SXO Intelligence · Data Report
AI search vs Google: where the queries went — and where the clicks didn’t
Every number below is sourced and dated. Where a monthly series was never published, the chart says so instead of drawing a smooth line through a guess.
One correction first
The premise needs one fix — and it’s good news.
Google’s search volume is not dropping. It is growing. Google went from ~13.7 billion searches a day (Jan 2025) to ~16.4 billion a day (May 2026). Organic traffic from Google to websites is down only −2.5% year over year. There is no collapse in Google demand to chart.
What is collapsing is the click. Google now answers the query on its own page: 68.01% of US searches end with no click to anyone, and when an AI Overview appears organic CTR falls ~61%. Same searches. Fewer exits.
Meanwhile AI is winning queries, not clicks. AI prompting is roughly 28% the size of search worldwide, but AI sends just 1.08% of all website traffic — and the clicks it does send convert 4.4× to 23× better than organic. Enormous attention, a tiny pipe, exceptional quality. That gap is the whole opportunity.
Right now — July 2026
Nine figures, each with the publisher and the date it was published. Check any of them.
Google searches ending with no click to any website (US)
68.01%
from 58.5% in 2024
SparkToro analysis of Similarweb clickstream, Jan–Apr 2026
Google searches per day
16.4B
from 13.7B (Jan 2025)
Demandsage, May 2026 — volume is up, not down
Google queries showing an AI Overview
48%
+58% year over year
Advanced Web Ranking / Digital Applied, Mar 2026. US-only methods: 60.32%
Organic CTR loss when an AI Overview is present
−61%
1.76% → 0.61%
Seer Interactive, Sep 2025. Ahrefs: position 1 loses 34.5% across 300k keywords
AI prompting as a share of search volume (worldwide)
~28%
12% in the US
Graphite, Mar 2026 — search-classified prompts only
AI referral traffic as a share of all website traffic
1.08%
vs 25% from organic search
Conductor, 2026. Google-to-ChatGPT referral ratio ≈ 190:1
AI visitor conversion vs organic visitor
up to 23×
0.5% of visitors → 12.1% of signups
Ahrefs, 2025. Semrush: 4.4×; Similarweb: 11.4% vs 5.3% (ecommerce)
ChatGPT citation rate (links shown in answers)
6.8%
from 1.6% (Jun 2025) — 4.25× in one year
Similarweb, May 2026
Generative-AI monthly web visits (worldwide)
9.5B
+70% YoY · 655M unique visitors (+57%)
Similarweb, average Jun 2025 – May 2026
The actual drop-off
Clicks, not queries
This is the “Google decline” people are describing — and it is a click decline, not a volume decline.
Note the methodology break: the 2019–2020 readings come from different clickstream panels than the 2024–2026 readings, so read the trend, not the point-to-point deltas. SparkToro characterises the recent move as a 22.9% decline in click-generating searches.
Share of Google searches that end without a click
MeasuredThis is the "Google decline" everyone describes — and it is a CLICK decline, not a volume decline. Note the methodology break: the 2019–2020 readings come from different clickstream panels than 2024–2026, so read the trend, not the point-to-point deltas.
Source · Jumpshot · Similarweb · Datos · SparkToro/Similarweb. Hollow markers and dashed runs mark interpolation or unpublished periods — we do not draw a smooth line through a guess.
The takeaway: Two thirds of searches now end without a click, and nearly half show an AI summary. The ranking that matters is being the source that summary cites.
The mechanism
AI Overviews are what took the click
Only two points in this window were ever published: Dec 2025 (34.5%) and Mar 2026 (48%). Those are the solid dots. The line between them is straight-line interpolation, and after March the dash means no data published — not “it plateaued.”
Anyone showing you a smooth monthly AI-Overview curve for this window is drawing, not measuring.
Google queries showing an AI Overview
2 measured anchors + interpolationThis is the mechanism behind the click loss. Only Dec 2025 (34.5%) and Mar 2026 (48%) were ever published — those are the solid dots. The line between them is straight-line interpolation, and after March the dashed run means NO DATA PUBLISHED, not "it plateaued". Anyone showing you a smooth monthly curve here is drawing, not measuring.
Source · Advanced Web Ranking / Digital Applied. Hollow markers and dashed runs mark interpolation or unpublished periods — we do not draw a smooth line through a guess.
The takeaway: Two thirds of searches now end without a click, and nearly half show an AI summary. The ranking that matters is being the source that summary cites.
The counterintuitive part
Google’s share of referrals went up
Google’s share of the clicks that do happen rose 8 points in five months. Both things are true at once: Google sends a smaller slice of its own searches onward, yet still dominates referrals — because AI platforms send so little traffic that they cannot take share.
AI is winning attention. It has not yet won distribution. Which is exactly why the window is still open.
Google's share of all referral traffic to websites
MeasuredGoogle's share of the clicks that DO happen went UP 8 points in five months. Both things are true at once: Google sends a smaller slice of its own searches onward, yet still dominates referrals — because AI platforms send so little traffic they cannot take share. AI is winning attention. It has not won distribution.
Source · Cloudflare Radar, REFERER dimension, weekly rolled to months. Hollow markers and dashed runs mark interpolation or unpublished periods — we do not draw a smooth line through a guess.
The takeaway: Two thirds of searches now end without a click, and nearly half show an AI summary. The ranking that matters is being the source that summary cites.
Inside AI referrals
Who is actually sending the traffic
“Optimise for AI search” is not one target. ChatGPT lost ~10 points of AI referral share in four months while Claude gained ~7. If your strategy is ChatGPT-shaped, it is already dated — which is precisely why citation-worthy structure beats chasing one engine’s quirks.
Click a platform in the legend to isolate it.
Share of AI referral traffic by platform
Measured"Optimise for AI search" is not one target. ChatGPT lost ~10 points of AI referral share in four months while Claude gained ~7. If your strategy is ChatGPT-shaped, it is already dated — which is why we build for citation-worthy structure rather than one engine's quirks.
Source · Goodie GA4 brand panel, 3-month centred rolling average. Hollow markers and dashed runs mark interpolation or unpublished periods — we do not draw a smooth line through a guess.
The takeaway: Two thirds of searches now end without a click, and nearly half show an AI summary. The ranking that matters is being the source that summary cites.
What this means for you
Stop optimising for the click. Optimise to be the answer.
With 68% of searches ending click-free and AI Overviews on 48% of queries, the ranking that matters is being the source the summary cites. FAQ schema, definitional first sentences and comparison tables are built to win exactly that.
The AI traffic gap is an arbitrage, and it closes.
AI is ~28% the size of search but sends 1.08% of traffic — while converting up to 23× better. ChatGPT’s citation rate went 1.6% → 6.8% in a year. As citation rates climb, that thin pipe widens fast, and it pays out to whoever is already cited. Being early here is cheaper than being early to Google ever was.
Your robots.txt is a real competitive asset.
Explicitly allowlist the AI crawlers. Most competitors block them by accident. You cannot be cited by an engine that cannot read you — and this is the cheapest advantage on the entire list.
Diversify past ChatGPT.
ChatGPT fell from 72.5% to 62.6% of AI referrals in four months while Claude went 11.8% → 18.5%. Track citations across all five engines, not one.
None of this is measurable without analytics.
You cannot see AI referrals you never collected, and AI platforms do not append utm_source. Install analytics before this window closes — the data is not backfillable.
Methodology & sourcesHow this was built, what is measured, and what is not.Show
The honest limitation
There is no authoritative public dataset reporting “AI search volume and clicks” against “Google search volume and clicks” on a common monthly basis. The two systems do not share a unit: Google counts queries and outbound clicks; AI platforms count prompts and citations. Any chart claiming a clean monthly head-to-head for Jan–Jun 2026 is modelled, whether or not it says so.
So this report does two things instead. It plots the series that were genuinely measured monthly — Google referral share, and the AI platform mix. For everything else it shows the measured anchor points, and marks the interpolation as interpolation.
Measured monthly series
- Google referral share, Jan–May 2026 — Cloudflare Radar, REFERER dimension, weekly timeseries rolled up to months. ChatGPT, Gemini and Perplexity fall below Radar’s reporting threshold, so no comparable monthly AI line exists from this source.
- AI platform referral mix, Jan–Apr 2026 — Goodie GA4 brand panel, 3-month centred rolling average. May is excluded as a partial month; no June data was published.
Anchor points (charted as points, not curves)
- Zero-click: 50.33% (2019, Jumpshot) · 64.82% (2020, Similarweb) · 58.5% (2024, Datos) · 68.01% (Jan–Apr 2026, SparkToro/Similarweb). Mixed panels — read the trend, not the point-to-point deltas. SparkToro characterises the recent move as a 22.9% decline in click-generating searches.
- AI Overview prevalence: 31% (Feb 2025) · 34.5% (Dec 2025) · 48% (Mar 2026), Advanced Web Ranking / Digital Applied. A separate US-only method reports 60.32% for 2026 — a different measurement basis, not a later reading, so it is not plotted on the same line.
Where the “right now” tiles come from
- 16.4B searches/day — Demandsage, May 2026 · 13.7B/day baseline — Google, Jan 2025 · −2.5% YoY organic traffic — Graphite, Jan 2026
- 28% / 12% AI prompt share — Graphite, Mar 2026 · 1.08% AI referral share — Conductor, 2026
- −61% CTR (1.76% → 0.61%) — Seer Interactive, Sep 2025 · 34.5% position-1 loss — Ahrefs, 2025 · 19.98% non-branded loss — Amsive, 2025
- 23× conversion — Ahrefs, 2025 · 4.4× — Semrush, 2025 · 11.4% vs 5.3% — Similarweb, 2026
- 6.8% citation rate, 9.5B visits, +70% YoY — Similarweb, May 2026
Chart construction
- No dual-axis charts: every series sharing an axis shares a unit.
- Solid dots are published readings. Hollow dots are interpolation. Dashed segments mean no figure was published for that period.
- Every chart has a table equivalent behind “Show data table”, and the direct label always anchors to the last measured point — never to an interpolated tail.
Compiled July 29, 2026 for RocketOpp LLC. Every figure is dated and attributed — none estimated by the author.
We build the structure this report says wins.
Schema, an llms.txt written for AI crawlers, a robots.txt that actually allowlists them, and content that leads with the answer. It is not an upsell — it is the reason the site is worth building. Right now that is $497.