form you can copy.
| Part | Rule | Why |
|---|---|---|
| Subject | 2 to 7 words, tied to the opening fact, no question mark | It reads like a note, not a campaign |
| Opening | One fact dated within six months, their own words first | It proves you looked, and anyone can check it |
| Insight | A problem seen in their evidence, never a truism | It earns the next line |
| Pitch | One result for companies like theirs, never a list of features | A list reads as a brochure |
| Proof | The closest case study, with its outcome | A claim with a customer behind it |
| Ask | One question answerable in one line, with room to say no | Easy to answer, easy to decline |
| Length | 60 to 150 words, no exclamation marks | Short enough to read on a phone |
| Claims | Each tied to evidence, with no web addresses | Every sentence can be traced |
| Dates | Written as dates, never as yesterday or last week | A message may be sent days later |
A draft that breaks a rule gets one fix. If what the buyer reads still breaks a rule, the draft is dropped. Nothing is cut or reworded to sneak it through.
A slip in a note to your rep never drops a draft. Code tidies those notes itself. Only what the buyer reads can sink a message.
Put it into practice
One draft, refused and then fixed
This example is made up, to show how the check reads. Tarang Labs, a software team in Pune, writes to Mara Ellis, chief operating officer of Alder Creek Bank.
Here is how a weak draft might open: "Loved your post yesterday about your new loan system!"
The check stops on three things at once. Yesterday will be wrong when the email goes out next week. The exclamation mark is out. And no dated fact stands behind the post.
The pitch then lists four services. That breaks the one-result rule. The proof names no customer, so it proves nothing.
The fix pass gets the draft and the list of problems. It must correct each one from the research, or leave the sentence out.
The fixed opening is dated, it is hers and anyone can check it. It reads: "On September 30 you wrote that Alder Creek Bank moves to a new loan system in March."
If the research held no dated words of hers, the fix could not invent them. The draft would open on a dated company event, or be refused.
Some refusals were our own checker's fault
Rule breaks after one fix happened 64 times: 20 on October 15 and 44 on October 19. Not all of it was the writer's fault.
October 19 ran in two batches. In the first, 23 messages were asked for. 3 were ready, and 20 broke the rules.
We read the refused drafts that day. Some were refused for nothing. The check read the verb in "you may want" as the month of May.
Others failed because the writer was handed things it was not allowed to use. It was offered signals from outside that run, and facts too old to cite.
We fixed both in the gap between the two batches. In the second, 85 messages were asked for. 48 were ready, and 24 broke the rules.
Later that day we changed one more rule. Now a ready draft that still breaks a buyer-facing rule gets a second rewrite before it is dropped.
A second rewrite costs another call. We pay it, because a dropped draft wastes the research behind it.
5 more drafts opened on the same fact as an earlier message to the same company. Two people at one firm should not get the same opening line, so those were refused too.
The lesson for anyone checking drafts: a refusal count tests your checker as much as your writer. Read a sample of refusals, not only the drafts that pass.
Stops before writing, and the writer's own no
On October 15, code stopped 17 messages up front. 9 accounts were judged not ready now, 4 had an open deal and 4 were not a match.
An open deal means someone on the firm's team is already selling there. A cold email on top of a live deal undercuts that person. The note says to ask them first.
We then changed the rule. Mach now writes to everyone who decides at the company, and where things stand with that company sets the angle. On October 19, no message was stopped before writing.
The angle matters. A booked meeting changes the message, not whether one exists. A draft that ignores a booked meeting is refused.
Stops still apply to a person who left, opted out or said no lately. They also apply after three of our touches in 30 days with no reply.
19 times, the writer read the research and declined. Each time it said why, and what would change its mind.
9 had no honest opening. There was no dated fact in the person's own words, or their role could not be confirmed. In one case, the linked profile belonged to someone else.
4 were people the firm was already talking to, or the same person twice. One duplicate record pointed at a founder with a meeting booked that day.
3 had received several emails lately with no reply. 3 more were companies outside what the firm sells.
Reading a person's posts first is how an honest opening is found, when one exists.
Where an honest opening comes from
No honest opening was the writer's commonest reason to decline. So it helps to know what counts as one. Mach climbs a ladder and takes the first rung that exists.
The rungs run from strongest to weakest. A message that reaches the last rung always goes to a person to check first.
| Rung | What it is | Example, made up |
|---|---|---|
| Their own words | A post or talk of theirs, dated within six months | On September 30 you wrote about your move to a new loan system |
| A company event | A dated change at their company, in their area | Your bank opened a new branch on October 2 |
| Common ground | Something you share that can be proven | We both spoke at the same meetup in September |
| Their role | Only the seat they hold, sent with a person's check | As the person who runs loan operations at your bank |
The last rung is weak on purpose. If your messages land there often, the research is thin, not the writer.
What saying no costs
Refusing has a price. 41 of the 87 people got no ready message from these runs. Your reps must write to them by hand, or wait.
The research behind a refused message stays with the account. A writer's decline also says what would reopen writing, such as a reply or a new dated post.
Ready still means a person checks it. Of the 62 ready messages, 56 carried a note asking for a check first. Often it was a surname, a title or a meeting date.
The ready share rose from 11 of 59 on October 15 to 51 of 108 on October 19. Fewer stops before writing and a fixed checker both played a part.
Ready does not mean it goes out alone. You approve each message before it is sent.
Your own refusal mix tells you what to fix. If most refusals are rule breaks, fix the writer or the checker. Read a sample to learn which.
If most are no honest opening, the research is thin. Read more of each person's own posts before writing.
If most are stops before writing, look at your company records. Open deals, owners and past touches decide those stops.
A pre-send check you can copy
If people or a tool write your first messages, run these checks before anything is sent. Refuse the message on any failed line.
| Ask | Refuse when |
|---|---|
| Is there a dated fact in their own words, or a dated company event? | Neither exists |
| Can every claim be traced to a source? | One claim has none |
| Does the message give a date, not yesterday or last week? | Time is counted from today |
| Does a colleague have an open deal or a meeting there? | Yes: ask them first |
| Has this person had three of your touches in 30 days with no reply? | Yes |
| Did someone else at the company get the same opening? | Yes |
| Is the pitch one result, not a list? | It lists features |
One mistake to avoid: judging a writer by how many drafts it turns out. A refusal with a reason beats a draft you must rewrite.
Common questions
Is refusing 105 of 167 messages a failure?
Not by itself. 69 were drafts that broke a rule or reused an opening. Fixing the writer raises the ready share. Loosening the rules would only hide the problem.
Where the numbers come from
Aggregates from one firm's stored first messages, written October 15 and 19, 2026. They are grouped by recorded reason and by time written. No draft or name is quoted. Exported October 21, 2026.
