In Q1 2026 we ran an audit for a B2B infrastructure company that had spent 8 months rebuilding their marketing site, and found that over 6 months their documentation subdomain had influenced more closed revenue than every marketing page combined. Not more traffic. More revenue. The marketing site had roughly 4 times the sessions.
The client's own team had suspected it. Nobody had measured it, because the docs sat outside the marketing analytics property and therefore outside every report anyone read.
What we found when we looked
The audit was meant to be routine. We were checking whether the new site had recovered its pre-migration rankings in Google Search Console, which it mostly had. As part of that we added the documentation subdomain to the property, largely for completeness, and the numbers that came back changed the engagement.
Documentation pages were attracting search traffic on the exact queries the marketing team had been trying to win for a year, and doing it with pages nobody in marketing had ever seen. More importantly, when we joined that against closed-won opportunities in HubSpot over 6 months, the accounts that had touched docs before a first sales conversation closed at a materially higher rate than those that had not.
The mechanism is not mysterious once you see it. Technical buyers do not read positioning. They go looking for whether the thing does what they need, and the only page on most company sites that answers that honestly is written by the engineering team for people who already bought.
We checked whether this was one unusual account. Across 5 client properties over 4 months we repeated the join between documentation sessions and closed-won opportunities. In 4 of the 5 the docs surface influenced a disproportionate share of revenue relative to its traffic, and in every one of those 4 the docs were excluded from the marketing analytics property. The exception was a company selling to a non-technical buyer, where the pattern reversed exactly as you would expect.
What we did not understand at the time
We had been treating documentation as a post-sale asset, and so had the client. It sat with engineering, it was excluded from the marketing analytics property, and it had never once appeared in a report I had written.
In retrospect that exclusion was not an oversight, it was a category error with an org chart behind it. Docs belonged to engineering, marketing owned the funnel, and the boundary between those two teams had become a boundary in our measurement without anyone deciding it should be. The buyer, obviously, does not experience that boundary at all.
There is a second thing I got wrong, which is more uncomfortable. When we started the engagement I had characterised the documentation as a risk, because it was inconsistent in tone and did not follow the brand guidelines. I was not wrong that it was inconsistent. I was wrong about what that inconsistency signalled to the reader, which was that a person who understood the system had written it rather than someone whose job was to make it sound good.
What changed
Three things, in order of how much they mattered.
First, the documentation subdomain went into the same analytics property and the same monthly report as everything else. That is a small technical change and it reframed every subsequent conversation, because for the first time the client could see a single picture of how buyers actually moved through their surfaces.
Second, we stopped trying to make the docs more marketing-like and started making the marketing pages more docs-like. Specifically, on 14 pages we replaced benefit statements with the concrete answer to the question the page title implied, put it in the first 2 sentences, and attached a real number or a code sample where one existed. Engagement on those pages improved but the more useful outcome was that sales stopped having to correct expectations set by the page.
Third, and this is the organisational change that made it durable, we got one person from engineering into the monthly content review. Not to write, just to say whether a claim was true. That single presence killed more bad copy than any process we could have designed, and it did it before publication rather than after a customer noticed.
That last one was harder to get agreement on than either of the others, because it costs an engineer an hour a month and the benefit is invisible if it works. The argument that landed was not about content quality. It was that sales had been spending time in first calls correcting expectations the website had set, and an hour of review upstream was cheaper than that correction repeated across every deal.
What I would tell a technology leader
Three things.
The first is to check where your documentation sits in your measurement. If it is on a different subdomain, owned by a different team, and excluded from the property your marketing reports read, you almost certainly have a blind spot in the exact place your most qualified buyers are spending their time. It costs an afternoon to find out.
The second is that the org chart tends to become the measurement boundary, and the buyer does not care about your org chart. Any surface owned by a team that does not attend the growth conversation will be systematically undervalued, not because anyone decided it was unimportant but because nobody is bringing its numbers to the meeting.
The third is about what technical audiences respond to, and I would state it more strongly than I would have a year ago. The thing that made those documentation pages effective was not that they were technical. It was that they were checkable. A sentence a reader can verify is worth more than a sentence engineered to be persuasive, and this is increasingly true beyond technical audiences as more research happens through systems that quote sources rather than rank pages. We now apply that standard to every page we build, which is covered in our answer engine optimization playbook.
The client did not need a better marketing site. They had an excellent asset that nobody in the growth conversation could see, and 8 months of rebuild had gone into the surface that was already working least hard.