In this article
The reading, measured
For one firm we work with, 294 person-research runs have finished. Each reads up to 20 of a person's recent public posts. The average person had 6.9.
| Measure | Value |
|---|---|
| Posts read per person (cap) | 20 |
| Posts read per person (average) | 6.9 |
| Of them about work (average) | 2.4 |
| People with a personal opener | 188 (64%) |
| People with no personal line | 106 (36%) |
Only 2.4 of those were about work: a hire, a launch, a tool choice, a talk. A repost with no comment counts for nothing.
Each run stores what it read and when. So any opener can answer one question: which post, from which date, says so?
An honest opener for 64% of people
The runs wrote an opener for 188 of 294 people. An opener is a first line built on something the person really said.
For the other 106, nothing honest was there, so Mach wrote no personal line. Their first message starts from a dated company event instead.
That one rule ends the fake 'loved your recent post' line for good.
What counts as read
A run reads recent public posts and the basics of the public profile: headline, role and place.
A work post is about their job. Mach uses work posts and reads past personal ones. A new hire is fair to mention; a holiday is not.
The opener is one of 196 kinds of text in Mach's writing rules. Each kind has length limits that code checks.
The two paths, on one made-up person
Turn the averages into one made-up person: Theo Bennett, head of loan operations at Alder Creek Bank in Ohio. Mach reads seven of his posts, and two are about work.
One is dated September 30. It asks how other banks kept loan applications moving during a switch to a new loan system.
The opener names that post, its date and one question it raises. Theo sees his own words.
A person with no work posts gets the other path. The first line uses a dated company event, such as a new loan system. It never pretends to a personal read.
Person research runs only for the people who decide at companies research said yes to. That order keeps the reading honest and the cost low.
We researched the company too much at first
At first, research spent most of its effort on the company: news, pages and filings. Those were easy to fetch and easy to sum up.
On 13 October we rebuilt research around a brief written by hand. Now about a fifth of the effort goes to the company. The rest goes to the people who decide.
The numbers above measure that second part, after the rebuild.
Rules you can copy
Read before you write, with a hard cap. Ours is 20 recent posts.
When nothing honest is there, write no personal line. Write to the person who decides, not to the company name. The people who decide have four roles.
We publish no reply rates yet. That data is too young to defend. When it can stand up, it gets its own post.
Common questions
Why stop at 20 posts?
Twenty is enough to find anything real and recent. It also keeps the cost and time the same for each person. Past 20, posts are usually too old to quote.
Where the numbers come from
Totals from the stored records of the person-research runs. One firm we work closely with. Exported 20 October 2026.
