How we scored 10,954 coaching websites
Every figure on this blog comes from one of two datasets we built ourselves. This page is what is in them, how they were made, and what they cannot tell you.
The websites
We started with a list of 12,294 coaching domains. 10,954 of them loaded with readable text on the homepage and those are the ones in every number we publish. 1,156 would not load at all. Another 184 loaded with nothing readable on them. We left all 1,340 out rather than scoring them zero, which would have flattered the average.
One page per site, the homepage, because that is the page a stranger arriving from a search actually meets. Nothing behind a click is scored.
The ten things we scored
Each one is scored 0 to 10 by a program reading the text of the homepage: the headline, the subheadings, the body copy, the buttons, any testimonial text, any price. The overall score is a weighted average, because a stranger deciding whether to book a call is not weighing the SSL certificate against the headline equally, and neither do we.
| What we check | Weight | Market average |
|---|---|---|
| Five-second read: can a stranger tell who it is for and what changes | 2.0 | 4.49 |
| Being specific: is the person and the problem named, or could it be anyone | 2.0 | 4.38 |
| Offer clarity: is there a defined thing to buy and a reason it is worth it | 1.5 | 2.89 |
| Proof: testimonials, named results, case studies, anything checkable | 1.5 | 1.51 |
| Lead capture: a way to stay in touch short of booking a call | 1.0 | 4.64 |
| Credibility: qualifications, affiliations, the things you say about yourself | 1.0 | 2.68 |
| Story: is there a human on the page, and does the copy reach the reader | 1.0 | 2.54 |
| Clear next step: one obvious action rather than none or seven | 0.5 | 5.60 |
| Technical health: loads, works on a phone, secure | 0.3 | 9.03 |
| Pricing shown: is there a number, a range, or anything at all | 0.2 | 1.77 |
Weights add up to 11. The weighted average is multiplied by 10 to give a score out of 100, and that is rounded to the score out of 10 you see in a post. A site scoring 80 or more is strong, 60 or more is decent, 40 or more is weak, below that is poor.
What that produced
The average live coaching website scores 3.65 out of 10. Of the 10,954: 13 are strong, 633 decent, 3,488 weak and 6,820 poor. Two thirds show no proof of any kind and 82.3% show no price. Technical health averages 9 out of 10, which is why we keep saying the build is not the problem.
The buying triggers
The second dataset is 117 coaching niches and 918 sub-markets inside them, built from 2,004 books those buyers actually paid for and 1,547 things real buyers wrote about their own situation. Each market is tagged for what drives the purchase, how aware the buyer is of her problem and of the solutions, how many times she has heard the claims before, and which forms of persuasion the evidence supports for that market.
Books and buyer quotes are the source because they are what people spent money and time on, rather than what they told a survey they wanted.
What none of this can tell you
- It is one page. Your about page, your booking flow and your first call are outside it.
- It is a program reading text, not a person forming an impression. It is consistent rather than sensitive, and on any single site a human would disagree with it somewhere.
- It is a snapshot. A site scored in 2026 may have been rebuilt since.
- It says nothing about whether a coach is any good. It measures what a stranger meets.
- Referrals, DMs, stages and reputation are all invisible to it, and plenty of coaches fill a practice on those alone.
How the numbers get into a post
They are recomputed from source rather than typed in. A script in the site's own repo holds every figure the blog asserts, recalculates each one from the scorecard file and the triggers dataset, and fails if a post and the data disagree. It has caught us out once already, which is the only reason it is worth having.
If you want to see the scoring run on your own homepage, the website X-ray uses the same ten checks against the same corpus.