For two years the story of AI search has been told in aggregate — clicks are down, traffic is falling — as if the loss were spread evenly across every reader. A Q2 2026 study from Advanced Web Ranking, built on anonymized Google Search Console data across thousands of sites, shows it is not. The decline is concentrated on desktop: top-two-position click-through fell 5.27 points quarter over quarter while mobile actually rose 6.59, and on desktop results pages carrying an AI Overview, the top two positions now take roughly 10% of clicks versus 29% on pages without one. That split matters because desktop is where the research-heavy, commercial-investigation, and B2B queries live — precisely the content that considered-purchase publishers spend the most to produce. This guide reads the desktop numbers honestly, including the logging-error caveat that complicates them, then works through what a click-based content-performance model does when the click is leaving the surface it was built on — why Search Console's new AI report closes only half the gap, why a modest 2026 rebound is not a reason to relax, and how to rebuild content performance around presence and citation rather than the desktop click alone.
Most of the AI-search coverage in 2026 has reported the click decline as a single blended figure — organic CTR is down by some large percentage, traffic is falling. Advanced Web Ranking's Q2 2026 study broke that blend apart, and the split is the story. Working from anonymized Google Search Console data across thousands of sites and millions of search terms, comparing April–June 2026 against January–March, it found the combined click-through rate for the top two organic positions fell 5.27 percentage points on desktop while rising 6.59 on mobile. The loss is not spread evenly across your audience. It is landing on the desktop reader.
The AI Overview effect inside that desktop number is sharper still. On US desktop results pages that carried an AI Overview, positions one and two together took 10.04% of clicks — against 29.05% on desktop pages without one. That is a 19-point gap between an AI-answer page and a classic page, and the study noted the gap had widened by roughly six points since Q1. The general shape of this decline is covered in AI Overviews are reducing organic clicks; what the Q2 desktop data adds is where the loss concentrates and, by extension, which of your content is most exposed.
The decline concentrates where AI answers do the most work. Desktop skews toward longer, research-heavy, and commercial-investigation queries — someone comparing options, reading documentation, planning a purchase at a keyboard rather than thumbing a quick lookup on a phone. Those are exactly the queries an AI Overview or an AI Mode answer resolves in place: the synthesized paragraph answers the question, so the reader never needs the ten blue links beneath it. Mobile's query mix leans more navigational and local, and its interface presents results differently, which is part of why its click share held or rose while desktop's fell.
For a content team, the uncomfortable implication is that the drop is aimed at your most expensive content. The desktop reader is disproportionately the B2B buyer, the considered-purchase researcher, the person reading the 2,000-word comparison you spent real money producing. That is the content whose whole job is to be found during research — and research is the exact query class the AI answer now intercepts. A publisher who only watches blended CTR will see a manageable-looking average and miss that the decline is falling hardest on the audience worth the most per visit.
The desktop numbers come with a real asterisk, and glossing over it would undercut the point. Google disclosed a logging error affecting Search Console impression data from May 13, 2025 through April 27, 2026 — a window that overlaps Q1 2026 entirely and part of Q2, the exact periods the study compares. Impression counts feed the denominator of any CTR calculation, so a distortion there can move the reported shift. Advanced Web Ranking flagged this itself. Read the 5.27-point figure as a strong directional signal about where the decline is concentrated, not as a number to quote to three decimals in a board deck.
This is the right posture for every AI-search statistic in 2026: the direction is well-corroborated across independent datasets even when specific magnitudes wobble with methodology and query set. Amsive's large keyword study put the aggregate decline near 15%, Seer's longitudinal work found declines above 60% on AI-Overview queries, and Ahrefs reported roughly a 58% average CTR drop for top pages using data through December 2025. The point estimates disagree; the direction does not. Desktop clicks on AI-answered queries are down, materially, and the trend has held long enough to plan around. The broader traffic picture, including Google's own framing of it, is worked through in Google AI Overviews and web traffic.
Content performance, as most teams still measure it, is a click-based model wearing several disguises. Pageviews, sessions, time on page, pages per visit, goal completions attributed to organic landing pages — every one of them starts with a click that lands a human on your page. That model was built for a web where the way to consume your content was to arrive at it. On the desktop surface, that assumption is now false often enough to break the model quietly: the content is being consumed, inside the AI answer, by a reader who never becomes a session. Your analytics record a non-event, and a piece that is genuinely doing its job — informing the buyer, shaping the consideration — reads as underperforming.
The failure is worse than a simple undercount because it is biased. It penalizes exactly the content most likely to be extracted into an answer: the clear, well-structured, authoritative, question-answering pages that are best at their job are the ones an answer engine most wants to quote and the reader most easily gets a full answer from without clicking. A click-only scorecard will tell you to make less of your best work. That is the trap, and it is why the fix is not a better attribution model bolted onto clicks — it is a second axis of measurement that does not depend on the click at all. The unified scorecard version of this is laid out in AI search and SEO KPIs.
Google did give publishers a new instrument. The Search Generative AI performance report, rolled out to a subset of sites in June 2026 and worldwide by the end of August, reports impressions inside AI Overviews, AI Mode, and AI features in Discover — broken down by page, country, device, and date, with data beginning mid-May 2026 and no historical backfill. For the first time you can see that a page was shown inside an AI answer. That is genuinely useful, and the workflow for reading it is covered in AI search performance reporting in Google Search Console.
But it closes only half the gap, and it is important to be precise about which half. The report gives you impressions inside AI surfaces; it does not isolate the clicks from them. So it can confirm that your desktop content is being surfaced in the answers — which is the presence half of the story — while staying silent on whether that presence still earns a visit, which is the click half the CTR decline made urgent. Presence without a click count is a real signal, not a consolation prize: it tells you the engine reads and trusts your page enough to show it. It just means the tool that reports your exposure and the tool that reports your clicks are now two different tools, and neither one alone describes performance.
There is a genuine bright spot in the 2026 data, and honesty requires stating it. Seer Interactive's longitudinal tracking shows CTR on AI-Overview queries recovering from a floor near 1.3% in December 2025 to about 2.4% by February 2026 — an 85% proportional gain in two months. Users appear to be re-learning that the citations in an AI answer are clickable, and Google has been iterating on how prominently it surfaces source links. The surface is not a one-way collapse to zero.
The reason not to relax is scale. A recovery from 1.3% to 2.4% is dramatic in relative terms and still a small fraction of the double-digit click rates a top desktop result used to earn. The rebound is best read as the surface stabilizing at a new, lower baseline, not as a path back to the old click economics. Planning as if clicks will return to pre-AI levels because the number is rising again is the same mistake as planning as if they will fall to zero. The realistic plan sits between: fewer clicks per impression on the desktop research surface, permanently, with presence and citation carrying more of the value the click used to.
The practical move is to run two scorecards side by side. Keep the click-based one — it still governs the queries and surfaces where clicking remains the behavior, and abandoning it would blind you to real losses. But add a presence-and-outcome scorecard for the AI surface that never asks 'did they click.' Four measures carry most of it. Impressions inside AI surfaces, from the Search Console report, tell you whether your content is even in the running. Citation — whether an answer engine names or links you when it answers your key questions — tells you whether presence converts to authority; tracking it is the subject of Google AI Mode visibility monitoring.
The other two catch the value that survives the vanished click. Share of voice against your competitors on the specific prompts your buyers ask tells you your relative position on the surface, which matters more than an absolute number that every publisher's is falling. And downstream demand — branded search volume, direct traffic, assisted conversions — captures the reader who consumed you inside an answer, formed an impression, and came back later by name rather than by link. A person who reads your framing in an AI answer and then searches your brand is a performance you earned and a click-only model cannot see. The full transition from page-level to per-asset measurement is worked through in content measurement in AI search.
The strategic conclusion of the desktop CTR data is uncomfortable for a certain kind of publisher: if your content-performance model depends on the desktop click, and the desktop click is where the decline concentrates, then the durable response is to stop being dependent on any single surface for your presence. Being present — and therefore measurable on the presence axis — across every surface an answer engine and an audience actually read is now a production problem before it is a measurement one. You cannot measure share of voice on ten surfaces you do not publish to. This is the specific gap Kompozy is built to close, and the honest framing is that it is a production-and-distribution answer, not a ranking trick.
Kompozy is a full AI content generation and multi-platform publishing engine — 18 output formats fanned across the eight social platforms plus blog and email. Where a single-URL SEO play puts all your presence on one desktop page competing for a click that is getting rarer, Kompozy turns one source idea into a native content set across the surfaces answer engines increasingly pull from: a blog article and newsletter for the owned long-form that Search Console's AI report can track, plus Persona Shorts, carousels, Photo Posts, and text posts on the social platforms that are themselves becoming AI-answer sources. Every piece is held to a Persona Brief and rendered brand-exact through HyperFrames, so a presence spread across many surfaces still reads as one consistent voice — which is what makes it a coherent brand entity an engine cites, rather than scattered posts.
The measurement reframe this guide argues for only works if the underlying presence is producible at cadence, and that is the part Kompozy removes the ceiling on. Being present on eight-plus surfaces every week, on brand, is precisely the workload that keeps most teams stuck on the one desktop page — so the click-based model persists by default because the multi-surface alternative is too much manual work. Autopilot schedules and publishes the approved spread across every destination on its own cadence, behind a per-post review gate, which turns 'be present everywhere the answer engines look' from a staffing problem into a settings decision. Kompozy will not tell you your AI-answer click rate — that is Search Console's and your monitoring tool's job — but it produces the multi-surface presence those tools are built to measure, so your content-performance scorecard has something real to read beyond a falling desktop click.
The Q2 2026 desktop data is the clearest signal yet that the AI-search click decline is not a uniform tax — it is concentrated on the desktop research surface, aimed at the considered-purchase and B2B content that costs the most to make and earns the most per visit. A content-performance model built on the desktop click will misread that content as failing at the exact moment it is doing its job inside an answer. The fix is not to chase the click back; the rebound proves it stabilizes low, not that it returns. The fix is to measure presence and citation beside the click, to accept that Search Console reports your exposure but not your AI clicks, and to make your presence broad enough across surfaces that there is something to measure when the desktop click is gone. Do that and a falling number on one surface stops being the whole verdict on your content.
Advanced Web Ranking's Q2 2026 study — anonymized Google Search Console data across thousands of sites, comparing April–June to January–March — found the combined top-two-position click-through rate on desktop fell 5.27 percentage points quarter over quarter, while mobile rose 6.59. On US desktop results pages that carried an AI Overview, positions one and two took 10.04% of clicks, against 29.05% on pages without one — a 19-point gap that had widened by about six points since Q1.
The decline concentrates where AI answers do the most work. Desktop skews toward longer, research-heavy, and commercial-investigation queries — the kind an AI Overview or AI Mode answer resolves in place, so the reader never needs the link. Mobile's query mix is more navigational and local, and its interface surfaces results differently. The takeaway is not that mobile is safe; it is that the content most exposed to the desktop drop is the considered-purchase and B2B content publishers invest in most.
Not yet. The Search Generative AI performance report Google rolled out worldwide by August 2026 reports impressions inside AI Overviews, AI Mode, and AI features in Discover, broken down by page, country, device, and date — but it does not isolate clicks from those surfaces, and it has no historical backfill before mid-May 2026. It tells you that you were shown, not whether anyone clicked, which is exactly the half of the story the CTR decline makes urgent.
There is a real rebound and it is real reason for caution, not relief. Seer Interactive's longitudinal data shows CTR on AI-Overview queries recovering from a December 2025 floor near 1.3% to about 2.4% by February 2026 — a strong proportional gain that is still a small fraction of pre-AI click rates. Treat the rebound as evidence the surface is stabilizing at a lower baseline, not as a return to the old click economics.
Stop treating the desktop click as the primary success signal and add presence-based metrics beside it: impressions inside AI surfaces, whether an answer engine cites you, share of voice against competitors on your key prompts, and downstream demand you can still see — branded search, direct traffic, and conversions. The click still matters; it just stops being the whole scorecard on the surface where it is falling fastest.
Desktop Google click-through is falling faster than mobile: an Advanced Web Ranking study of Q2 2026 Search Console data found desktop top-two CTR dropped 5.27 points quarter over quarter while mobile rose 6.59, and on desktop pages with an AI Overview the top two positions take roughly 10% of clicks versus 29% without one. Because desktop carries the research and buying queries publishers invest in most, content performance must shift from counting desktop clicks to measuring presence and citations across every surface an answer engine reads.
Get started → · ← All guides · Compare Kompozy vs other tools