Automotive Search Marketing · Second Edition · Chapter 15 # Measuring AI Search: Search Console, the GA4 “Direct Traffic” Mystery, and Your AI Readiness Checklist By Alex Melen. The full chapter as printed, followed by notes from the SmartSites accounts and reader questions, updated Oct 20, 2026. Or [get the book](https://amzn.to/3VJsZxD) for all 20 chapters. [Buy on Amazon](https://amzn.to/3VJsZxD) In this chapter - [Impressions Up, Clicks Down: The New Normal in Search Console](#impressions-up-clicks-down-the-new-normal-in-search-console) - [Google’s Generative AI Report: Your AI Scoreboard](#google-s-generative-ai-report-your-ai-scoreboard) - [The GA4 “Direct Traffic” Mystery](#the-ga4-direct-traffic-mystery) - [Free First-Party Data: Bing Webmaster Tools](#free-first-party-data-bing-webmaster-tools) - [The Monthly AI Spot-Check](#the-monthly-ai-spot-check) - [Your AI Search Readiness Checklist](#your-ai-search-readiness-checklist) - [Questions to Ask About AI Search (Before You Pay Anyone for It)](#questions-to-ask-about-ai-search-before-you-pay-anyone-for-it) - [The Road Ahead](#the-road-ahead) - [Notes from the accounts](#notes-from-the-accounts) - [Reader questions](#faq) Free chapters - [What Is AI Search?](https://www.alexmelen.com/book/chapters/what-is-ai-search/) - [Winning AI Search: GEO for Your Dealership](https://www.alexmelen.com/book/chapters/geo-for-car-dealerships/) - [AI Search and Your Ad Dollars](https://www.alexmelen.com/book/chapters/chatgpt-ads-and-ai-search-advertising/) - **Measuring AI Search** **Automotive Search Marketing** A Car Dealer's Guide to Google & AI Search · Second Edition [Buy on Amazon](https://amzn.to/3VJsZxD) Why this chapter is free Every chapter in this book ends up at the same place: measure it, or you will be sold a story about it. AI search makes that harder than it has ever been, which is why this chapter is free. It is the one I most want dealers to read before they sit down with a vendor. After the chapter text you will find the measurement findings that have come in since the manuscript was locked, including the one that I now open most talks with, and the questions I am asked about tracking AI traffic. The readiness checklist at the end of the chapter is still the single most useful page in the book; print it. The chapter text below is reproduced from the printed book; the sections after it are web-only and dated. If you take one thing from this book’s measurement chapters, it’s that being data-driven is your best defense against wasted spend and vendor snake oil. AI search makes measurement genuinely harder—but if you know what to look for, the picture is clear. There are three things you need to understand. ## Impressions Up, Clicks Down: The New Normal in Search Console Open Google Search Console for almost any dealership in 2026 and you’ll see the same puzzling chart: impressions climbing, clicks flat or falling. In the old world, that meant something was wrong—maybe your titles were weak, maybe a competitor out-copied you. In the AI world, it’s often the expected signature: your pages are being surfaced and cited in AI Overviews (which counts as an impression) without generating a click. Google folded AI Mode data into Search Console’s performance report without a separate filter, so the AI effect is baked into your main numbers whether you like it or not—though a new report, covered next, finally breaks part of it out. Clicks still matter—but raw click counts no longer tell the whole story of your search visibility. Judge your SEO (and your agency) on the full picture: impressions, the intent mix of the clicks you do get, and most importantly, the leads and sales those clicks produce. ## Google’s Generative AI Report: Your AI Scoreboard In June 2026, Google finally gave site owners a direct look behind the AI curtain: a dedicated Generative AI performance report inside Search Console. It shows how many times links to your site appeared inside Google’s generative AI experiences—AI Overviews and AI Mode—broken out by page, country, and device. After two years of squinting at blended numbers and third-party estimates, this is first-party data from Google itself, and it answers the question every dealer asks me first: are we showing up in the AI answers at all? The report is young, and it shows. As I write this it reports impressions only—no queries yet, and no separate click data—and Google has been rolling it out to properties gradually, so don’t panic if yours doesn’t have it yet. But even with those limits, this is the AI-era scoreboard I’d check monthly: a rising line means your pages are being pulled into AI answers, the page-level view shows which content is earning it, and a flat line means the GEO work hasn’t landed yet. And when the content standard from Chapter 3 is working, this report is where you see it first. In late July 2026 we began shipping a set of model-research pages for a Honda dealer client in Virginia—answer-first, real local pricing, spec tables, customer review quotes, FAQ markup, all of it. Over that summer, the store’s weekly impressions inside Google’s AI features nearly quadrupled, from roughly 350 a week in late June to a peak of about 1,400 by mid-August—more than 6,600 AI-feature impressions for the summer, with the curve visibly steepening as the new pages entered the index. Every one of those is a shopper seeing this dealership surfaced inside an AI answer—visibility the referral reports we’re about to look at would have missed entirely. Pair this report with the Bing Webmaster Tools AI Performance report (coming up in a moment) and you have free, first-party AI visibility data across both major ecosystems—something that simply didn’t exist a year ago. Neither replaces the monthly spot-check later in this chapter, because impressions tell you that you appeared, not what was said. But together they turn “trust me, the AI likes you” into a chart you can pull up yourself. ## The GA4 “Direct Traffic” Mystery Here’s a puzzle we solved with our own data at SmartSites, and it’s one of my favorite examples of what having a lot of data—and AI to analyze it—makes possible. Through 2025 and 2026, dealerships everywhere watched the same pattern in GA4: organic traffic falling, direct traffic mysteriously rising. Direct traffic is supposed to mean people typing your website address or using a bookmark. And I kept asking the obvious question: do we really believe that, suddenly, twice as many people are typing dealership URLs into their browsers? Of course not. That behavior doesn’t change overnight. So we tested it against our data—nearly 1,500 client GA4 properties, year over year. Direct sessions jumped about 66 percent across the board. And here’s the decisive part: the jump was almost identical for clients whose SEO we manage and clients where we don’t touch SEO at all. When something hits everyone equally regardless of what anyone did, it’s not a performance story—it’s a measurement artifact. What’s actually happening: when someone clicks a citation inside an AI Overview, or follows a link from ChatGPT, the referrer information often gets stripped along the way. GA4 sees a visitor with no origin and files it under Direct. Your SEO and GEO work is generating visits that your analytics can’t credit. Google eventually acknowledged the problem and gave AI traffic its own channel—more on that in a moment—but the fix catches only a sliver of what’s really happening. One honest caveat, because I promised you education, not spin: it’s not all mislabeling. Real search clicks (per Search Console, which measures at the source) declined too—that’s the genuine zero-click effect we covered earlier. And notably, dealerships are the vertical where the least lost organic “reappears” as direct—because on car-shopping queries, the AI so often answers completely and the click never happens at all. Which brings the whole chapter full circle: for a dealership, the answer to AI search isn’t recovering the click. It’s being inside the answer. What to actually do in GA4: first, know that Google finally gave AI traffic its own bucket—in May 2026, GA4 added a built-in “AI Assistant” channel that automatically labels visitors arriving from ChatGPT, Gemini, Copilot, DeepSeek, and Grok. It’s a start, but it catches only a sliver of the real AI-driven traffic (a fraction of a percent of what sits in Direct), misses some assistants (Perplexity, for one) entirely, and only counts forward from its launch date. One more gap to know about: clicks from Google’s own AI surfaces—AI Overviews and AI Mode—never show up in this channel at all; Google counts those as organic search, so your “AI traffic” bucket understates the biggest AI surface of them all. (The place to see that footprint is the Generative AI report we just covered.) So still have your team or agency build a custom channel group that catches what the default rules miss; watch “Direct” visitors who land deep on inventory or research pages (real type-in traffic goes to your homepage—deep-page direct landings from new visitors are almost always re-routed search/AI clicks); and stop treating GA4’s organic number as your SEO scoreboard—use Search Console clicks plus impressions as ground truth, with leads and sales as the final word. ## Free First-Party Data: Bing Webmaster Tools Here’s a measurement bright spot most dealers have never heard of. In February 2026, Microsoft launched an “AI Performance” report inside Bing Webmaster Tools—the first time a major platform has shown site owners, from its own first-party data, exactly where their content is being cited in AI answers. The report covers Copilot, Bing’s AI summaries, and Microsoft’s partner AI experiences: which of your pages get cited, what questions triggered those citations, and how it all trends over time. By June 2026, Microsoft had added intent and topic breakdowns and even a “citation share” metric—your slice of the citations for a given question versus everyone else’s. And it’s completely free. The catch: Copilot and Bing are a much smaller slice of AI search than Google’s AI Overviews, so treat this as directional insight rather than the full picture. But directional first-party data beats third-party guesswork every time. If your dealership’s website isn’t verified in Bing Webmaster Tools yet, this report alone is worth the ten minutes it takes. ## The Monthly AI Spot-Check Finally, the free habit every dealership should build. Once a month, ask the major AI tools (ChatGPT, Gemini, and Google’s AI Mode) the questions your buyers actually ask: “best [your brand] dealer near [your city],” “most reliable used SUV under $25,000,” “should I buy from [your dealership]?” Ask each question several times, in different phrasings—remember, single answers are random; patterns are real. Log three things: how often you appear, what the AI says about you, and which competitors keep showing up. It costs nothing, takes half an hour, and tells you more about your AI visibility than most dashboards. (Yes, paid AI-visibility tracking tools are emerging—we touch on them in Useful Tools—but for a single-rooftop dealership, this manual routine is where to start.) ## Your AI Search Readiness Checklist We covered a lot in this chapter, so here it is as a single checklist. My suggestion: hand this page to whoever runs your marketing—in-house or agency—and ask them to score your dealership honestly on each item. (In the first edition, I told you a good agency should welcome accountability. That hasn’t changed.) - AI crawlers (GPTBot, ClaudeBot, PerplexityBot, Google-Extended) are allowed in robots.txt. - Your vehicle data and key page content are readable without JavaScript, and you’ve verified it. - Structured data is in place: AutoDealer/LocalBusiness, vehicle schema, FAQ schema on key pages. - Every service and department has its own dedicated, detailed page. - Key pages answer the core question in the first sentence, with specifics and tables. - You’re producing the two proven AI-era content formats: numbered buying guides and vehicle attribute pages—and you’ve stopped funding generic how-to content. - Your business facts (name, address, phone, hours, brands) are identical everywhere on the web. - You have an active plan for earning mentions in “best of” lists, local press, and industry sites. - Review generation is running continuously, with prompts that produce specific, quotable reviews—for sales and service. - Your Merchant Center vehicle feed is clean, complete, and updating daily. - GA4 tracks your AI referrals (the built-in AI Assistant channel plus a custom channel group for what it misses), and you judge SEO by Search Console (including its Generative AI report) plus leads—not raw GA4 organic. - Someone runs the monthly AI spot-check (multiple questions, multiple runs, multiple platforms) and logs visibility over time. Score twelve out of twelve and your dealership is ahead of the overwhelming majority of the industry—including plenty of stores twice your size. ## Questions to Ask About AI Search (Before You Pay Anyone for It) Just like I gave you questions for vetting SEO and PPC vendors, here’s the AI-era set. These five questions will separate the professionals from the snake oil in the first ten minutes: - “How do you measure AI visibility?” (Right answer: presence rate across many prompts and runs. Wrong answer: any promise about “ranking #1 in ChatGPT.”) - “What exactly would you change on our website, and why?” (Right answer: specifics—crawler access, rendering, schema, dedicated pages, answer-first content. Wrong answer: proprietary magic they can’t explain.) - “How does this build on our existing SEO instead of replacing it?” (Right answer: GEO is a layer on the same foundation. Anyone selling you GEO instead of SEO doesn’t understand either.) - “Show me a before-and-after for a client.” (Same as it ever was: real work has receipts.) - “What happens to this strategy when the platforms change next quarter?” (Right answer: humility plus fundamentals. This space changes monthly, and anyone who claims certainty is selling it.) ## The Road Ahead I’ll close this chapter the way I’ve tried to write this whole book: honest about what I know, and honest about what nobody knows. Nobody knows exactly what search looks like in five years. When I wrote the first edition of this book, GA4 was brand new, AI Overviews didn’t exist, and ChatGPT hadn’t launched. Think about what that means: I wrote an entire book the old way—typing every letter myself, one at a time, like some kind of animal. I personally strategized over how many chapters it needed and how best to present everything you’ve just read, with no AI to bounce a single idea off. A strange world—and at the time, we didn’t even know it. I wrote then that in a few years we’d look back on Universal Analytics as archaic—that prediction aged well, and it aged well because it was based on a direction, not a date. So here’s my direction-based prediction for this edition: AI will keep moving deeper into every step of the car buyer’s journey—from research, to comparison, to eventually contacting dealerships and negotiating on the buyer’s behalf. EMARKETER frames it well in their Digital Shelf 2026 report: brands now have to win over three kinds of shoppers—humans, AI-assisted humans, and eventually AI agents shopping on a human’s behalf. Hybrid AI-assisted shopping is already mainstream; the “agentic” era, where a customer’s AI assistant calls your BDC or fills out your lead form for them, is closer than most dealers think. The dealerships that are visible and citable—and easy for an AI to do business with— will compound their advantage every year. The plumbing for that future is being built right now: in late 2025, OpenAI launched Instant Checkout in ChatGPT on an open standard called the Agentic Commerce Protocol, letting users complete purchases from major retailers without ever leaving the conversation—and Google and the payments industry are building their own equivalents. Nobody is buying a car inside ChatGPT tomorrow. But “agent-ready” is worth translating into dealership terms today, because every piece of it is something this book has already asked you to do: inventory and pricing that are machine-readable (your feeds and your schema), a service-booking path a machine can complete (a clean scheduling link, not a phone-only process), lead forms short and simple enough for an assistant to fill out, and business facts consistent enough everywhere that an agent can trust them. When the first customer’s AI assistant contacts your store—and it will—the dealership that answers cleanly wins the deal the same way the fastest BDC rep wins it today. But here’s what I’m even more confident about, because it’s been true through every disruption I’ve watched since the mid-1990s: the fundamentals outlast the formats. Be where your customers look. Give them genuinely useful information. Make every fact about your business consistent and easy to verify. Measure what actually matters, and demand transparency from everyone you pay. That playbook won the directory era, the search era, and the mobile era. It will win the AI era too. The tools changed. The game didn’t. ## Notes from the accounts (Oct 20, 2026) What has come in from SmartSites' client data and Alex's 2026 talks since the manuscript was locked. Not in the printed book; revised as the platforms change. Sources and tables are on the [research page](https://www.alexmelen.com/research/). ### AI Overview clicks are landing in Direct, not Organic The chapter explains the GA4 "direct traffic mystery" from the data available when it was written. We have since confirmed the mechanism specifically for Google's AI Overviews: measuring across SmartSites' GA4 properties, clicks out of an AI Overview are not being attributed to organic search or any other channel; they are being filed as Direct. Across roughly 1,500 client properties, Direct sessions rose about 66 percent year over year, and the lift was the same whether or not we manage the client's SEO. The tell is new visitors landing deep on inventory or research pages; genuine type-in traffic goes to the homepage. ### The measurement framework I now use on stage Three layers, in order. First, presence: ask the questions your buyers ask, in several phrasings, several times, across ChatGPT, Gemini and Google AI Mode, and log how often you appear and what is said; this is the only honest replacement for rank, because there is no rank. Second, first-party scoreboards: Search Console's Generative AI report for Google's AI surfaces and Bing Webmaster Tools' AI Performance report for Copilot. Third, outcomes: leads and sales, with AI referral and misattributed Direct traffic examined together. Judge an agency on all three, not on a dashboard that claims a position in ChatGPT. ### No, FAQ schema will not fix this Dealers often ask whether a particular piece of markup will get them cited. We tested it: 500 client sites with FAQ structured data against 500 without showed no statistically significant difference in AI citations. The markup is cheap and we keep it, but the levers that moved citations were first-sentence answers, clear entity information and mentions on third-party sites. Spend your time there. ### A reminder from the SparkToro data In Rand Fishkin's January 2026 research (600 volunteers, roughly 3,000 prompt runs), the chance of the same shopping prompt returning the same brand list twice was under one in a hundred. Anyone who shows you a single screenshot of your dealership missing from an AI answer, or present in one, has shown you nothing. Patterns across many runs are the data; a screenshot is an anecdote. ## Reader questions **How do I see AI search traffic in Google Analytics 4?** Partially. GA4's built-in AI Assistant channel labels visits from ChatGPT, Gemini, Copilot, DeepSeek and Grok, but it misses Perplexity, counts only from its launch date, and never sees clicks from Google's own AI Overviews or AI Mode, which Google counts as organic or which lose their referrer and land in Direct. Build a custom channel group for the referrers it misses, watch for new visitors landing deep on product or research pages from Direct, and use Search Console's Generative AI report as the first-party view of Google's AI surfaces. **Why are my Search Console impressions up but clicks down?** Because being cited inside an AI Overview counts as an impression without producing a click, and Google folds AI Mode data into the main performance report with no filter. In the old world that pattern meant weak titles; in 2026 it is usually the signature of pages being read inside AI answers. Check the Generative AI report for the AI-feature impressions, look at which pages are earning them, and judge the work by leads and sales rather than raw clicks. **What is the right way to measure AI visibility for a dealership?** Presence rate, not position. Once a month, ask ChatGPT, Gemini and Google AI Mode the questions your buyers actually ask, in several phrasings, several times each, and log how often your dealership appears, what is said about it, and which competitors keep showing up. Pair that with Search Console's Generative AI report and Bing Webmaster Tools' AI Performance report, both free and first-party, and tie the whole thing back to leads. Any tool or vendor promising a ranking position inside ChatGPT is selling something that cannot exist. Keep reading This is Chapter 15 of *Automotive Search Marketing: A Car Dealer's Guide to Google & AI Search* (Second Edition, 2026). The book covers SEO, PPC, AI search, measurement and choosing an agency in 20 chapters. [Buy on Amazon](https://amzn.to/3VJsZxD)[More excerpts](https://www.alexmelen.com/book/inside/) Copyright © 2023, 2026 Alex Melen. Quotations of brief passages with attribution and a link are welcome. Chapter text as printed; web sections last updated Oct 20, 2026. The AI platforms change quickly, so this page is revised as they do. ## Want this delivered to your team in person? Alex presents the AI search material as a keynote, dealer-group session or 20 Group workshop, tailored to your market and brands. [Check availability](https://www.alexmelen.com/contact/) [Speaker kit](https://www.alexmelen.com/speaking/#kit) --- Source: https://www.alexmelen.com/book/chapters/measuring-ai-search/ · Markdown version of this page for AI agents and LLMs. Site guide: https://www.alexmelen.com/llms.txt