What GEO Traffic Actually Looks Like: 6 Months of Real Data from AI Search Optimization

If you've been reading about Generative Engine Optimization, you've probably seen the stats: 140% increase in LLM traffic, 23x conversion rates, 1,200 monthly sessions from AI referrals. Those numbers come from case studies — and they're real. But they don't tell you what the day-to-day of GEO traffic actually looks like when you're the one doing the work.

We've been optimizing for AI search engines at Mintec for about six months now. Not just for ourselves — for clients across SaaS, professional services, and e-commerce. Here's what the numbers actually show, what Google's new Search Console report reveals (and hides), and which levers actually move the needle.

The measurement problem: nobody has a complete picture

The first thing you learn when you start tracking GEO is that the data is fragmented across at least three different places, and none of them tell the full story.

Google Search Console's Generative AI Performance report, launched in June 2026, shows impressions your pages get inside AI Overviews and AI Mode. It tells you which queries triggered your content. What it doesn't show: clicks, click-through rate, or which AI engine actually cited you. As 99signals put it, this is the "not provided" moment for GEO — Google is showing you just enough to know something is happening, but not enough to act on it with confidence.

Google Analytics 4 can track referral traffic from known AI platforms — chatgpt.com, claude.ai, perplexity.ai, gemini.google.com — but only when the AI links out. When an AI Overview synthesizes your content without a link, there's nothing to track. This means GA4 undercounts your actual AI visibility.

Third-party tools like Ahrefs and Semrush have started adding AI citation tracking, but the data is spotty and delayed. Ahrefs recently found that AI traffic accounted for 12.1% of signups in one study while representing only 0.5% of total visitors — a 24:1 conversion ratio. That's the kind of math that makes GEO worth pursuing even when the raw traffic numbers look small.

The checklist we use at Mintec for every client property:

1. Set up a GA4 custom segment for LLM referral sources
2. Monitor the GSC Generative AI report weekly — not for clicks, but for query trends
3. Use Ahrefs or Semrush AI citation tracking as a directional signal
4. Manually spot-check ChatGPT, Claude, and Perplexity for your target queries monthly

If you only do one thing, set up the GA4 segment. It's the only source that shows actual humans arriving on your site.

Real numbers from 6 months of GEO work

Here's what we've actually seen across our content and client properties.

Our own site (mintec.co): 15 LLM-referred sessions in a recent 7-day window — 8 from ChatGPT, 7 from Gemini. That's small. But we have 376 blog posts and only 30.6% have FAQ schema. The posts that do have structured data get cited disproportionately. Our top-cited page, a web design trends piece, pulled 3 ChatGPT sessions in one week with zero paid promotion.

Client A — B2B SaaS (6 months of GEO optimization): Started from zero. After implementing FAQ schema across 40 product and comparison pages, adding citation-focused introductions (direct answers to likely queries), and building a structured glossary of industry terms, they now see 60-80 monthly sessions from AI referrals. That's about 3% of their total organic traffic. But those visitors convert at nearly 5x the rate of Google organic — they arrive with specific intent and trust the AI's recommendation.

Client B — professional services (3 months): 20-30 monthly AI sessions so far. The pattern is consistent: the pages that rank for "what is X" or "how does Y work" queries pull AI traffic. Comparison pages ("X vs Y") are the second-best format. Opinion pieces and news-driven content rarely get cited.

These numbers won't blow up your analytics. But the conversions per session are genuinely different. Enrich Labs reported 14 demo requests attributed to AI referral traffic in 90 days, versus zero in the prior 90 days. That's not a traffic story — it's a lead quality story.

Google's AI report: what it actually shows

The new GSC Generative AI Performance report deserves a closer look because most of the commentary about it has been wrong in one direction or the other. It's neither useless nor the answer to your measurement problems.

What you get:

• Impressions in AI Overviews and AI Mode, broken down by page and query
• A separate view for generative AI features in Discover
• An opt-out toggle (currently being tested in the UK) that removes your site from AI search features without affecting regular rankings

What you don't get:

• Click data
• Which AI engine cited you (AI Overviews vs AI Mode vs Discover AI)
• Competitor comparison data
• Any visibility into ChatGPT, Claude, or Perplexity citations

The report is useful as a directional signal. If a page suddenly spikes in AI impressions, something is working. If a page you've optimized for GEO shows zero impressions after 60 days, your approach needs adjustment. But you cannot measure ROI from this report alone — you need GA4 to close the loop.

One pattern we've noticed: pages with FAQPage structured data appear in the GSC AI report 3-4x more frequently than pages without it, even when both rank similarly in regular search. Google may have deprecated FAQ rich results for visual snippets, but the structured data still feeds the AI citation pipeline.

What actually moves the needle

After six months of experimentation, here's what we've learned works and what doesn't.

What works:

Structured data, specifically FAQPage and Article schema. Our data shows a clear correlation between schema implementation and AI citation frequency. This aligns with what Stackmatix and the broader GEO community have documented: well-structured content with proper JSON-LD markup is the highest-signal input for AI citation.

Direct-answer introductions. Pages that open with a clear, concise answer to a likely query — before diving into depth — get cited more often. AI engines pull from content that resolves a query quickly.

Comparison and definition content. "X vs Y" and "What is X" formats consistently outperform how-to guides and opinion pieces for AI citation. The pattern holds across industries.

FAQ content that mirrors real search queries. Not keyword-stuffed FAQ sections. Actual questions your audience asks, answered in 2-3 substantive sentences. Google's AI pulls from these more than from body paragraphs.

What doesn't:

Keyword optimization of existing content without structural changes. Rewriting H2s to include target phrases doesn't move the GEO needle if the underlying structure isn't citation-friendly.

Chasing AI trends with news-style content. AI engines cite evergreen, definitional content far more than time-sensitive posts.

Treating GEO as separate from SEO. The Enrich Labs finding is consistent with what we see: GEO optimization amplifies existing SEO strength. It's not a parallel channel — it's a layer on top.

The conversion math changes the conversation

Every GEO skeptic we talk to points at the small traffic numbers. And they're not wrong — the volume is low. But the conversion behavior is different enough that the volume argument misses the point.

AI Thinker Lab found AI search visitors convert at 4.4x to 23x the rate of standard organic traffic. Ahrefs found a 24:1 conversion ratio. Our own client data shows a consistent 3-5x uplift in conversion rate for AI-referred sessions versus Google organic.

Why? Because AI-cited traffic arrives with pre-built trust. The AI already validated the source. The user didn't click result #4 out of 10 hoping it might be useful — they clicked the specific source the AI named as authoritative.

This is why GEO matters even when the traffic numbers look small. A post that gets 40 AI-referred sessions at a 15% conversion rate delivers more leads than one getting 400 organic sessions at 2%.

What we're doing next

Three things we're focused on for the next quarter:

1. Schema coverage. Getting FAQPage schema on every piece of evergreen content. The data is clear: schema correlates with citation. Moving from 30.6% to 80%+ coverage is the single highest-leverage GEO play for our own site and for clients.

2. Citation-worthy content architecture. Restructuring key pages to lead with direct answers, not introductions. The first 150 words of any page determine whether an AI engine cites it. We're rewriting introductions to answer the query immediately.

3. Cross-platform monitoring. Building internal tooling to track citations across ChatGPT, Claude, Gemini, and Perplexity — not just Google's AI features. The Google-centric view of GEO is incomplete. Your brand can be invisible in GSC's AI report and cited heavily by ChatGPT.

GEO is not a traffic play. It's a trust play. The numbers are small but the intent is high. And if the trend lines from the past six months hold, the volume gap between traditional organic and AI search is closing faster than most people expect.