ChatGPT links a source in 6.8% of answers: travel near 23%, professional services under 4%

What Similarweb published

Similarweb's AI search statistics page, published on 29 July 2026, tracks how often ChatGPT answers include a citation. In its US data, about 1.6% of ChatGPT prompts carried a citation in June 2025, and about 6.8% did in May 2026.¹ Similarweb describes a sharp rise from October 2025, a pause of about four months and then a renewed climb.

The sector split is the more useful part. In May 2026, travel and hospitality answers carried a citation about 23% of the time, automotive about 20%, and professional services under 4%.¹ Similarweb's explanation is that categories built around specific, checkable options draw more citations than categories built around general guidance.

The same page reports two other figures worth knowing. ChatGPT's share of generative AI website visits worldwide fell from about 76% in June 2025 to about 53% in May 2026, and, citing its 2026 Generative AI Landscape report (data period not stated), that roughly 26% of ChatGPT responses contained an ad.¹

What the citation rate counts

It counts the share of US ChatGPT prompts in Similarweb's data whose answer carried at least one source link. It is not a count of how many sources an answer cites, and it is not a measure of any one brand. A rate of 6.8% means that in most observed prompts, the user read an answer with no link to follow.¹

Two limits follow from the method. The data is US only, so other countries may look different, and the page does not break the figure down by device. And a panel sees what users asked, which skews toward everyday questions; a B2B buyer's research prompts are a small part of it.

Compare Semrush's 2026 AI Visibility Index, built from 126 million US prompts tracked from January to April 2026. It found ChatGPT cited about 15 sources per response, against about 3 for Gemini.² That looks like a contradiction and is not one. Semrush counts sources inside answers to prompts it chose and ran, while Similarweb counts how often real users' prompts got any source at all. Quote each figure with what it counts, as our note on reading benchmark dates argues for referral data.

ChatGPT answers carrying a citation, US, May 2026

Travel and hospitality¹~23%

Automotive¹~20%

All answers¹~6.8%

Professional services¹under 4%

Why do sectors differ so much?

Because the answer a traveller needs points somewhere specific, and the answer a client of an accountant needs usually does not. "Which hotels near the old town have parking" has checkable answers with booking pages behind them. "Should I incorporate my consultancy" gets general guidance, and the engine rarely sees a reason to link.

Automotive sits high for the same reason: models, trims, prices and dealer stock are specific and change often. Professional services sit low because most prompts ask for explanation or judgement. Our view: the gap is less about how engines treat sectors and more about the shape of the questions people ask in them.

Which KPI should you track in your sector?

Pick the KPI that matches how often answers in your sector link out. In a low-citation sector, most buyers read your name, or a competitor's, in an answer with no link at all. Being named, and described accurately, is the outcome you can measure.

Our recommended KPI split. Check your own sector's rate before you choose a row.

Referral visits are a weak lead KPI everywhere for now. SE Ranking's study of 101,574 websites from January to April 2026 found that AI search accounted for 0.32% of visits.³ ChatGPT sent 74.78% of those AI referrals and Gemini 11.56%.⁴ When the channel is that small, a monthly swing in visits says little about whether engines recommend you.

Our view: whichever row you are in, measure across engines, not ChatGPT alone. Similarweb's own figures show ChatGPT's share of generative AI visits shrinking, and the engines differ in how much they cite. Our buyer's spec for an AI visibility score lists what a credible measure should disclose, including prompts, engines and run counts.

How we would use this figure

Use the sector rate to set expectations, not targets. A professional services firm should not promise its board AI referral growth when fewer than one answer in twenty-five links anywhere. A travel brand should expect citations to matter and should check which of its pages are being linked.

Our AI answer monitor records, for each prompt and engine, whether a brand is named, whether it is cited and which page is linked. In a Sigzen AI audit, every prompt runs on each of six engines (ChatGPT, Gemini, Perplexity, Claude, Copilot and Google AI) on three separate days. That separates being named from being linked, which is exactly the distinction Similarweb's sector split makes visible.