Labellerr builds data annotation and computer vision tooling for teams training AI models. They had published 300–400 posts and almost none of it was doing commercial work: the blog ranked for curiosity, the product pages ranked for nothing, and the site generated no leads. The engagement had a three-month window, which ruled out the obvious answer of publishing more.
AI-first software: data annotation and computer vision platform
ML engineers, data ops leads, AI product teams
300–400 published blog posts
Growth-stage B2B SaaS
Generate qualified leads from organic within one quarter
Labellerr had done the hard part. Three to four hundred posts on annotation workflows, model training and computer vision: genuinely technical material written for a genuinely technical reader. The problem was that every one of them was written for the top of the funnel.
An ML engineer who reads a post about annotation quality has no reason to end up on a pricing page, because nothing connected the two. Meanwhile the commercial terms that describe what Labellerr actually sells (data annotation platform, data labelling platform) were owned by competitors.
Underneath that sat the usual accumulated debt: broken links, slow pages, missing metadata, and a set of posts quietly competing with each other for the same queries.
Deep technical content with nothing behind it. No comparison pages, no use-case pages, no bottom-of-funnel material for a buyer who had finished researching and wanted to evaluate.
Broken links, slow page speed and missing metadata at scale. On a site with 400 URLs, small per-page problems become a site-wide signal.
Multiple posts targeting the same queries with no hierarchy, and a site layout that gave the commercial pages no prominence and no internal authority.
Stage 1 examines five dimensions before anything is recommended. Here’s what each one revealed.
With three months on the clock, publishing was the wrong instrument. New content takes three to six months to rank on a healthy site; on this one it would have added cannibalisation to an estate that already had it.
The diagnosis found something more useful: a large body of content that had never been triaged. Sorting 400 posts by buyer stage, then building the missing bottom-of-funnel layer and pointing the existing library at it, was work that could show signal inside the window.
The commercial terms were the other half. “Data annotation platform” and “data labelling platform” describe exactly what Labellerr sells, and nothing on the site was seriously targeting either.
Anything published in month one would still have been finding its footing at the end of the engagement. Restructuring shows signal in four to eight weeks.
Technical buyers evaluate hard and late. The pages that serve that moment simply didn’t exist on the site.
Broken links and slow templates across 400 URLs put a ceiling on what any content work could return.
Three months. No time to publish our way out of it, so we didn’t.
Specific. No “we optimised your content strategy.” Here’s exactly what happened, in order.
Broken links resolved, page speed addressed at template level, metadata rebuilt across the estate, and existing pages brought in line with their target keywords.
All 300–400 posts classified by funnel stage. The classification is what revealed the scale of the bottom-of-funnel gap.
Head terms assigned to specific pages, cannibalisation conflicts resolved before any rewriting started.
The missing evaluation layer created and the existing library pointed at it, so research traffic finally had somewhere commercial to go.
The blog redesigned around trust and usability signals (author expertise, evidence, structure) rather than article count.
New pages created where a real query had no page to answer it, targeted rather than speculative.
Navigation and conversion path rebuilt so the highest-authority page on the site finally pointed visitors at the product.
Authority routed from the content library into the pages that needed to rank, instead of circulating inside the blog.
Fixes specced for Labellerr’s own developers and briefs written for their content team, so the work shipped inside the window.
Measured at Month 3 against the pre-engagement baseline. No paid traffic changes during the engagement.
YMYL, local, and competing against NHS.uk. Rather than fight for the head terms, we took the specific high-intent searches a family makes days before booking a visit.
Two decades of accessible-learning expertise that search engines read as a generic services site. No new content budget: the existing pages were remapped to buyer intent and the technical drag cleared.
A genuinely differentiated organic range that search couldn’t distinguish from anyone else’s. We fixed the cannibalisation, worked the products that already sold, then moved authority.
Every case on this site, including this one, started with a diagnosis before any work was recommended. That’s Stage 1 of The Clarity Method™. That’s where we start.
Book a free strategy call. No commitment, no pitch deck.