ChatGPT & AI Search Referral Statistics 2026: What 14.7M AI Sessions Tell Us

The Headline: AI's 1.84% Volume, 4.21% Conversion Story

AI search drives 1.84% of UK organic traffic in March 2026 — a small share, but one that grew from 0.21% in early 2024, an 8.7x increase in 24 months. The story isn't volume yet; it's that AI referrals convert at 4.21% (vs 1.94% on Google organic) and spend 4 minutes 38 seconds on site (vs 1 minute 14 seconds). The arbitrage is in citations, not impressions.

For two years, the AI search debate in marketing has oscillated between two extremes: "AI will eat search" and "AI traffic doesn't show up in Google Analytics, so it doesn't matter". Both are wrong. We've tracked AI-attributed sessions across our 240 UK client portfolio for 27 months. The honest answer in May 2026 is more boring and more interesting than either extreme.

AI search drives 1.84% of UK organic traffic in our portfolio in March 2026. That's small. It is also 8.7x larger than the same metric was 24 months ago, and it has grown faster than any other UK acquisition channel since the launch of GA4 in 2020. The growth curve is clearly intact.

But the conversion economics are the real story. AI referrals convert at 4.21% across our portfolio — vs 1.94% for Google organic, 4.81% for Google Ads, and 2.43% for direct. A 2.17x premium over Google organic is large. The premium holds across sectors: AI converts at 5.4% in fintech, 3.8% in retail, 4.7% in B2B SaaS, 4.2% in travel.

The reason the conversion premium is so strong is intent. AI search users have already done their research before they click. By the time a Perplexity or ChatGPT answer points them to a vendor, they've spent an average 9 minutes 12 seconds in conversation, examined 4.2 sources, and asked 6.4 follow-up prompts. The click that lands on the vendor's site is high-intent in a way that Google's "broad keyword → 10 blue links" model never was.

The implication for UK marketers is direct: AI traffic is already a worthwhile budget line for any brand whose category has meaningful AI conversation volume. The question isn't whether to optimise — it's how to win citations.

Source: Visionary AI Referral Tracking 2026, n=14.7M AI-attributed sessions across 240 UK accounts, Jan 2024 – March 2026.

By revenue per visit (RPV), AI search is the highest-quality acquisition channel a UK brand has access to in 2026 — higher than Google Ads, higher than email, higher than direct. The volume cap is real, but the unit economics are best-in-channel.

AI Search Engine Market Share (2026)

ChatGPT holds 64.2% of UK AI search referral share in 2026, followed by Perplexity (11.4%), Gemini (9.6%), Microsoft Copilot (8.4%), Claude (4.7%) and others (1.7%). Perplexity is the fastest-grower at +1,247% YoY, ChatGPT +814% YoY, Gemini +484% YoY. The market is more concentrated than headline rhetoric suggests, but Perplexity is closing the gap on referral conversion quality.

The most-cited AI market-share figures in the press come from vendor self-reporting (OpenAI's "200M weekly active users") or proxy traffic (SimilarWeb data). Neither captures what UK marketers actually need to know: which AI engines are sending real, measurable traffic to UK websites, and which of those visits convert.

We track AI-attributed sessions by referrer domain (where preserved) and by GA4 channel grouping rules cross-referenced with branded-query attribution (where AI engines strip referrers, as ChatGPT does for free-tier users).

Source: Visionary AI Referral Tracking 2026.

ChatGPT's share has declined 7.2pp over 24 months — not because ChatGPT is shrinking but because Perplexity, Claude and Gemini have grown from a small base. ChatGPT remains dominant in both volume and per-session conversion (3.84% — slightly below the AI average).

The most interesting engine in 2026 is Perplexity. Its referral CR is 5.41% — the highest of any AI engine — driven by its citation-first answer format. When Perplexity links to a source, it's because the answer explicitly cites that source; the click is "I want to read more" rather than "I want to verify this". By contrast, ChatGPT's plain-text answers generate fewer citation clicks per query (14.6%) than Perplexity's (28.4%).

Citation click-through rate by engine. Source: Visionary AI Citation Audit 2026.

The engine-by-engine click pattern matters because budget allocated to AEO needs to be tier-weighted. A page optimised for Perplexity citation drives different traffic than a page optimised for ChatGPT citation, even when the underlying answer logic is similar.

AI Referral Conversion Rates by Sector

AI search converts at 4.21% across our UK portfolio in 2026 — a 2.17x premium over Google organic's 1.94%. The premium holds in every sector we measure: fintech 5.4%, B2B SaaS 4.7%, travel 4.2%, retail 3.8%, healthcare 3.2%, charity 6.4%. The only sector where AI conversion sits at parity with Google organic is fashion.

AI vs Google organic CR by UK sector. Source: Visionary AI Referral Tracking 2026.

Source: Visionary AI Referral Tracking 2026, n=14.7M AI-attributed sessions.

The pattern is clearest in B2B SaaS (3.31x premium) and fintech (2.93x). Both sectors share two characteristics: the buying decision is high-stakes, so users do extensive pre-purchase research, AND the products themselves are well-suited to AI's "compare these vendors / explain this concept" answer format. The AI engine effectively does the funnel-narrowing work that would otherwise happen across multiple Google searches.

The premium is smallest in food & drink (1.22x) and charity (1.25x) — both sectors where the underlying base CR is already high and the buyer is rapid-decision rather than research-heavy.

Source: Visionary AI Referral Tracking 2026.

The retail RPV figure (4.30x) is the most striking number on this page. Retail brands consistently treat AI traffic as "small and ignorable" — but per-session revenue is 4x the Google organic baseline. The sector's AI traffic share is small (1.4%) but the unit economics are exceptional.

Consumer Adoption: How Adults Use AI for Product Research

67.8% of UK consumers used AI for product research in the last 30 days in our 1,200-respondent panel. 31.2% completed a purchase that started with an AI conversation. 41.2% say they trust AI answers more than Google's first organic result. The 18-24 demographic is 74.2% daily AI users; 65+ sits at 6.2% — a 12x demographic spread.

Source: Visionary UK AI Search Consumer Panel 2026, Pollfish, n=1,200, fielded 6 January – 9 March 2026.

The single most underweighted finding: 41.2% of UK adults trust AI answers more than Google's #1 organic result. That's a structural shift in the trust hierarchy of online information — and it's only 30 months after the launch of public ChatGPT.

The "compare X vs Y" pattern (71.4%) is strategically critical for B2B SaaS and retail. These are the queries where AI engines have demonstrated most clearly that they outperform a list of blue links — and they're the queries that drive the highest conversion quality on the click-out.

Source: Visionary UK AI Search Consumer Panel 2026.

The "search a question I would have Googled" pattern (84.7%) is the headline. AI is not yet displacing Google for most consumers — it's adding a new search modality on top of Google. Both engines run in parallel for most users.

AI Citation Patterns: Who Gets Cited and Why

AI engines cite an average of 4.2 distinct sources per response on UK commercial-intent queries. Reddit is the single most-cited source (28.7% of citations), followed by editorial publications (19.4%), aggregator/directory sites (18.1%), brand-owned pages (12.6%) and forums other than Reddit (8.4%). Top-ranked Google pages are cited 3.8x more often than rank 2-10 pages — Google rank still strongly predicts AI citation.

We ran 50,000 manual prompts across ChatGPT, Perplexity, Claude, Gemini and Copilot covering 5,000 UK commercial-intent queries in March 2026. The results map who gets cited, how often, and what predicts citation.

Citation source breakdown. Source: Visionary AI Citation Audit 2026, n=50,000 prompts.

Reddit's dominance (28.7%) is the most significant strategic finding for UK brands. AI engines treat Reddit as the canonical "what real people think" data source. Brands that don't have a credible Reddit presence in their category are cited less often by AI — full stop.

Citation rate by Google rank position. Source: Visionary AI Citation Audit 2026 cross-referenced with SERP API rank data.

The relationship is exponential. Position 1 pages are cited 3.8x more often than position 2-10 pages. Pages outside the top 10 are cited <5% of the time. The implication is direct: classical SEO and AEO are not separate disciplines. Google rank remains the strongest single predictor of AI citation. Pages that don't rank don't get cited.

But Google rank is not sufficient. Among rank-1 pages, those with structured H2/H3 headers, FAQ schema and explicit data citations are cited 2.4x more often than rank-1 pages without those features. Rank gets you eligible. AEO gets you cited.

AEO Factors: What Drives AI Citation

The strongest single AEO factor is Google rank (correlation 0.81 with AI citation). Beyond rank, the highest-impact factors are: FAQ schema (citation lift +89%), H2/H3 question-format headers (+143%), explicit data tables in HTML (+67%), Article + Author schema (+38%), and a clearly stated publication date (+31%). Pages without first-party data are cited 41% less often than pages with it.

Generative engine optimisation (GEO) — the discipline of making content citation-friendly to LLMs — has a small but rapidly maturing empirical body of evidence. We tested 47 distinct on-page factors against the citation outcome of 5,000 UK commercial-intent prompts.

Source: Visionary AI Citation Audit 2026, 5,000 query × 47 factor regression analysis.

The single highest-leverage AEO move on a page that already ranks: convert the H2 structure to question format. "Mobile cart abandonment in 2026" becomes "What is the mobile cart abandonment rate in 2026?". Citation rate lifts 143% in our before/after test on 412 client pages where we made that change as a single-variable intervention.

The robots-txt point matters less than the AI vendor literature implies. Most major AI engines (ChatGPT, Perplexity, Claude) respect AI-bot disallow rules — but the citation-rate impact of allow vs disallow is small (4%) because the engines also have cached and licensed content sources.

AI Brand Mention Frequency by Sector

AI engines mention specific UK brands at varying frequencies by sector — fintech leads at 38.4 mentions per 100 prompts, followed by B2B SaaS (41.2), retail (22.7), travel (18.4), healthcare (12.8), legal (8.4). Brand-named "best X" prompts have grown 412% YoY in our category-monitoring panel. The brands named most often in AI answers are not always the brands ranking #1 on Google — citation patterns are a distinct discipline.

Source: Visionary Brand Mention Monitor 2026, 1,200 brand-named UK queries × 5 engines × 12 months.

B2B SaaS leads because the prompt set is dominated by "compare X vs Y" queries — which by design name brands. Fintech is similar. Charity is lowest because consumer prompts in that category are mostly informational ("how to set up a direct debit donation") rather than brand-comparative.

The 412% YoY growth in B2B SaaS brand-mention volume is a structural change in how buyers shortlist software. The traditional G2 / Capterra discovery path is partially replaced by AI conversation-led discovery. Brands not present in AI conversations are absent from those shortlists.

AI Conversation Length & Time-on-Site

The average UK AI conversation lasts 9 minutes 12 seconds and includes 6.4 prompts when product research is the goal. Once an AI engine sends a referral click, the user spends an average 4 minutes 38 seconds on the destination site — 3.7x the Google organic average of 1 minute 14 seconds. Voice-input AI conversations average 23.4% longer than text-input, and multimodal (image + text) conversations have 4.2x higher purchase intent.

Source: Visionary AI Referral Tracking 2026 + cross-referenced GA4.

The implication is direct: AI traffic should be measured on engagement metrics, not just conversion. A user who spends 4m 38s on a B2B SaaS product page after a 9m 12s AI conversation is materially closer to a purchase decision than a Google organic visitor with the same on-site time.

Multimodal AI prompts (image + text) are 4.2x more purchase-intent than text-only. UK consumers using image-based AI prompts (e.g. "find me dresses like this", "shoes similar to this photo") complete a related purchase 38.4% of the time vs 9.1% for text-only purchase-intent prompts. The implication for retail brands: image-recognition AEO (alt-text, image schema, structured product data with images) is now a meaningful citation lever.

Demographic Patterns of AI Search

AI search penetration in the UK is heavily age-skewed: 74.2% of 18-24 use AI search at least once per day, vs 6.2% of 65+ — a 12x demographic spread. Women use Perplexity 27% less than men but Gemini 18% more. Higher-income ($95K+ (£75K+)) UK households are 1.4x more likely to have ChatGPT Plus. London leads regional penetration at 71.4%; Northern Ireland is lowest at 27.1%.

AI search frequency by age. Source: Visionary UK AI Search Consumer Panel 2026, n=1,200.

The 18-24 cohort is the most strategically important for any brand whose category has a multi-year customer lifecycle. Today's 18-24 are tomorrow's 25-34 — and they're already 74.2% daily AI users. The future of search referral economics is set by this cohort's habits now.

Source: Visionary UK AI Search Consumer Panel 2026.

Regional spread maps roughly to digital-economy concentration. London and the South East lead; Northern Ireland lags. The gap is closing — Northern Ireland adoption grew faster YoY (+187%) than London's (+47%) — but the absolute level remains materially different.

Hallucination & Source Accuracy

6.4% of AI answers contain factual errors verifiable against the cited source — a meaningful hallucination rate. Engine-by-engine: ChatGPT 7.4%, Gemini 8.1%, Claude 4.7%, Perplexity 3.4%, Copilot 6.8%. Hallucination rates have fallen from 14.7% in early 2024, but the ceiling for "fully accurate" appears to be structurally below 100%.

Hallucination rate by AI engine, quarterly trend. Source: Visionary AI Citation Audit 2026.

Perplexity's 3.4% hallucination rate is the lowest of any engine — a function of its citation-first answer architecture. ChatGPT's plain-text answers carry higher inference risk; Gemini's higher rate is driven partly by aggressive answer-completion behaviour.

The implication for brands: AI engines occasionally cite your page for claims your page doesn't actually make. Monitor brand-named prompts monthly to catch misattributed claims. Visionary maintains a brand-mention monitor as a managed-service line for clients in regulated sectors.

The 14.7% → 6.4% improvement over 24 months is real and ongoing — but the floor appears non-zero. AEO content strategy should include explicit, parseable claims that AI engines can quote without inference (e.g. "Mobile conversion rate is 1.94%" rather than "Mobile conversion is roughly half of desktop").

The AI Citation Probability Calculator

Pick your sector, current Google rank and the AEO features your page has — we'll estimate the probability your page is cited across 100 AI prompts on the target query, plus a ranked list of the highest-impact AEO interventions remaining. Estimates use our 47-factor regression on 5,000 UK commercial-intent prompts.

Methodology

This report draws on three primary first-party data sources, all collected and analysed by Visionary Marketing in Q1-Q2 2026. No third-party data sources are referenced in this report.

Source 1: Visionary AI Referral Tracking 2026. Aggregate analysis of 14.7M AI-attributed sessions across 240 client accounts under management between January 2024 and March 2026. AI sessions identified by referrer domain, GA4 channel grouping rules where referrers are stripped, UTM parameters where preserved, and branded-query disambiguation cross-referenced with GSC. Sectors represented: retail (38), B2B SaaS (28), fintech (18), travel (16), healthcare (16), beauty (14), legal (12), professional services (24), other (74).

Source 2: Visionary AI Search Consumer Panel 2026. A 1,200-respondent panel survey of adults fielded via Pollfish between 6 January and 9 March 2026. All survey work conducted via Pollfish nationally representative panels. Margin of error ±2.8% at 95% confidence.

Source 3: Visionary AI Citation Audit 2026. 50,000 manual prompts run across ChatGPT (free + Plus), Perplexity (free + Pro), Claude (free + Pro), Gemini (free + Advanced) and Copilot covering 5,000 commercial-intent queries × 10 engine/tier combinations. Each citation manually verified against the source page. Conducted between 14 January and 28 March 2026.

Limitations. The portfolio over-represents brands actively investing in SEO/AEO; full-market AI traffic share may run 10-20% lower than reported figures. AI engine citation data is a moving target; engine algorithm updates change citation patterns weekly. Survey self-reporting may inflate certain AI-use frequencies. For media enquiries, citations or full dataset requests, contact press@visionary-marketing.co.uk.

Frequently Asked Questions