ChatGPT's share of AI referrals: a new dataset widens the gap between studies

What BrightEdge published on 24 September

BrightEdge released new referral data on 24 September, and it pushes ChatGPT's lead further out than the studies we compared in August. BrightEdge sells an enterprise SEO and AI search platform, so it sells into the same category as Sigzen AI does.

The release says ChatGPT's share of AI referral traffic fell to a low in the spring, then climbed every month to a 2026 high in August, with Gemini, Perplexity and Claude splitting a small remainder.¹ It also reports that ChatGPT's referral volume grew strongly across the year. What it does not say is how many sites the data covers, which countries, or how a referral was classified.

That is why we quote no figure from it here. A vendor share without a sample or method is a direction, not a benchmark. The direction is still worth noting: in this panel, ChatGPT's lead grew through the summer rather than shrinking.

How it sits against the other datasets

Put BrightEdge next to two other 2026 studies and the spread in ChatGPT's share gets wider, not narrower. Each study counts a different thing over a different period, so the numbers were never measuring the same quantity.

SE Ranking, which also sells SEO and AI visibility tools, studied 101,574 websites in 250 countries with data from January–April 2026. ChatGPT sent 74.78% of AI referrals and Gemini 11.56%.² In the same data, AI search accounted for only 0.32% of all visits to those sites.³

Similarweb, another vendor in the category, measured visits to AI tools rather than clicks out of them. On its worldwide panel, ChatGPT's share of generative AI web traffic fell from about 76% in June 2025 to about 53% in May 2026.⁴ That is a usage share, and it moved in the opposite direction to BrightEdge's referral share.

Sources as footnoted. Shares are not comparable across rows: each study defines its sample, period and referral differently.

Why the studies keep drifting apart

The gap widens because the studies differ on four inputs, and new releases rarely line them up. Before quoting any share, check each one.

• Which sites. A panel of large brands, a broad crawl of small sites and a set of agency clients will each draw a different mix of engines.
• Which referrers. Some studies count only chat assistants. Others fold in AI Mode, Copilot inside a browser, or nothing from Google at all, because Google's AI features mostly arrive as organic search.
• Which months. A single month, a four-month average and a year-on-year comparison will not agree while shares move this fast.
• Which countries. Engine mix varies by market, and a release that does not state geography may be mostly US traffic.

The oldest of the four studies shows why the period matters. Conductor's benchmark, first published in November 2025 and updated in July 2026, put ChatGPT at 87.4% of AI referrals in traffic data from 1,215 US enterprise customer domains. Its referral data ran from May to September 2025, so it is now about a year old.⁵ Read it as a baseline from before Gemini's growth, not as a current figure.

Two shares that are easy to confuse

Share of AI usage

• Visits people make to ChatGPT, Gemini, Claude and others
• Tells you where questions are asked
• Similarweb's measure

Share of AI referrals

• Clicks those tools send to websites
• Tells you which engine sends visitors to sites
• BrightEdge's, SE Ranking's and Conductor's measure

Can ChatGPT lose users and gain referrals at once?

Yes, and the data suggests that is roughly what happened. A shrinking share of usage can still send a growing share of clicks if each answer links out more often.

Similarweb tracked exactly that change. The share of US ChatGPT prompts whose answer carried a citation rose from about 1.6% in June 2025 to about 6.8% in May 2026.⁴ More links per answer means more referrals per user, even as Gemini wins users. And Google's own AI features sit inside Search, where a click is usually logged as organic search rather than as an AI referral.

Our view: the studies are less contradictory than they look. Usage is spreading across engines, while the clicks that analytics tools can label as AI still come mostly from ChatGPT. Both statements can be true in the same month.

What to do with your own split

Use the panels for context and your own analytics for decisions. Your site's engine mix depends on your audience, your markets and which of your pages get cited, none of which a panel knows.

1. Set up an AI channel in analytics if you have not, following our guide to seeing AI referrals without fooling yourself.
2. Pull the last three full months by source and report each engine's share with the date range beside it.
3. Compare month to month, not against a vendor's headline. A change in your own split is a signal; a gap to a panel is not.
4. Read the referral split as a floor. App traffic, copied links and AI Overview clicks land in other channels.

Referral share also says nothing about where you are mentioned. An engine can describe your company in an answer and never send a click. Our August flagship on why referral studies disagree covers the method differences in more depth, and our AI visibility work measures mentions directly.

Our view: report your own engine split quarterly, with its date range, and cite panel figures only with their sample and period attached. Our AI answer monitor checks six engines (ChatGPT, Gemini, Perplexity, Claude, Copilot and Google AI) through official APIs where they are offered and a licensed AI-answer data provider otherwise, so mentions in Gemini or Google AI are counted even when they never show up as a referral.