Case studies

The company sells software to enterprise buyers worldwide. Organic search was growing. The real question was whether that growth reached the pages that sell.

Mapping the keyword universe against the site made the gap obvious. Almost all organic visits landed on informational pages. The product and solution pages, the ones that create pipeline, got almost none. Publishing more content would have widened the gap, not closed it.

SEO strategy, site architecture, the redirect plan and the rollout. All of it, start to finish.

• Split the site into two engines with separate jobs. Traffic pages earn category authority. Revenue pages turn that authority into demand.
• Rebuilt the silo structure around that split and shipped the full redirect plan.
• Built the bottom of the funnel. Comparison, alternatives, pricing, migration and competitor-capture pages aimed at buyers already evaluating.

Monthly organic traffic grew from 180K to 315K+. Blog traffic grew from 10K to 28.7K. Branded search grew from 5K to 12K+. The USA became the #1 traffic-contributing region globally.

Traffic pages and revenue pages are different products. Run them as one engine and the gap between them stays invisible.

Traffic is the easy number to grow and the easy number to report. The harder question is what that traffic is worth. Answering it means connecting organic demand to the CRM, and the CRM to deals.

The HubSpot stack. Workflows, forms, lifecycle stages, lead scoring and deal-level attribution. Plus the bottom-of-funnel pages that fed SDR conversations in the USA and the rest of the world.

• Set up workflows, forms, lifecycle stages and lead scoring in HubSpot.
• Added enrichment with Clay, Bitscale and Factors.ai. ICP fit became a property, not a guess.
• Attributed pipeline at the deal level, not at the form fill.
• Fed SDRs from comparison, alternatives, pricing, migration and competitor-capture pages.

On a PLG-led model, MQLs grew from 200 to 1,500 a month. SEO-influenced pipeline grew 4x, from $2M to $8M.

Sessions, MQLs and revenue live in different worlds. A blog post can move sessions without touching MQLs. A sharp BOFU page can move revenue while barely registering on the traffic chart. Watch the gap between the three, not any single line.

Ask ChatGPT, Perplexity, Gemini or an AI Overview for the best tools in a B2B category and the answer is usually assembled from a handful of third-party listicles and comparison pages. If a brand isn’t on those pages, it usually isn’t in the answer, however good its own site is.

Most brands respond to AI search by publishing more on their own domain. That helps search. It rarely moves the AI answer for category and comparison questions, because the engines cite sources they already trust on those queries.

The whole loop. Prompt research, citation-source mapping, target selection, sourcing through publishers and placement partners, negotiation, client approval and live tracking.

• Built the buyer prompt set for each category. Best-of, alternatives, comparison and use-case questions.
• Mapped every prompt to the exact pages AI engines cite for it, across engines, and ranked those pages by how often they appear.
• Kept only targets that AI engines were already citing for the client’s category.
• Sourced placements through publishers and partners, and kept a claims-to-verify register so every description of a client was accurate before it went live.
• Tracked every placement from approval to live.

100+ listicle placements acquired for clients, each inside a page AI engines were already citing for the client’s category.

On category questions, AI visibility is mostly an off-site problem. Own-site content earns rankings. Placement on the pages the engines already trust earns the mention.

Internal linking is one of the highest-leverage fixes in SEO and one of the worst done by hand. Someone skims a few posts, links the obvious phrase, and moves on. The pages that need authority most, the ones ranking on page two, rarely get it.

A repeatable pipeline run from Claude Code on any site.

• Inventory. Claude runs the crawl through Screaming Frog MCP, or reads a saved crawl export. It keeps indexable 200 pages and splits them into source pages that carry links and target pages that should receive them. Orphan and thin pages get flagged.
• Demand. Pulls three months of Search Console data per page. Pages with strong impressions, positions around 5 to 20 and weak CTR get pushed up. The queries a target already ranks for become its anchor candidates, because they are proven demand.
• Value. Batches every anchor candidate into one SE Ranking call for volume, difficulty and intent. Commercial and buyer intent ranks above generic phrases.
• Matching. Embeds every source paragraph and every target (title, description and ranking queries) with BAAI/bge-base-en-v1.5 from Hugging Face, then keeps only pairs above a cosine similarity floor.
• Placement. If the anchor already sits in the paragraph, the instruction is to link those words, no rewrite. If not, a local llama3.2:3b on Ollama drafts one short sentence that carries it.
• Guardrails. At most two new links per source page. Caps per target. No self-links, no links that already exist on the live page, no generic or brand-only anchors.
• QA. Every row is checked. One bracketed anchor, the right target title, no duplicates. The semantic fit distribution is reported so the team can see how strong the set is.

One sheet, sorted by a 0 to 100 priority score. Each row tells the implementer the exact sentence to find, the words to link, the target, whether it is an exact or topical match, the similarity score, how often that anchor is already used site-wide, and why the link matters.

Small details decide whether a pipeline like this gets used. qwen3 leaked its reasoning into drafts, so drafting moved to llama3.2:3b. The model matters less than the guardrails around it.

The full method is published as a free Claude skill, the Internal Linking Engine. Read it or download it.

A lot of SEO and growth work repeats. Clustering, crawls, audits, reporting, small page changes. Done by hand, it takes the time a team needs for diagnosis and strategy.

• An SEO NLP engine. BERT models from Hugging Face and local LLMs running on Ollama, wired into Claude over MCP. It handles keyword clustering, intent classification, content gaps, entity extraction and internal linking.
• Agentic SEO operations. n8n and Claude agents that push SEO changes to staging without dev tickets and deploy landing pages straight to the domain.
• An audit library. Claude skills for technical, AI visibility, schema, page speed, ranking-factor, brand-fact and entity audits. Screaming Frog crawls run through MCP and come out as audits a dev team can act on.
• Reporting on live data. Weekly SEO reports, performance trackers and MBR decks built from the source, not assembled by hand.
• Tools without an engineering queue. A calculator shipped with zero engineering, and a Next.js leads dashboard on the HubSpot API.

SEO changes reach staging without a dev ticket. Landing pages deploy without an engineering queue. Reports and decks come from live data instead of manual assembly.

The models are the easy part. The value sits in small, dull systems around them that run the same way every week and check their own numbers.