At a glance: ChatGPT is still the biggest single source of AI referral traffic, but its dominance is loosening. On the B2B panel behind Goodie’s 2026 AI Search Traffic Report, ChatGPT went from 89.1% of AI referrals to 62.6% in eight months, with Claude, Gemini and Perplexity absorbing the difference. Not every panel agrees on the magnitude — but the direction is consistent enough that optimising for one engine is now a concentration risk. The 2026 play is one strong content asset, reinforced per engine, measured per engine.
What the data actually says about ChatGPT’s referral share
Start with the headline number. In May 2026, Goodie published the second wave of its AI Search Traffic Report, comparing two measurement windows: 2,802,519 AI referral sessions across 41 brand sites (May–August 2025, via GA4) against a fresh anonymised GA4 brand panel for March–April 2026, triangulated against SimilarWeb traffic data covering 25.77 billion visits to the top five AI surfaces. The finding, in Goodie’s own words: “Eight months ago, ChatGPT held 89% of B2B AI referrals. Today, it holds 63%.”
Where did the share go? On Goodie’s panel, Claude jumped from 1.4% to 18.5% of AI referrals, Gemini roughly quadrupled from 2.4% to 10.6%, and Perplexity more than doubled from 3.1% to 7.3%. Four engines now account for nearly all measurable AI referral traffic — it’s just no longer one engine plus rounding errors.
Usage data points the same direction. Similarweb’s 2026 generative-AI statistics show ChatGPT’s share of generative-AI website visits falling from about 76% in June 2025 to roughly 53% by May 2026, while Gemini rose from under 9% to around 27–28% and Claude grew from barely 2% to close to 9%. And Gemini’s audience keeps compounding: on Alphabet’s Q2 2026 earnings call the company reported over 950 million monthly Gemini app users with daily actives tripling year on year, and on 11 August 2026 Sundar Pichai announced the app had passed 1 billion monthly active users.
The important caveat: measurement panels disagree
Here’s what most coverage of the “ChatGPT is falling” story skips. Previsible’s July 2026 AI traffic report — 6.77 million LLM-driven sessions across 166 GA4 properties, November 2024 through May 2026 — puts ChatGPT at 92.4% of trackable LLM referral traffic and still gaining relative share on that panel. Same year, same methodology family (GA4 referrers), wildly different answer to “how big is ChatGPT’s share?”
Both reports can be honest and still disagree: panel composition drives the result. Goodie’s panel is B2B-weighted; Previsible’s spans e-commerce, publishing, ticketing and more. Even Previsible’s ChatGPT-dominant panel shows the same second-order trend, though — Claude referrals grew 64x over the tracked period and Gemini grew steadily into the #2 position. The takeaway isn’t “ChatGPT holds exactly 63%” or “exactly 92%”. It’s that (a) the non-ChatGPT engines are growing fast on every panel, and (b) blended industry averages tell you very little about your traffic. Which is precisely the argument for measuring per engine on your own site — more on that below.
And yes — ChatGPT referral volume surged in June 2026. Both things are true.
If you saw headlines about ChatGPT referrals exploding mid-year, that’s real too. seoClarity measured ChatGPT-specific referral traffic growing 130% month-over-month in June 2026, driven by OpenAI’s 7 May 2026 interface update that embedded clickable inline brand citations directly in answers. That 130% figure is ChatGPT’s own referral traffic — not total AI traffic — and the other engines barely moved in the same window (seoClarity logged Gemini +8%, Claude +16%, Perplexity −10%).
Share and volume are different questions. The whole AI-referral pie is growing — Similarweb counts average monthly visits across generative-AI platforms up 70% year on year to 9.5 billion — so ChatGPT’s absolute referrals can surge while its slice of the pie shrinks over the longer window. For a marketer, both facts lead to the same conclusion: AI referral traffic is worth real optimisation effort, and betting the whole effort on one engine’s product decisions is how you end up exposed.
Single-engine AEO is concentration risk
Think about what the June 2026 surge actually demonstrates: one interface change by one company moved referral traffic 130% in a month. That cuts both ways. An engine can turn your traffic up — or design it away — with a UI decision you’ll read about after it ships. Similarweb also noted that after ChatGPT’s May 2026 update, roughly six in ten referred visits landed on homepages rather than deep pages, which changed overnight what “winning a citation” delivers.
If 100% of your AEO effort targets ChatGPT, your pipeline inherits every one of those product decisions. Meanwhile the engines growing fastest — Claude and Gemini on both major panels — are the ones almost nobody is deliberately optimising for yet. That’s the actual opportunity: less competition for citations on the engines gaining share.
Engines have different source diets (what our monitoring shows)
We run daily citation monitoring across ChatGPT, Gemini, Claude, Perplexity and DuckDuckGo for our own market — thousands of logged polls, written up in our first-party citation-poll data post. The consistent qualitative pattern: each engine has a different “source diet”, so the work that earns a citation on one engine often does nothing on another.
- ChatGPT leans heavily on third-party directories and listicles. When we trace what ChatGPT actually cites for commercial queries in our category, it’s review platforms and “best X” roundups far more often than vendors’ own sites.
- Gemini rewards on-site content and entity signals — well-structured pages on your own domain, a clean Google Business Profile, consistent entity information across the web.
- Claude and Perplexity favour clean, indexable HTML that answers the question directly, with visible freshness.
- DuckDuckGo’s AI answers follow classic SERP inclusion: if you rank in the underlying results, you’re in the pool; if not, you don’t exist.
These are our observations from our own monitoring — qualitative, one market, stated as such. But they match the practical experience of anyone who’s tried to reverse-engineer citations engine by engine: there is no single “AEO checklist” that covers all five.
The multi-engine playbook: one asset, engine-specific reinforcement
The good news is that multi-engine AEO is not five separate content programs. It’s one strong asset — a genuinely useful page with a direct answer up top, real data, clean HTML — plus a reinforcement layer per engine.
| Engine | Typical source diet (our observation) | Optimisation lever | How to measure |
|---|---|---|---|
| ChatGPT | Third-party directories, review platforms, listicles | Get listed and reviewed on the directories it already cites in your category | GA4 referrer chatgpt.com; run your buying prompts monthly and log citations |
| Gemini | On-site content, entity signals, Google Business Profile | Structured on-domain answers, complete GBP, consistent NAP/entity data | GA4 referrer gemini.google.com; poll Gemini with grounding on your prompts |
| Claude | Clean, crawlable HTML with direct answers | Fast, semantic pages; answer capsule near the top; allow its crawler | GA4 referrer claude.ai; prompt-level citation checks |
| Perplexity | Indexable pages with visible freshness | Update dates, current-year data, no JS-walled content | GA4 referrer perplexity.ai; citation polls on target queries |
| DuckDuckGo | Classic organic SERP results | Standard SEO: rank in the underlying index | Referrer duckduckgo.com; track SERP inclusion for target queries |
We’ve written the full engine-by-engine detail up separately, based on the same monitoring:
- How to get cited by ChatGPT — the directory-and-listicle route, step by step
- How to get cited by Gemini — entity signals, GBP and on-site structure
- How to get cited by Claude — the engine growing fastest on both 2026 panels
- How B2B brands get cited in Perplexity — freshness and indexability
Sequencing matters less than coverage. A sensible order for a services business: publish the core asset with a direct answer capsule and current-year data (helps every engine); same week, fix GBP and entity consistency (Gemini); then work the two or three directories ChatGPT already cites in your category; then keep the asset visibly fresh (Perplexity) and make sure it ranks for the underlying query (DuckDuckGo). One asset, five reinforcements.
Measure per engine, not blended
A blended “AI traffic” number hides exactly the information you need. If your AI referrals are 95% ChatGPT in a market where Gemini serves a billion monthly users, that’s not a win — it’s an exposure you haven’t noticed yet.
Two measurement layers, both cheap:
- Referral layer: in GA4, segment session referrers for chatgpt.com, gemini.google.com, perplexity.ai, claude.ai and duckduckgo.com separately. Watch each engine’s trend line, not the sum. (Caveat: AI referrals undercount reality — plenty of users read the answer, then search your brand name later. Referrer data is a floor, not a ceiling.)
- Citation layer: maintain a fixed list of the 10–20 prompts your buyers actually ask, run them against each engine on a schedule, and log whether you’re cited. This leading indicator moves weeks before referral traffic does. It’s what our own daily monitoring does, and it’s the layer that told us the source-diet differences above.
Why bother measuring a channel this small? Because the visitors convert. AI-referred visitors arrive pre-qualified by the answer that sent them — we’ve covered why AI-search leads convert higher than organic with the conversion data behind that claim. Small volume, outsized pipeline contribution.
Honesty box — what we couldn’t verify, and what’s easy to mislabel:
- The 89%→63% figure is one panel, not gospel. Goodie’s is B2B-weighted GA4 data; Previsible’s broader 166-property panel still shows ChatGPT at 92.4% of LLM referrals. We cite both because the honest claim is the direction (diversification), not a precise universal share.
- Perplexity query volume: the last primary-sourced figure is 780 million queries in May 2025, from Perplexity’s CEO. The “1.2–1.5 billion monthly queries” numbers circulating for 2026 are third-party extrapolations with no primary statement behind them, so we’ve left them out.
- The seoClarity +130% stat is ChatGPT-specific referral traffic (June 2026, month-over-month, after the inline-citations update). It gets widely misquoted as “total AI traffic grew 130%”. It didn’t — other engines were roughly flat in that window.
Where this leaves you for 2026
ChatGPT still sends the most AI referral traffic, and that’s worth optimising for. But every credible 2026 dataset — Goodie, Similarweb, even the ChatGPT-bullish Previsible panel — shows Claude and Gemini growing at multiples while most businesses optimise for one engine or none. The window where multi-engine AEO is cheap is exactly now, while your competitors are still arguing about whether AI traffic matters.
We do this for clients on a pay-per-result basis — 50,769+ AI-booked sales appointments since 2017 and 1M+ leads generated — and multi-engine citation coverage is now a standard part of how we fill calendars. If you’d rather see what your per-engine citation footprint looks like before committing to anything, that’s a 20-minute conversation. Book a call and we’ll walk through it with your actual buying prompts.
FAQ
Is ChatGPT still the biggest AI traffic source?
Yes, on every major 2026 panel. Goodie’s 2026 report puts ChatGPT at 62.6% of B2B AI referrals (down from 89.1% eight months earlier), while Previsible’s broader July 2026 panel measures 92.4%. Biggest, yes — but the panels agree the non-ChatGPT engines are the ones growing fastest, which is why we’d optimise for coverage, not just for the current leader.
Should I optimise for Gemini or ChatGPT?
Both — they respond to different work, so it’s not either/or. In our monitoring, ChatGPT citations mostly come from third-party directories and listicles, while Gemini rewards your own site’s structure, entity consistency and Google Business Profile. The core content asset is shared; the reinforcement differs. With Gemini’s app passing 1 billion monthly users in August 2026, ignoring it is the riskier call.
How do I track AI referral traffic by engine?
In GA4, segment referrers separately: chatgpt.com, gemini.google.com, perplexity.ai, claude.ai and duckduckgo.com. Report each engine’s trend line rather than a blended “AI traffic” total. Treat referrals as a floor — many AI-answer readers arrive later via branded search — and pair the referral data with scheduled citation checks on your key buying prompts.
Why did ChatGPT referral traffic surge in June 2026 if its share is falling?
Because share and volume moved on different clocks. seoClarity measured ChatGPT-specific referrals up 130% month-over-month in June 2026 after OpenAI’s 7 May inline-citations update — a product change, not a market shift. Over the longer eight-month window, Goodie’s panel shows ChatGPT’s share of all AI referrals falling as Claude, Gemini and Perplexity grew. The pie grew and ChatGPT’s slice grew slower.
Does DuckDuckGo matter for answer engine optimisation?
More than its market share suggests, because it’s nearly free to win. In our monitoring, DuckDuckGo’s AI answers draw from classic organic results — if your page ranks for the underlying query, you’re in the citation pool. That means ordinary SEO work you should be doing anyway doubles as DuckDuckGo AEO, with no separate program required.
How long does multi-engine AEO take to produce leads?
Citation wins typically show up before traffic does — a fresh, well-structured page can be cited within weeks of indexing, while directory-driven ChatGPT citations depend on the directories’ own crawl and review cycles. We track citations as the leading indicator and booked appointments as the lagging one. If you want a realistic timeline for your category, bring your prompt list to a call and we’ll map it against what we’re seeing engine by engine.