A SaaS contract-management platform, $20M ARR

The challenge

The client had spent three years building a strong organic position in legal operations search. Their domain ranked top three for most category terms, traffic was healthy, and demo requests were predictable. Then, over a single quarter, demo volume from organic dropped 22 percent. The drop was not visible in Search Console. Sessions held flat, impressions even ticked up. What had changed was the path: prospects were asking ChatGPT and Claude about contract management software, getting answers, and never clicking through. When the team ran the obvious prompts, two competitors showed up by name in every response. They did not. Internal SEO had built a roadmap focused on more long-tail blog posts. The CMO suspected that was the wrong move, brought us in for a second opinion, and asked for a 90-day reset. The board had given marketing one quarter to either reverse the demo decline or restructure the channel mix, which made the timeline non-negotiable.

What we shipped

We started with a citation audit across ChatGPT, Claude, Perplexity, and Gemini, running 80 category and comparison prompts twice weekly. The baseline confirmed it: zero citations on category prompts, four citations on bottom-funnel comparison prompts where the brand name was already in the query.

We rebuilt the on-domain corpus around three things. First, a category-defining pillar at /contract-management with a clear definition, a feature-by-feature comparison matrix against the four named competitors, and a structured FAQ that mapped to the real prompts buyers were asking. Second, we stripped the existing thin product pages and consolidated them into nine deep solution pages with concrete pricing context, integration lists, and named customers. Third, we shipped llms-full.txt at 740KB, a complete plaintext export of the product, security, pricing, and integration documentation, hand-curated so the dense parts came first.

On schema, we added SoftwareApplication, FAQPage, and Organization markup site-wide, plus a sameAs graph linking to G2, Capterra, Crunchbase, LinkedIn, and the founder's verified profile. Off-domain, we placed eight bylined posts in three legal-ops trade publications and pushed for inclusion in two analyst-maintained vendor lists. Finally, we set up a weekly citation tracker so the marketing team could see movement without depending on us.

The numbers

The most useful shift was not the citation count, it was lead quality. Sales reported that prospects arriving via demo in months four and five were coming in further along the buying cycle. Discovery calls got shorter. Reps stopped having to explain what contract management software is, because the prospects had already had that conversation with an LLM and showed up asking specific questions about redlining workflows and SOC 2 evidence. Sales cycle from first call to close compressed from 71 days to 58. Branded search volume nearly doubled by month six, which is the lagging indicator we trust most for AI-driven discovery. The team also saw fewer competitive bake-offs at the late stage, because the LLMs were framing the category around the client's positioning rather than the incumbent's. Win rate on competitive deals moved from 31 percent to 42 percent over the engagement, and the head of sales attributed roughly half of that lift to prospects arriving with a more accurate prior on the category.

What we'd do differently

We underweighted the importance of customer-named case studies in the early corpus build. The first wave of pages had strong feature depth but used pseudonymous customer references. When we replaced those with named, logo-permission customers in month four, citation rate jumped another 14 points within three weeks. That should have been wave one, not wave four. We also spent too long debating llms.txt vs llms-full.txt structure before shipping. The right call was to ship something serviceable in week two and iterate, not to design the perfect file in week six. We lost roughly five weeks of compounding to that.

What's next

The 90-day engagement renewed into a retainer focused on three areas. First, defending the category position as competitors notice and respond, which we expect by Q3. Second, extending the citation work into adjacent buyer queries around CLM plus procurement and CLM plus revenue operations, where the client has real product depth but no surface coverage in LLM responses. Third, we are building a citation-attribution model that ties LLM mentions to pipeline by week, so the CMO can defend the line item to the board with something more rigorous than a screenshot of a ChatGPT response. The goal for the next two quarters is to push category-prompt citation rate past 70 percent and hold it. We are also instrumenting a competitor-monitoring layer so the team can see when other vendors start rebuilding their own llms-full.txt files or making structured changes to their entity graphs, because the early movers in this category will keep an edge only as long as the rest of the field stays slow to react.