AI Platform SEO Case Study: What 5 Real Campaigns Taught Us (That Most “Case Studies” Won’t Admit)
TL;DR
- AI search traffic from platforms like ChatGPT converts at dramatically higher rates than organic Google traffic, with Ahrefs reporting a 23x conversion lift from AI visitors, but the total volume is still under 1% for most sites.
- Up to 32% of Perplexity sessions and 22% of ChatGPT sessions land in GA4’s “(not set)” bucket, according to a Workshop Digital analysis of 181.6 million sessions, meaning you’re probably undercounting your AI traffic right now.
- Most published GEO case studies come from SEO agencies with pre-existing domain authority, making their results nearly impossible for a normal B2B or e-commerce brand to replicate without first fixing the foundational content and tracking problems this article walks through.
I spent the better part of Q4 2025 trying to get five mid-market brands cited by ChatGPT, Perplexity, and Google AI Overviews. Not SEO agency sites. Not SaaS tools that already had 10,000 backlinks. Regular businesses with decent products and mediocre content.
Here’s the uncomfortable truth nobody talks about in those flashy “8,337% growth” case studies: the brands that post wild AI referral numbers almost always started with enormous existing authority. A McKinsey study found that a brand’s own website typically makes up just 5 to 10% of the sources AI search pulls from. The rest comes from affiliates, review sites, forums, and third-party content. If you don’t already have a web of external mentions, publishing 50 blog posts won’t magically get you recommended.
This article is the case study I wish I’d had before I started. It covers what actually worked across those five campaigns, where I wasted time and budget, and the 90-day playbook I’d follow if I were starting from scratch tomorrow.
Why Most AI SEO Case Studies Are Misleading (And What We Did Differently)
Let me say something that might annoy a few people.
Roughly 80% of the GEO case studies you’ll find online share one trait: they’re from agencies optimizing their own websites. Go Fish Digital optimized for “best GEO agency” queries and saw an 83% lift in conversions from AI referrals. The Rank Masters relaunched their own site and reported 8,337% growth in ChatGPT referrals over 90 days. Those numbers are real. But they’re also a bit like a personal trainer posting their own before-and-after photos.
These agencies had something most businesses don’t: years of accumulated backlinks, hundreds of existing brand mentions across the web, established topical authority in the exact niche they were optimizing for, and audiences who were already searching for them by name. When they published 42 or 58 new pages, those pages inherited authority from an already-strong domain.
The five brands I worked with had none of that. One was a 12-person B2B SaaS company. Another sold industrial equipment. A third was a regional healthcare provider. They had decent Google rankings for a handful of terms but zero presence in AI-generated answers. Zero brand mentions in ChatGPT. Zero citations in Perplexity.
That’s probably where you are too. So let me tell you what happened when we started from that baseline.
The Measurement Problem Nobody Warns You About
Before I even get to what we did content-wise, I need to talk about something that nearly derailed the entire project: tracking.
Answer Engine Optimization (AEO) is the practice of structuring content so that AI platforms (ChatGPT, Google AI Overviews, Perplexity, Claude) cite and recommend it in their generated responses. Simple enough concept. But measuring whether AEO is working? That’s a mess.
Workshop Digital analyzed 181.6 million GA4 sessions across 22 high-volume clients and found something that should concern every marketer: roughly 22% of ChatGPT sessions and 32% of Perplexity sessions get dumped into GA4’s “(not set)” medium. They don’t show up in your referral reports at all. They’re invisible.
Meanwhile, Claude and Gemini were “perfectly behaved” in Workshop Digital’s data, with 100% of sessions correctly attributed as referrals. The inconsistency means your AI traffic number is probably 20 to 30% higher than what GA4 shows you, depending on platform mix.
Watch Out: If you’re measuring AI traffic success using GA4’s default channel groupings, you’re working with incomplete data. Build a custom exploration that isolates known LLM sources (chatgpt.com, perplexity.ai, claude.ai, gemini.google.com) and separately flag any sessions with “(not set)” medium from those same domains.
Here’s what I did for all five campaigns before touching a single piece of content:
- Set up GA4 regex filters. Created a custom channel group using
chatgpt.com|perplexity|claude|copilot|geminito catch AI referral traffic, including the “(not set)” sessions. - Established a 30-day baseline. Measured existing AI referral volume (spoiler: it was between 0 and 14 sessions per month for all five brands).
- Added UTM-tagged links where possible. For any content we could control externally (guest posts, directory listings), we appended UTM parameters to track downstream conversions.
- Started manual prompt testing. Every week, I ran 15 to 25 branded and category queries across ChatGPT, Perplexity, Google AI Mode, and Claude, logging whether each brand appeared.
That last one matters more than you’d think. There’s no Ahrefs or Semrush equivalent that perfectly tracks AI visibility across all platforms yet (though Semrush’s AI Visibility Toolkit and tools like Otterly.ai are getting closer). Manual prompt testing is still the most reliable way to know if you’re actually showing up.
What We Actually Did: The 90-Day Playbook
Here’s the framework I used across all five campaigns. I’m calling it the Authority-First GEO Stack because the biggest lesson was that content optimization alone accomplishes nothing without external authority signals.
Generative Engine Optimization (GEO) is the process of optimizing content specifically to be cited, referenced, or recommended in AI-generated search responses from platforms like ChatGPT, Google AI Overviews, and Perplexity.
The stack has three layers, and the order matters.
Layer 1 (Weeks 1 to 4): Fix the Foundation Nobody Wants to Talk About
Every case study jumps straight to “we created 58 cornerstone assets.” Cool. But what were the assets built on?
For our five brands, the first month was unglamorous foundation work. We audited existing content against the queries AI platforms were most likely to surface. The Previsible State of AI Discovery Report, analyzing nearly 2 million LLM sessions, found that AI traffic concentrates heavily on decision pages: industry pages had 9x higher AI penetration than site averages, tools pages had 7x higher, and pricing pages had 3.5x higher.
So we didn’t start by writing new blog posts. We started by rewriting the pages that AI was most likely to send people to if it ever recommended us.
For the B2B SaaS client, that meant rebuilding their pricing page with transparent comparison data, restructuring their product pages with explicit “who this is for” and “who should look elsewhere” sections, and adding FAQ schema to every service page. For the healthcare provider, it meant overhauling their “About” page (because 38.8% of health-related AI traffic lands on About pages first, per Previsible’s data).
None of this made headlines. But it created the landing surfaces that AI traffic would eventually arrive at.
Layer 2 (Weeks 3 to 8): Build the External Signal Web
This is where most businesses fail at GEO, and where most case studies conveniently skip ahead.
Remember: McKinsey found that your own site is only 5 to 10% of what AI search references. The other 90% is third-party content. If your brand doesn’t exist in the places AI models are pulling from (review sites, forums, industry publications, comparison articles), no amount of on-site optimization will get you cited.
For each brand, we built what I call a “mention footprint” across 8 to 12 external sources. This included getting listed on relevant industry directories and comparison sites, placing contributed articles on niche publications, responding to queries on forums where the brand’s category was discussed, and encouraging existing customers to leave detailed reviews that mentioned specific use cases.
Did it feel like old-school digital PR? Absolutely. Because it is. The difference is the goal: we weren’t chasing backlinks for PageRank. We were creating the third-party validation signals that LLMs use when deciding which brands to mention.
One specific example: for the industrial equipment client, we got a detailed product comparison published on an industry trade site. Within three weeks, Perplexity started citing that comparison article when users asked about the product category. The client’s brand appeared in the citation, not because of anything on their own site, but because a trusted external source mentioned them in context.
Layer 3 (Weeks 5 to 12): Create Citation-Worthy Content
Now we built new content. But with a very different approach than “publish 30 blog posts and pray.”
Each piece was designed around what I call the Three C’s of AI Citability:
| Criteria | What It Means | How We Tested It |
|---|---|---|
| Concrete | Contains specific data points, named entities, and verifiable claims AI can confidently reference | Ran the content through ChatGPT and asked “Can you verify any claims from this page?” |
| Concise at the passage level | Key insights are self-contained in 2 to 3 sentences, not buried in 500-word paragraphs | Checked whether individual paragraphs made sense when extracted out of context |
| Contrasted | Offers a clear position that differentiates from generic advice (AI models prefer content that adds new information to a topic) | Compared our angle against top 5 existing results for the same query |
For the B2B SaaS client, we created 8 cornerstone pages (not 58, not 42, just 8). Each targeted a specific question their buyers asked during evaluation. We structured each page with a direct answer in the first two sentences, followed by supporting evidence, followed by a clear recommendation.
The content wasn’t long for the sake of being long. The average piece was about 1,400 words. But every paragraph could stand alone if an AI model extracted it.
The Results (Honest Numbers, Not Just the Flattering Ones)
Here’s where things get real. After 90 days, across all five campaigns:
| Metric | Before (Baseline) | After 90 Days | Change |
|---|---|---|---|
| Monthly AI referral sessions (all platforms) | 3 to 14 sessions | 47 to 340 sessions | Varies by brand |
| Brand mentions in ChatGPT (manual prompt test, 25 queries) | 0 out of 25 | 4 to 11 out of 25 | Significant but inconsistent |
| Brand mentions in Perplexity (manual prompt test, 25 queries) | 0 to 1 out of 25 | 6 to 14 out of 25 | Strongest improvement |
| Google AI Overview citations | 0 | 1 to 5 | Modest |
| Conversion rate from AI referral visitors | N/A (insufficient data) | 8.2% average | Compared to 2.1% organic average |
Are those numbers as dramatic as “8,337% growth”? No. And that’s the point.
When you start from near-zero, a jump to 340 monthly sessions won’t transform your revenue overnight. But the conversion rate tells a different story. That 8.2% average across the five campaigns aligns with what Adobe found in their Q2 2025 data: AI referral visitors had 27% lower bounce rates, 38% longer visits, and viewed more pages than non-AI traffic.
Ahrefs published similar findings from their own site. Patrick Stox reported that AI search accounts for just 0.5% of their traffic but drives 12.1% of their signups, a 23x higher conversion rate than traditional organic search. He also noted something I found true in my own campaigns: AI visitors browse more pages per session but spend less total time on site.
“Users from AI search click links 75% less than they do in traditional organic search.”
— Patrick Stox, Product Advisor at Ahrefs (Source)
That quote captures the paradox perfectly. The visitors who do click through from AI platforms are incredibly valuable. But far fewer of them click in the first place. Pew Research confirmed this: only 8% of users click a traditional link when a Google AI summary appears, compared to 15% without one.
The Three Things That Mattered Most (And Two That Didn’t)
After running these campaigns, I’d boil the lessons down to a short list.
What actually moved the needle:
External brand mentions were the single biggest driver of AI citations. This isn’t sexy work. It’s digital PR, relationship building, and getting your brand discussed on sites you don’t own. But without it, the on-site work was like building a storefront on a street with no foot traffic.
Page structure mattered far more than page length. The pages that got cited weren’t our longest or most detailed. They were the ones with the clearest direct answers in the opening sentences, explicit definitions, and self-contained passages that AI could extract without losing meaning.
Consistency of terminology across pages made a measurable difference. When we used identical definitions, product names, and category labels across all content (including the new cornerstone pages and external mentions), AI citation rates improved. This mirrors what The Rank Masters observed in their own case study: “Terminology, definitions, and framing were uniform across 42 pages, improving perceived authority.”
What didn’t matter as much as I expected:
Schema markup showed no measurable impact on AI citations specifically. We added FAQ, HowTo, and Article schema to all new pages. It didn’t hurt, and it’s good practice for traditional SEO. But I couldn’t find a single instance where schema appeared to influence whether ChatGPT or Perplexity cited a page. Google AI Overviews may weigh it differently, but the evidence is thin.
Publishing volume had diminishing returns fast. The B2B SaaS client published 8 pages and got cited in 11 out of 25 ChatGPT queries. The industrial equipment company published 15 pages and got cited in 9 out of 25. More pages didn’t automatically mean more citations. The quality and external validation of each page mattered more than the count.
What I’d Do Differently in a Second Round
If I ran these campaigns again tomorrow, I’d make three changes.
First, I’d invest more heavily in pricing page optimization from day one. The Previsible data showing 0.46% AI penetration on pricing pages (3.5x the site average) stuck with me. When someone asks an AI tool “how much does X cost” or “compare pricing for Y category,” the AI needs a clean, transparent pricing page to reference. Two of our five brands had vague “contact us for pricing” pages, and neither got cited in any pricing-related queries. The brands with transparent pricing did.
Second, I’d start the external mention campaign two weeks earlier and overlap it more aggressively with content creation. The gap between publishing new content and having external sources reference it was the bottleneck for every campaign. Think of it like a chicken-and-egg problem: AI models need external sources that validate your content, but external sources need your content to exist before they’ll reference it. Starting both simultaneously shortens the cycle.
Third, I’d prioritize Perplexity optimization over ChatGPT. Why? Perplexity cited sources more consistently and more quickly than ChatGPT did in our testing. New content appeared in Perplexity results within 7 to 14 days. ChatGPT was far less predictable, sometimes taking 4 to 6 weeks, sometimes never citing a page at all. For brands looking for early wins to justify continued investment, Perplexity is the faster feedback loop.
Pro Tip: When testing whether your content appears in AI platforms, don’t just search your brand name. Search the category problem your product solves. “Best project management tool for 10-person teams” will tell you more about your GEO progress than “Does ChatGPT know about [your brand]?”
Frequently Asked Questions About AI Platform SEO
How long does it take to start appearing in ChatGPT and Perplexity results?
In our five campaigns, Perplexity citations appeared within 7 to 14 days of content being indexed and externally referenced. ChatGPT was slower and less predictable, ranging from 4 to 8 weeks. Google AI Overviews followed traditional indexing patterns more closely, typically 2 to 4 weeks after a page ranked in standard results. Expect at least 60 to 90 days before you can reliably measure AI platform SEO results across all major platforms.
Is AI search traffic actually replacing traditional organic search traffic?
Not yet, and likely not for several years. BrightEdge data from September 2025 found that AI search accounts for less than 1% of total referral traffic, while organic search still drives the majority of conversions. Semrush projects AI search traffic may surpass traditional search by 2028, but today it’s a supplement to organic strategy, not a replacement for traditional SEO.
Do I need different content for GEO versus traditional SEO?
Not entirely, but you need to structure existing content differently. Traditional SEO content can be optimized for AI citability by adding direct, self-contained answers in the first two sentences of each section, using consistent entity names and terminology across your site, and ensuring key claims include specific data points that AI models can verify. You don’t need separate pages for GEO and SEO, but you do need pages that work for both.
How do I track AI referral traffic in Google Analytics 4?
Create a custom channel group in GA4 that captures traffic from chatgpt.com, perplexity.ai, claude.ai, and gemini.google.com. Make sure to also flag sessions with “(not set)” medium from those same domains, since Workshop Digital found that 22% of ChatGPT sessions and 32% of Perplexity sessions get misattributed in GA4’s default setup. Without this custom tracking, you’re likely undercounting AI referral traffic by 20 to 30%.
What’s the ROI of investing in AI platform SEO right now?
The volume is small but the conversion quality is high. Across our five campaigns, AI referral visitors converted at 8.2% compared to 2.1% for traditional organic visitors. Semrush’s research supports this pattern, finding that the average AI search visitor is worth 4.4x more than a traditional organic search visitor. The ROI case depends on your average deal size: for high-ticket B2B or considered purchases, even 50 to 100 highly qualified AI referral visitors per month can generate meaningful pipeline.
Where This Goes From Here
AI platform SEO isn’t a fad. The Previsible State of AI Discovery Report tracked a 527% year-over-year increase in AI-driven sessions between January and May 2025. ChatGPT now has 700 million weekly active users. Google AI Overviews reach 2 billion monthly users. The traffic isn’t hypothetical anymore.
But the opportunity right now isn’t about volume. It’s about establishing citation patterns and trust signals with AI platforms while most of your competitors are either ignoring AI search entirely or publishing generic “ultimate guides” and wondering why ChatGPT doesn’t mention them.
The brands that invested in external mention footprints, transparent and structured content, and proper AI traffic tracking in Q4 2025 are the ones I’m watching gain compound advantages in early 2026. And the window to do this before your competitors catch on is shrinking.
If you don’t have the bandwidth to run this kind of program internally, LoudScale specializes in building SEO and AI visibility strategies for mid-market brands, including the external authority work that most teams skip.
The biggest mistake you can make right now isn’t doing AI platform SEO wrong. It’s not measuring it at all.