Written by: Arjun Karnik, Growth Marketing Specialist
Key Takeaways
- ChatGPT not sending traffic reflects the zero-click era, where AI assistants deliver answers inside their own interfaces.
- Impressions are climbing while clicks fall, with 68% of U.S. Google searches ending without a click and 71% of B2B buyers using AI chatbots for research.
- Success measurement now depends on citations, mentions, and share of answer, because many LLM citations never appear in Google’s top 100 results.
- GA4 undercounts AI traffic because 70.6% of AI-adjacent visits arrive without referrer headers and appear as Direct or Organic Search.
- Arjun Karnik’s fan-out query mapping and freshness loop tactics deliver measurable citation gains. Book a demo to see your current citation footprint.
The Problem: Why Impressions Rise While Clicks Shrink
Pew Research Center tracked 900 U.S. adults across 68,879 Google searches in March 2025 and found a sharp drop in clicks when AI summaries appeared. Users clicked a traditional result 8% of the time with an AI summary, compared with 15% when no summary appeared. SparkToro’s analysis of Similarweb clickstream data showed that 68.01% of U.S. Google searches ended without a click during January–April 2026, up from 60.45% in 2024. That jump is the fastest acceleration in a decade. Only 276 out of every 1,000 Google searches now result in a click to the open web.
The B2B buyer shift follows the same pattern. G2 surveyed 1,076 B2B software buyers and decision-makers across North America, EMEA, and APAC in March 2026 and found that 71% use AI chatbots for software research. More striking, 69% chose a different vendor than originally planned based on what the assistant recommended, and 33% bought from a vendor they had never previously heard of. This progression from research to vendor discovery to purchase shows that being in the answer is a vendor-selection event, not a vanity metric.
The audience scale makes this shift unavoidable. OpenAI reported 900 million weekly active ChatGPT users in February 2026. At Google I/O in May 2026, Sundar Pichai reported more than 2.5 billion monthly active users for AI Overviews and over 1 billion for AI Mode within its first year.
Inside Google Search Console, this shift appears as the scissors pattern. Impressions climb while clicks fall. The content is being read and used to construct answers, but it no longer sends visitors to the site at previous levels. Most owners notice only the half that hurts.
The Solution: Measure Citations and Share of Answer, Not Just Rankings
If traditional metrics no longer capture AI-driven visibility, the measurement framework itself must change. Arjun Karnik runs a public test lab for generative engine optimization under his own name. He documents what gets a business mentioned, cited, and recommended in AI answers and publishes the receipts, misses included. The approach is self-verifying. Ask an AI assistant about his topics and see who gets cited.
The core shift is the measurement target. Rankings measure position on a list buyers increasingly skip. Citations, mentions, and share of answer measure whether the business appears inside the answer the buyer actually receives. Share of Answer is calculated as the percentage of tracked category prompts on which an AI engine names a brand or cites its domain. Strong B2B SaaS programs reach 30–60% on targeted prompts. Weak programs sit at 0–10% regardless of organic ranking.
The structural differences between SEO and GEO explain why the old measurement framework creates false confidence.
| Dimension | SEO | GEO | Why It Matters |
|---|---|---|---|
| Optimizes for | Human-ranked lists and domain authority | Machine retrieval and citation | Each channel targets a different surface |
| Query model | The query the buyer typed | Dozens of hidden fan-out queries triggered by one prompt | Many LLM citations do not rank in Google’s top 100 for the original query |
| Success metric | Rankings | Citations, mentions, share of voice | Rank reports measure a surface buyers often skip |
| Where authority comes from | Backlinks and domain authority | Expert topical coverage and freshness | Brand authority signals correlate more strongly with AI citations than Domain Authority or referring domain counts |
Fan-out query mapping provides a practical starting point. A single buyer prompt triggers dozens of hidden retrieval queries. In a test on Arjun’s site, pages rewritten to match extracted ChatGPT fan-out queries earned citations while control pages did not. The before state used practitioner jargon and focused on the visible keyword. The after state aligned URLs, titles, H1s, and H2s to buyer-language questions the model actually retrieved against. Citations followed.
Why GA4 Misses Most AI-Influenced Traffic
The GA4 attribution gap is structural. Configuration tweaks cannot fully fix it. An analysis of nearly 450,000 AI-adjacent visits found that 70.6% arrived without referrer headers and were recorded as Direct in GA4. When an AI answer mentions a brand without a clickable link, users often run a follow-up Google search for the brand. GA4 then attributes the visit to Organic Search, while the AI discovery that caused it leaves no trace.
| Scenario | What GA4 Records | What Actually Happened |
|---|---|---|
| User clicks a ChatGPT citation link on desktop | Referral from chatgpt.com (or AI Assistant channel if GA4 recognizes it) | Direct AI referral |
| User opens link in ChatGPT mobile app | Direct (referrer stripped by in-app browser) | AI referral |
| User reads AI answer, then types brand into Google | Organic Search (branded) | AI-influenced brand search |
| User copies brand name from AI answer, then types URL directly | Direct | AI-influenced direct visit |
The GA4 AI Assistant channel should be treated as the visible floor, not the full AI-influenced demand signal. It measures only recognized AI assistant referrals and misses zero-click citation exposure, AI Overview impressions that stay inside Google, answer-assisted branded search, and sessions collapsed into Direct when source data disappears. Every measured value is a floor. The practical response is to instrument for citations and share of answer instead of grading a channel on clicks it no longer sends.
The most measurable scenario is desktop citation clicks, although even these undercount mobile traffic. For mobile app referrals, GA4’s in-app browsers strip referrer information, which prevents accurate attribution. When users read an AI answer and then search for your brand, monitor branded search volume lift in Search Console as a proxy signal. For direct visits that follow AI exposure, no fix exists, so treat all measured AI traffic as a conservative baseline.
The Mechanics Behind Arjun Karnik’s GEO Results
Content decay quietly erodes AI visibility. In Arjun’s tests, pages dropped 78% to 99% in two months without updates. That decay remains invisible unless you instrument for it, and by the time it appears in a monthly report, the position has already vanished. Seer Interactive analyzed 47,097 AI citations across 7,683 pages in ChatGPT, Gemini, and Perplexity between March and June 2026 and found that 75% of cited pages had been updated within the last year. Pages cited consistently across all four months averaged under six months since their last update. The page you refreshed beats the page you wrote.
Content freshness accounts for 40% of Perplexity’s ranking signal, and pages under 30 days old receive 3.2× more citations than older content. More than three quarters of pages cited by ChatGPT were updated within the prior 30 days.
The mechanics Arjun runs through AI Growth Agent address this decay directly.
- Fan-out citation test: Pages rewritten to match extracted ChatGPT fan-out queries earned citations, while control pages did not. The rewrite covered URLs, titles, H1s, and H2s, not just body copy.
- Buyer-language alignment: A page titled “What is GEO” was relabelled “How to Get Your Business Recommended by AI Search,” with slug, title, H1, and H2s realigned to buyer questions. Citations followed within weeks of that specific change.
- Impression-decay tripwires: Automated triggers wired to Search Console signals queue content updates when performance drops, using the 78–99% decay range as the trigger band. Content repairs itself on a loop instead of waiting for a quarterly audit.
- Structured publishing at machine cadence: AI Growth Agent runs 5–8 autonomous actions per day, including new articles and updates. On Arjun’s site, the GEO subfolder went from zero to the only source of new impressions on the entire domain in 60 days. New articles reached thousands of monthly Google impressions within weeks.
AI Growth Agent clients average more than 12,000 additional AI citations and mentions, over 100,000 additional bot visits, and a 20% or greater lift in impressions across the first twelve weeks. These are AI Growth Agent’s published client results.
See how the fan-out mapping and freshness loop work in practice. Book a demo.
FAQ: Common GEO Questions From B2B Teams
Why does my competitor show up in ChatGPT and I do not?
The most common reason is structural, not qualitative. AI retrieval systems match a question to the best available answer across dozens of hidden fan-out queries triggered by a single buyer prompt. If a competitor’s pages use buyer-language questions, carry schema markup, and have been updated recently, they win the retrieval match even when your domain authority is higher.
Freshness comes next. A competitor that refreshes content on a regular cadence will consistently outperform a better-written but stale library, because the game resets weekly. Topical coverage then decides the long tail. If a competitor has structured answers across the full question space in your category and you have covered only the head terms, the model cites them for the long-tail queries where most buyer research starts.
Why does AI ignore my business even though I rank on page one?
Traditional Google rankings and AI citations behave as largely independent channels. Only a minority of AI Overview citations overlap with traditional Google page-one rankings, and many LLM citations never appear in the top 100 Google results. Ranking measures position on a list. Citation measures whether a model can extract a clean, structured, current answer from your page and match it to a buyer’s question.
A page can rank first and still go uncited if it uses practitioner jargon instead of buyer language, lacks schema, has not been updated recently, or fails to answer the fan-out queries the model retrieves against. Backlinks do not fix this gap. Restructuring the page to answer the model’s questions in the words buyers use does.
How do I measure AI visibility when GA4 undercounts it?
Use three measurement layers at the same time. First, track the GA4 AI Assistant channel and any custom channel group you build with regex to consolidate fragmented AI referrers. Treat this as the visible floor, not the ceiling.
Second, monitor branded search volume in Google Search Console as a proxy for AI-influenced discovery. When an AI answer names your business, a share of buyers will search your brand name on Google, and that lift remains measurable even when the original AI referral does not.
Third, run share-of-answer tracking across ChatGPT, Google AI Overviews, Perplexity, and Gemini using a fixed set of 40–75 buyer queries on a consistent cadence. Citation frequency on that prompt set becomes the headline metric that replaces rank position. Keep one honest caveat in mind. Buyers often copy a name from an AI answer and type it directly into a browser, which appears as direct traffic and never receives attribution. Everything you measure is a floor.
Should I stop doing SEO and focus only on GEO?
SEO remains essential. Technical fundamentals, structured content, and freshness serve both channels. Content built for AI citation still performs in Google organic search. On Arjun’s site, GEO-focused articles reached thousands of monthly Google impressions within weeks, and the GEO subfolder became the only source of new impressions on the domain.
What changes is the target and the metric. The measurement target moves from rankings to citations, mentions, and share of voice. The content structure moves from keyword-density tactics to answer-first formatting with schema on everything. The cadence moves from a quarterly content calendar to continuous publishing and refreshing at machine cadence. Traditional SEO plumbing such as crawlability, site structure, and internal linking remains foundational. The rank report retires as the primary success metric.
Next Steps: A Checklist You Can Test in Your Own Stack
The following checklist covers the sequence Arjun runs on his site. Each step is independently testable. Start with technical plumbing, because nothing downstream works if the retrieval layer cannot read the site.
- Unblock AI crawlers. Check robots.txt for rules that block ChatGPT-User, PerplexityBot, ClaudeBot, and Google-Extended. This silent blocker appears frequently and should be fixed first.
- Add schema markup to everything. Article, FAQ, HowTo, and Organization schema act as structural requirements, not enhancements. Pages without schema are harder for retrieval systems to parse.
- Run a baseline visibility audit. Query ChatGPT, Gemini, Perplexity, and Google AI Overviews with your 10–15 most important buyer questions. Document who gets cited. This becomes your control group for every later test.
- Extract fan-out queries directly from ChatGPT. For each buyer prompt, note the sub-questions the model asks or implies. These sub-questions are the retrieval queries your pages need to answer, not the visible keyword.
- Rewrite URLs, titles, H1s, and H2s in buyer language. Replace practitioner jargon with the words buyers use. Test one page at a time and hold a control. In Arjun’s tests, citations followed within weeks of this specific change.
- Set impression-decay tripwires in Search Console. Define a percentage drop over a rolling window that triggers a content update queue. Use the 78–99% decay range described earlier as your trigger point. Monthly audits catch this too late.
- Publish and refresh at machine cadence. The channel requires continuous new articles plus continuous updates to existing ones. AI Growth Agent runs 5–8 autonomous actions per day on Arjun’s site. A human team writing 7–10 articles a month without a refresh loop cannot match the freshness math.
- Set up share-of-answer tracking. Build a fixed prompt set of 40–75 buyer queries. Track citation frequency across all four surfaces on a consistent cadence. Treat this metric as the replacement for rank position.
- Monitor branded search volume as an AI proxy. A lift in branded searches in Search Console, without a corresponding campaign, provides the most reliable signal that AI answers are naming you and buyers are following up.
- Audit what AI currently says about you. Run a defensive GEO check before any growth work. A wrong AI answer hurts more than no answer. Correct the record first.
AI answers are being written right now. Every week that passes without structured, fresh, buyer-language content in the retrieval layer is a week when a competitor’s answer hardens into the default. Early citations become tomorrow’s record, and the cost of entry rises as settled answers gain incumbency.
See how to get your business into the record AI reads from. Book a demo.
