Case Study: How a Cybersecurity SaaS Went From 0 to 18 AI Answer Mentions

This is a Reddit marketing case study about a Series B cybersecurity SaaS, an EDR platform selling into mid-market IT and security teams. We are publishing it with the client anonymized and every number exactly as it appeared in our reporting, because the numbers are the point.

When the engagement started, the brand had presence on 4 of the 68 Reddit URLs that AI engines were retrieving for buyer-intent queries in their category. That is 6%. Across the first 14-day baseline window, zero AI answers in our tracked prompt set named them. Their competitors were named in 73% of those same answers. Buyers were asking ChatGPT and Gemini for EDR recommendations every day, and the brand did not exist in the reply.

The product was not the problem. The evidence was. Security buyers pressure-test tools in comparison threads, incident post-mortems, migration regret threads, and "what should we buy" posts, and this brand was absent from all of them. The competitors being recommended were not better. They were described more often, in the places the machines read from.

What we actually did

Two weeks of baseline measurement first, because a program without a baseline is a story, not a program. We ran a fixed set of buyer-style prompts daily across ChatGPT, Gemini, and Google AI Overviews and logged which Reddit URLs the answers retrieved and which brands they named. That produced the tracked set of 68 threads and the two numbers above.

Then a quarter of careful participation. Every post and in-thread comment was written to hold up if you removed the brand name entirely: incident response lessons, deployment failure modes, honest tradeoffs between architectural approaches, field notes from production environments. Everything was client-approved before publishing, posted by hand from accounts with years of ordinary history, and subject to the usual guardrails. No fake accounts, no vote manipulation, no undisclosed affiliation where affiliation mattered.

In comparison threads we answered the buyer's actual question and named tradeoffs, including ones that did not favor the client. In security, the follow-up question always comes, and a comment that cannot survive it is worse than no comment.

Paid amplification came last, and only behind angles that had already survived organic scrutiny, pointing at evaluation assets like benchmarks and deployment guides rather than hype pages.

The results, with denominators

Brand presence went from 4 of 68 tracked Reddit URLs to 47 of 152 by the end of the engagement. The tracked set itself grew as the category's AI search surface expanded; presence grew faster, from 6% to 31%.

AI answers naming the brand went from 0 per 14-day window to 18. Of those 18, 12 cited at least one Reddit thread our team was active in. That linkage is what attribution looks like in this channel: not a click path, but a visible line from the thread we worked in to the answer that named the client.

AI retrievals of tracked Reddit URLs went from 84 to 402 per window. Delivery ran at an average of 23 client-approved actions per window across the final three cycles.

And the number the CFO cared about: new leads went from 11 in the 45 days before the engagement to 29 in the final 45 days, with a 42% close rate on the cohort that self-reported finding the brand through AI tools or Reddit. Attribution came from a post-sale form field that offered "AI / LLM (ChatGPT, Claude, Perplexity)" and "Reddit" as explicit options. Self-reported attribution undercounts; it does not overcount.

What this proves and what it does not

It proves that in any B2B category where buyers ask the internet before they fill in a demo form, moving the content of the threads AI engines read from moves both AI presence and pipeline, and that the movement is measurable week by week rather than defensible only in a 12-month retrospective.

It does not prove that volume works. The 23 actions per cycle figure is meaningless without the quality bar attached to it: client approval, defensible specifics, willingness to name tradeoffs. LLMs preferentially cite content that survives scrutiny, and so do the humans in the thread. A program that scales activity without scaling that bar will not reproduce these numbers.

If your category has the same shape, buyers comparing options in public while your brand sits outside the conversation, the first step is the same baseline measurement we ran here. We can show you what yours looks like in a 30-minute call: cal.com/aodhan-upvote/30min.