SaaS
Getting recommended by the LLMs.
How a mid-market SaaS went from zero AI share of voice to the default recommendation in two of its three core category queries.
CASE / 04
Confidential
Where does a B2B buyer start a vendor search now? For this client's pipeline, it was not Google. About 38% of qualified demo requests in Q2 named a language model by hand in discovery calls. The client's brand showed up in roughly 6% of those answer sets.
GEO
Global (EN-first)
Setup
AI discovery audit
Duration
18 weeks
Shipped
Q1 2026
Why the number was low
Why the number was low
Two reasons. First, the content that ranked well on Google was not the content language models cited. Google rewards completeness, breadth, internal linking. Models cite content that is structured, canonical, and answers one question cleanly. Most of the client's content was the first kind, almost none was the second.
Second, there was no structured data on the answer pages. No schema, no FAQ markup, no product-comparison tables that a model could lift. The brand was invisible to the one layer of infrastructure that had started routing real buying intent.
Query audit
Query audit
Forty seed questions were pulled from the client's top three buyer personas and run weekly across ChatGPT, Perplexity and Gemini for six weeks. For each answer we logged: who was cited, why they were cited (structured answer, third-party review, comparison page), and whether the citation survived across sessions.
Content that models actually cite
Content that models actually cite
Fourteen canonical pages were rewritten to the answer shape models reward. One question per page, structured answer block in the first 120 words, supporting detail below, citations to primary sources. Three comparison pages were added against the two competitors that dominated the answer sets. This was the most uncomfortable piece for the client, because those pages read like they were written for a machine, not a human. They were.
What did not work
What did not work
A thought-leadership playbook we inherited from the previous agency produced zero model citations in eight weeks. Long-form posts without structured answer blocks are invisible at the model layer, even when they rank on Google. We stopped publishing them in week five.
Launch With Us