Across 16 websites over twelve months, organic search delivered 9.1% of the sessions and 86.4% of the commercial actions.
By commercial actions we mean forms submitted, numbers called, quotes requested. We measure those because traffic is the input and intent is the outcome, and a report that leads with sessions is reporting the input and calling it a result.
Here is the whole thing.
The channel table
Portfolio-wide, 6 October 2025 to 5 October 2026. Intent means a lead or a direct contact attempt.
| Channel | Sessions | % of traffic | Intent actions | % of intent | Per 100 sessions |
|---|---|---|---|---|---|
| Organic Search | 5,075 | 9.1% | 393 | 86.4% | 7.74 |
| Direct | 49,127 | 88.0% | 47 | 10.3% | 0.10 |
| AI Assistant | 254 | 0.5% | 5 | 1.1% | 1.97 |
| Referral | 913 | 1.6% | 2 | 0.4% | 0.22 |
| 81 | 0.1% | 2 | 0.4% | 2.47 | |
| Unassigned | 182 | 0.3% | 6 | 1.3% | 3.30 |
| Total | 55,718 | 455 | 0.82 |
Read the two percentage columns against each other and the whole argument is in one row. Organic carries a tenth of the traffic and nearly nine tenths of the outcome.
The Direct number is not what it looks like
88% of sessions from Direct shouldn’t be believed, and we’re not asking you to believe it.
When we checked engagement, Direct ran at roughly 19% engaged sessions with a 9 second average, against organic at 55% and 160 seconds. A quarter of the portfolio’s sessions arrive from cloud datacentre regions, and on two sites it approaches half.
So Direct is inflated by automation. That matters twice over. It makes your traffic look better than it is, and it drags every blended average down with it, which is how a site that converts perfectly well ends up with a conversion rate of 0.3%.
If your Direct channel is large, low-engagement and short-duration, you’re reporting bots to yourself.
The spread between sites is the part nobody publishes
Same measure, organic sessions only, so the comparison is fair. Client sites are unnamed.
| Site | Organic sessions | Intent per 100 |
|---|---|---|
| A local plumbing site | 278 | 20.14 |
| A dental directory | 1,402 | 15.34 |
| A healthcare directory | 876 | 11.87 |
| margen.net, our own site | 2,243 | 0.53 |
We’ve got the most organic traffic in the portfolio and the worst conversion rate in it, by a factor of 38.
We’re publishing that rather than quietly leaving ourselves out of the table. It’s what happens when you compare businesses that aren’t comparable. Someone whose boiler has failed decides in minutes. Someone choosing a retainer at £2,950 to £12,950 a month takes weeks, visits several times, talks to two other agencies and converts on a call that no single-session metric will ever attribute correctly.
Which is the actual lesson: conversion rate benchmarks do not survive a change of business model. Anyone quoting you an industry average conversion rate is quoting you a number assembled from businesses that aren’t yours.
What we got wrong first time
We’d built our portfolio reporting on sessions, like everyone else, and it hid the only thing that mattered.
The moment it became obvious was a fortnight on one directory where organic sessions were flat, 194 against 187, and enquiries went from 4 to 10. Conversion went 2.06% to 5.35%. Judged on traffic the fortnight was nothing at all. It was the best result the site had produced.
Every metric that mattered was invisible in the traffic report, and we had been sending traffic reports.
How to do this on your own site
Map your events to tiers. Sites name things differently, so pick three: a completed action, a
direct contact attempt, and a started-but-abandoned. One site fires generate_lead, another fires
phone_click, and both mean a human tried to buy something. Without a shared mapping you cannot
compare anything.
Calculate intent per 100 sessions per channel. Not conversion rate across the site. Per channel, because the whole point is that channels differ by two orders of magnitude.
Put the two percentage columns side by side. Share of traffic against share of intent. Any channel where intent exceeds traffic is underfunded. Any channel where traffic exceeds intent is flattering your dashboard.
Then switch your reporting to engaged sessions, so the automation stops dragging the averages down.
It takes an afternoon. In our case it moved the argument from “organic is about a tenth of our traffic” to “organic is essentially all of our commercial outcome”, which is a different conversation with a finance director.
Related
This sits in a series where we publish our own numbers, including the ones that do not flatter us: 6,482 page-one rankings that produced 218 clicks, an experiment we ruined by treating all ten pages at once, the audit of our own estate that measured one search engine, the method we use to tell AI-generated queries from typed ones, and a free tool that converts at 13% and nobody sees.
Common questions
What is a good conversion rate for organic search traffic?
It depends so heavily on what you sell that a single benchmark is misleading.
Across 16 sites we measure, organic search produced 7.74 intent actions per 100 sessions on average, but the spread ran from 20.14 on a local trades site down to 0.53 on our own agency site. An emergency plumber and a £2,950-a-month retainer are not the same purchase.
Why does Direct traffic convert so badly?
Usually because a large share of it is not people. Across our portfolio Direct accounted for 88% of sessions and 10.3% of intent actions, at 0.10 per 100.
Direct ran at roughly 19% engaged with an average of 9 seconds, against organic at 55% and 160 seconds. That gap is automation, and it inflates your session count while contributing nothing.
Should I report sessions or conversions?
Conversions, or whatever counts as a committed action on your site: a form submitted, a number called, a quote requested.
Sessions are the input. If your reporting leads with traffic you can have a quarter where sessions are flat, enquiries double, and the report says nothing happened.
How do you measure intent across different websites?
We map each site’s events onto three shared tiers. Lead means someone completed the action the site exists to produce. Contact means they tried to reach you directly. Engaged means they started and did not finish.
Without a shared mapping, cross-site comparison is impossible, because one site fires
generate_lead and another fires phone_click for the same human act.
Does AI assistant traffic convert?
In our data yes, at roughly twice the portfolio average, on very small volume. Across 12 months AI assistant referrals produced 254 sessions at 1.97 intent actions per 100, against a portfolio average of 0.82.
On one directory the rate was far higher, but on 11 sessions, which is too small to mean anything on its own.
Why is your own agency site the worst converter in your portfolio?
Because we sell a considered purchase and the sites that beat us sell an urgent one. Our organic traffic converts at 0.53 per 100 against 20.14 for a local trades site we manage.
Someone whose boiler has failed decides in minutes. Someone choosing a retainer takes weeks and several visits, and most of that never shows up as a single-session conversion.