A GEO engagement that took a workflow SaaS from invisible in AI answers to cited across nearly half of tracked buyer prompts in six months.

Client identity withheld. Figures are from the engagement’s monthly prompt-panel tracking.

The situation

A workflow automation SaaS ranked respectably in classic search but had noticed a pattern in sales calls: prospects arrived having already asked ChatGPT for a shortlist, and the shortlist rarely included them. When it did, the description was two funding rounds out of date. Their category’s AI answers were being written by review sites and two competitors, and every month that consensus hardened.

What we found

The baseline panel of forty buyer prompts across ChatGPT, Perplexity and Gemini put numbers on the anxiety: the client appeared in nine percent of answers, competitors in over half. Diagnosis took a week and surprised nobody who has done this work. A security plugin was blocking GPTBot and PerplexityBot outright. Key pages rendered their substance through JavaScript that retrieval fetchers never execute. Product pages were written in marketing prose with no passage a machine could quote as an answer. And the third-party record was thin: three credible sources discussed the product, two of them stale.

The strategy

Sequence by response speed. Access and quotability move retrieval-based answers within weeks; the external record moves everything else over months. So: unblock, restructure, then build the record while the early wins compounded, with the panel re-run monthly as the single source of truth.

The work

Phase 1: open the doors (weeks 1-2)

Bot access was corrected, llms.txt shipped pointing at the twenty most citable pages, and the rendering issue was fixed by moving critical content server-side on the templates that mattered.

Phase 2: become quotable (weeks 2-8)

The top pages were restructured around extraction: direct answers under question-phrased headings, a plain-language definition of the category they wanted to own, honest comparison pages against the two competitors dominating answers, and a pricing page that stated its posture clearly instead of hiding it. Every page gained passages that stand alone as complete, correct answers.

Phase 3: build the record (weeks 4-24)

Interrogating the assistants revealed which sources they actually trusted for this category. Those became the target list: profile depth on the two review platforms models cited constantly, contributed expertise on three industry publications, a data piece that earned pickup, and consistency sweeps so every description of the company agreed on the facts.

The results

Perplexity moved first, exactly as retrieval logic predicts: citations appearing within five weeks of the access and structure work. By month six, the client appeared in 37% of tracked prompts across assistants, a fourfold lift, with descriptions current and framing neutral-to-recommended. Branded search grew 41% over the same period as AI-referred buyers verified what they had been told. Sales began hearing their own comparison page’s language read back to them in discovery calls.

What made the difference

Fixing access before content, and letting the assistants themselves nominate the citation targets. Most GEO effort is wasted guessing what to build; the engines will tell you which sources they trust if you ask them systematically.

Explore the services behind this engagement: GEO Optimization, AI Citation Building and AI Search Visibility.