Playbook

Most AI search case studies show you a visibility chart going up and to the right, and nothing else.

No pipeline. No demos. No answer to the only question that matters: did any of this turn into revenue conversations?

This piece is the opposite. It's the full breakdown of a real 90-day AI search and SEO program we ran for one client, at $8K/month, with the forecast we committed to before work started and the measured results against it.

Every number below comes directly from the client's HubSpot and GA4, not from estimates.

The client is a B2B SaaS company selling into mid-sized and enterprise organizations. Operational buyer, 60 to 120 day sales cycle, an established category with entrenched competitors.

Before the engagement, they had essentially zero bottom-funnel search presence and appeared in about a third of the AI prompts their buyers actually ask.

Here's the exact program we built. The framework, the deliverables, the forecast, what we deliberately chose not to do, and the attribution problem that hides most of this channel's results if you don't know where to look.

Most AI search engagements fail this step. The agency promises "visibility," the client expects pipeline, and three months in nobody can say whether the program is working.

We built a 12-month forecast model before the engagement started, with two scenarios the client could hold us against.

Our GEO Stack has three layers, built in order.

1. Layer 1: Solution-aware presence
   The content AI assistants pull from when answering buying questions. Category listicles, competitor alternatives pages, comparison content. BOFU first, because that's where intent is highest and time to first result is shortest.

   Deployed
2. Layer 2: Citation velocity
   Third-party mentions in high-authority sources the models already trust. Presence puts you in the conversation. Citation velocity is what makes the model confident recommending you over the incumbent.

   Deployed
3. Layer 3: Problem-aware bridges
   Content that connects a buyer's process pain to your product category, so AI brings you into the conversation before the buyer knows the category exists.

   Deliberately deferred

For this engagement, we ran Layers 1 and 2 only. Layer 3 was deliberately deferred, and I'll explain why in the "what we said no to" section.

Citations pointing at pages that don't exist do nothing. Problem-aware content with no solution-aware foundation sends buyers to a shortlist you're not on.

The content plan was 15 BOFU pages. Eight shipped inside the 90-day window, seven more in production behind them.

Every page was net-new. None of these URLs existed before the engagement, which makes the attribution unusually clean: any result from these pages is program-attributable by definition.

What the pages did

By the end of the window, the results at the page level:

Citation velocity ran alongside the content from month one.

Across the window, we secured 24 links and citations at an average domain rating of about 69, every one pointing at a published bottom-funnel page or the homepage.

This is not link building in the traditional sense. The target list came from the AI side.

We extracted the domains the assistants were already citing when answering the client's 89 tracked prompts, found the overlap with pages ranking for solution-aware keywords, and prioritized the publications the models demonstrably trust.

Outreach led with value: updated data, corrections, genuine inclusions. Not PR pitches.

The effect you're building is mention frequency. Each placement raises the probability the model treats the brand as a credible answer on the next adjacent prompt.

Against the forecast both sides agreed to.

Traffic: 519 captured BOFU visits against 319 Conservative

That's 63% ahead of the cautious scenario and closing on the Realistic line of 638.

And this number understates reality, because GA4 can't see most AI-referred visits. More on that below.

Demos: 59 against a Realistic forecast of 9.4

Fifty-nine buyers who named Google Search or AI Search on the intake form booked a demo inside the window.

That's 6.3 times the Realistic scenario and roughly 17 times Conservative. Of everyone who self-reported Google as their source, 96% booked. Of everyone who self-reported AI search, 100% booked.

Those two rows are the contract metrics. Underneath them, the leading indicators that drive the next two quarters.

AI citation rate went from 33% to 47%

Across the 89 tracked prompts, the client's own content is now cited as a clickable source in 47% of AI answers, up 13 points from baseline, with 56 of 89 prompts improving.

The gains concentrated exactly where we published: the competitor alternatives clusters jumped between 35 and 59 points, with the best cluster going from 23% cited to 82% cited.

Buyers naming Google as their source rose 159%

From 22 to 57 versus the prior 105 days.

Counting AI search alongside, the two channels became the single largest named source of inbound.

AI-sourced buyers were the highest quality in the funnel

Of the buyers who named AI search, 8 of 9 reached MQL or beyond. An 89% qualification rate, roughly double the inbound baseline.

This matches what we see across our client base: AI search delivers a fraction of the volume of Google and several times the intent, because the buyer arrives having already received a recommendation.

Here's the section that justifies the parenthetical in the title.

During the window, 13 buyers explicitly told the client they found them through AI search. The number HubSpot machine-tagged as AI: 2.

That's not a settings error. It's structural.

HubSpot builds its source field from the referring URL. When ChatGPT, Perplexity, Gemini, or Claude sends a buyer, the referrer is stripped, so the visit arrives with no source data and gets logged as Direct.

The same distortion hits Google: 73 buyers named Google on the form against 43 the dashboard tagged as organic.

There's a second-order effect that's even sneakier. Buyers who see a brand recommended in an AI answer often don't click at all.

They remember the name, then come back later by typing the brand into Google or going direct. That behavior gets logged as branded search or Direct traffic, and Direct sessions rose about 14% during the engagement. The downstream signature of upstream AI mentions.

So the operating rule we use with every client: treat the dashboard numbers as a floor, treat form self-reports as the recovered signal, and build the reporting around the gap between them.

Without it, this entire channel is invisible, and you'll conclude a working program is failing.

The program is defined as much by what we didn't do.

Everything above is a proven starting point, not a finished machine.

That last point is the honest caveat every AI search vendor should give you. This channel compounds, which means the early window understates the steady state.

But it also means the brand that builds the citation base first becomes exponentially harder to displace. Every month you're not the cited answer, someone else is training the model that they are.

At $8K/month, the program delivered 519 measured BOFU visits, 59 demos from search and AI buyers against a forecast of 9, a citation rate that went from one in three prompts to one in two, and an AI-sourced buyer cohort qualifying at 89% MQL+.

None of it required a secret.

It required sequencing (presence before citations, citations before expansion), restraint (no TOFU, no programmatic volume, no tactic chasing), and an attribution setup honest enough to see the channel at all.

We run AI search and SEO programs exclusively for B2B SaaS companies. If you want a senior operator to audit your category's AI search landscape, map the prompts that matter for your ICP, and hand you a prioritized 90-day plan, let's talk.