18 Months of AI Search Data: How Traffic, Citations, and Conversions Actually Moved for SMBs

Eighteen months is roughly how long AI search has been a measurable business channel rather than a demo — and the data from that period now exists in enough independent studies to assemble the honest picture. So that's what this post is: a synthesis of published research from Similarweb, Adobe Digital Insights, Conductor, Seer Interactive, Ahrefs, Contentsquare, SparkToro, Microsoft Clarity, and cohort benchmarks of hundreds of SMB sites, arranged as one timeline from January 2025 to mid-2026. Not our client data, not one vendor's dashboard — the cross-study consensus, with the contradictions kept in. If you run an SMB and want to know what actually happened to traffic, citations, and conversions while everyone argued about it on LinkedIn, this is the record.

The 18-month story in one paragraph: organic clicks got scarcer, AI referrals grew explosively on a small base, and the quality of AI-referred visitors turned out to be the real headline. Zero-click search climbed from ~58–60% of queries to roughly two-thirds. AI referral traffic grew 3–7× year-over-year depending on the study — 165× faster growth than organic search — yet still sits around 1% of total visits for a typical site. And the visits that do arrive convert at multiples of organic (studies range from ~1.3× to well over 10× depending on industry and methodology), because the AI pre-qualifies the visitor before the click. The strategic translation: the channel is small, fast-growing, high-intent, and badly mismeasured — which is precisely the combination early movers get paid for.

The timeline: six quarters, quarter by quarter

Q1 2025 — the baseline quarter. AI referrals exist but round to zero for most SMBs. ChatGPT's citation behavior is embryonic — in January 2025, only 0.6% of ChatGPT answers included citations at all. Zero-click search sits around 58–60% of US queries. This is the "measured before" against which everything else moves.

Q2 2025 — the acceleration. Similarweb's clickstream shows total AI referral visits growing more than 3× between September 2024 and September 2025, with the steepest acceleration concentrated in January–May 2025 before leveling. Meanwhile the first rigorous CTR damage reports land: Seer Interactive's longitudinal tracking of informational queries measures organic CTR collapsing (their 15-month study ultimately showed drops from ~1.76% to ~0.61%), and Ahrefs quantifies position-1 CTR falling roughly 58% when an AI Overview is present.

Q3 2025 — volume becomes undeniable, concentration peaks. AI platforms generate 1.13 billion referral visits in June 2025 — up 357% year over year. In panel data across 41 brand sites (May–August 2025), ChatGPT accounts for 89% of all measurable AI referrals: a one-engine channel. Citations mature too — ChatGPT's citation rate has grown from 0.6% to 2.8% of answers by August — small absolutely, but nearly 5× in eight months.

Q4 2025 — the commercial proof quarter. The 2025 holiday season delivers the numbers that move budgets: Adobe measures AI-driven referral traffic to US retail sites up ~693% year over year, and — the key finding — AI referrals converting 31% better than non-AI traffic. Contentsquare's cut is soberer on volume (AI referrals at just 0.2% of total traffic in Q4) but confirms the same quality signal: fast growth, improving conversion. The pattern that defines the whole dataset is now visible: tiny share, superior quality.

Q1 2026 — diversification and the measurement reckoning. The engine mix breaks open: in the same panel methodology that found ChatGPT at 89%, its share of B2B AI referrals falls to ~63% by March–April 2026, with Claude jumping to ~18%, Gemini ~11%, and Perplexity ~7% — a one-engine channel becoming a four-engine channel in eight months. Simultaneously, the attribution problem gets quantified: cohort tracking of 200 Stripe-connected SMB sites finds a median 34% of GA4 "Direct" traffic is actually AI-referred once server-side detection is applied (41% for B2B SaaS) — meaning most SMBs have been dramatically undercounting this channel the entire period. TechCrunch reports AI traffic to US retailers up another 393% in Q1.

Q2 2026 — the channel professionalizes. Blended AI referral share reaches ~1.08% of all website traffic (Conductor), growing roughly a percentage point of share per month, with heavy vertical skew — IT at ~2.8%, and legal, finance, health, SMB services, and insurance together accounting for over half of LLM-sourced sessions. Conversion benchmarks stabilize into ranges (more below). Similarweb's April–May clickstream puts ChatGPT referral conversion at 7.1% — second only to paid search at 7.8%, ahead of organic, direct, email, and social. And the SMB cohort data shows AI-attributed sessions compounding at 13.4% monthly, with Perplexity and Claude growing faster than ChatGPT.

The three findings that survive every methodology

Cross-study synthesis means embracing that numbers disagree; these three conclusions hold everywhere:

1. Conversion superiority is real, but the multiple depends on who's counting. Adobe says +31% for retail. The Stripe cohort says B2B SaaS converts AI traffic at 2.7% vs 1.4% organic (~1.9×) — but finds the pattern reverses for e-commerce (organic 2.1% vs AI 1.6%). Similarweb's panel says 7.1% for ChatGPT referrals. Publisher-side studies (Microsoft Clarity, 1,277 domains) report Copilot referrals converting at up to 17× direct traffic, and enterprise analyses claim 30–40% in specific contexts. The honest read: methodologies, industries, and conversion definitions differ wildly — but no credible study finds AI referrals converting worse than organic for B2B and services, and the mechanism is structural: the AI pre-qualifies the visitor — they arrive having already compared, which organic search never guaranteed. (E-commerce is the exception worth respecting: transactional AI traffic doesn't yet out-convert Google Shopping-era organic.)

2. The channel is systematically undercounted — probably in your analytics too. A third of "Direct" traffic being disguised AI referrals is the single most actionable finding of the 18 months. If your GA4 shows negligible AI traffic, that's likely a measurement artifact, not a fact about your business. The fix costs an afternoon: referrer segmentation for the AI domains, and treating unexplained Direct growth as a signal to investigate rather than dismiss.

3. Per-engine value and per-engine volume are inverted. In the SMB cohort's revenue-per-visit ranking, Perplexity ($1.42) and Claude ($1.18) beat ChatGPT ($0.87), Gemini ($0.41), and AI Overviews ($0.29) — nearly the mirror image of session share, where ChatGPT carries ~71% of volume. Claude's RPV on B2B SaaS specifically hit $1.94 on just 6% of sessions. Strategy implication: the highest-volume engine is not the highest-value one, and the fastest-growing engines (Perplexity +21.6%/mo, Claude +18.3%/mo in that cohort) are the high-RPV ones — exactly the platform-specific optimization argument from our GEO comparison post, now with revenue attached.

What did NOT happen (the doomsday audit)

Eighteen months also falsified some 2024-era predictions, and honesty requires the list: organic traffic did not go to zero — it declined for informational queries while branded and transactional held far better; SEO's foundation role, if anything, strengthened (citation studies throughout the period kept finding organic visibility upstream of AI citations). AI traffic did not replace search traffic — at ~1% share it supplements; the disruption is in the invisible layer (zero-click answers shaping decisions that later arrive as "Direct" or branded search). And citations did not stay a ChatGPT monopoly — the 89%→63% share shift in eight months is a warning against optimizing for any single engine's current behavior. Meanwhile one prediction aged well: Gartner's early call that traditional search volume would drop ~25% by 2026 landed close to the observed zero-click and volume trends.

The SMB playbook, extracted from the arc

If the 18 months teach one operating posture, it's this: instrument now, optimize for quality over volume, and diversify across engines. Concretely — segment AI referrers in GA4 today and reclassify your Direct anomaly (the afternoon fix with the highest information yield); judge the channel on conversion and revenue-per-visit, not sessions (1% of traffic at 2–7× conversion is worth more than its share suggests); do the visibility work from the AEO/GEO playbooks across all four engines, weighting the high-RPV ones your dashboard will initially tell you to ignore; and check you're not one of the sites that opted out by accident — roughly a third of B2B SaaS companies currently block AI crawlers in robots.txt, removing themselves from the entire channel this post just quantified.

The next 18 months start from a very different baseline than the last: four viable engines instead of one, conversion proof instead of promises, and monthly compounding that turns 1% into a first-page line item on someone's revenue report — most likely the competitor who instrumented earliest. The data phase of AI search is over; the operating phase has started.

We run this instrumentation and visibility work as a standard engagement — referrer segmentation, the four-engine audit, and the monthly citation tracking that turns this post's industry averages into your numbers.