Reach. In March 2025 Google attracted 269.6 million U.S. visitors; 6.8 times the amount of ChatGPT’s 39.6 million.
Total referrals. Google pushed 3X more clicks out to other sites than ChatGPT.
Efficiency. The average ChatGPT user clicks 1.4 external links per visit; Google’s users click 0.6 times per visit.
Stickiness. A typical Google visit now triggers 10+ in-platform pageviews (or “queries”), up from 8.4 two years ago.
Destinations.64% of all ChatGPT referrals land on just 120 domains. These range from YouTube to NIH PubMed.
Why this resource exists
Over two hundred marketers joined the LinkedIn thread on my post. They asked hard questions about sample bias, summer traffic spikes, and impact. This article addresses those questions and adds fresh context so you can cite, link, or challenge the numbers with confidence.
Audience size
Google still dwarfs every rival. Its U.S. audience was nearly 270 million in March 2025, while ChatGPT climbed 47% year-over-year to 39.6 million in March 2025.
Bar chart: U.S. unique visitors, Mar 2025: Google 269.6 M vs ChatGPT 39.6 M.
<div class="post-note-cute">Reach sets the ceiling for potential traffic. You plan strategy based on potentia brand reach.</div>
Outgoing traffic volume
Google remains the firehose to the open web. Google sends three times the total clicks to the open web, but ChatGPT is not a rounding error.
Column chart. Outgoing visits, Mar 2025: Google 175.5 M, ChatGPT 57.7 M.
<div class="tbl-wrap"><table style="border-radius:12px;"><thead><tr><th>Engine</th><th>Outbound clicks, Mar 2025</th><th>Change vs. Mar 2024</th></tr></thead><tbody><tr><td>Google</td><td>175.5 million</td><td>+66%</td></tr><tr><td>ChatGPT</td><td>57.7 million</td><td>+558%</td></tr></tbody></table></div>
Clicks per user
When comparing search engines of very different sizes, ratio beats volume. ChatGPT sends users to external websites more than twice as often.
Line chart. visits-out per individual, Jan 2024–Mar 2025: ChatGPT steady near 1.4, Google near 0.6.
Why did traffic spike in May2024/July 2024?
In July, Google referrals jump, then fall back. Three overlapping things may lead to an answer to why, but I’m not confident enough in the evidence to say any these ARE the reasons.
<div class="tbl-wrap"><table style="border-radius:12px;"><thead><tr><th>Date</th><th>What happened</th><th>Why it drives clicks</th></tr></thead><tbody><tr><td>5 Mar – 19 Apr</td><td>The largest Core + Spam updates in Google history</td><td>Cleaner SERPs exposed more trustworthy pages, rewarding clicks.</td></tr><tr><td>14 May</td><td>AI Overviews rolled out at Google I/O</td><td>Links placed inside an Overview get higher click-through than the same blue link.</td></tr><tr><td>14 Jun – 14 Jul</td><td>Euro 2024 and Copa América ran side-by-side</td><td>Millions of real-time sports queries created an outsized need for fresh sources.</td></tr><tr><td>Jun 2024</td><td>Google cut AI Overviews from 84% of searches to ≈ 15 %</td><td>Fewer AI answers pushed more searchers into organic results.</td></tr><tr><td>16–17 Jul</td><td>Amazon Prime Day 2024</td><td>Shopping frenzy sent high-intent buyers to comparison pages.</td></tr><tr><td>26 Jul – 11 Aug</td><td>Paris 2024 Olympics</td><td>Global event coverage spiked demand for schedules, scores, and highlights.</td></tr></tbody></table></div>
Annotated line chart. Referral spike May–Jul 2024.
<div class="post-note-cute">When core updates, interface shifts, and cultural events collide, even a stable source like Google can swing a large percent in referral output.</div>
Google is growing stickiness on purpose... probably
Pages-per-visit on Google desktop has climbed to double digits. Every extra in-SERP click is one less opportunity for your site to earn traffic. Google claims this as “query growth”. It's good for Google's stock.
Bar chart. Pages per visit: Google up to 10.1 in April 2025.
Where does ChatGPT send referrals though links?
Technology, health, and news sites absorb the bulk, but 64 percent of all external clicks go to just 120 domains, including YouTube, Wikipedia, and Amazon.
Bar chat. Volume of ChatGPT referrals by industry, Mar 2025.
Top 1,500+ websites that ChatGPT sends users to
What the dataset cannot see
Browser only. Similarweb’s panel measures desktop and mobile browsers. App activity, including most ChatGPT iOS/Android usage, stays hidden.
±10 % typical variance, ±30 % in edge cases. We validated against 100+ first-party 1st party analytics sets. Results fell inside those bands.
OpenAI/Google self-referrals removed. Links to any openai.com subdomain (≈27.7 % of all ChatGPT clicks in March) are stripped before analysis.
<div class="post-note-cute">Watch branded-search lifts. If ChatGPT cites you without a link, the next step is might be a branded Google query.</div>
Community Q&A
Can we trust Similarweb?
Our tests show ≈ ±10 % average error against first-party logs, sometimes wider. Treat trends as directional, not exact counts.
“ChatGPT is less than 1 % of Google in my GA4. Am I doing something wrong?” (Joy Hawkins)
UTM-less links from the ChatGPT app arrive without referrer data. They blend into “Direct”. Your browser numbers will look bigger than your app numbers. I also think that the browser-based app is inconsistent in sending referral data but do not have solid proof of this yet.
“Do users click away because they distrust AI answers?” (Steven Johnson)
Possibly. Google hides links behind AI Overviews; ChatGPT shows them inline. When an LLM quotes a source, skeptical users click to verify. This faith-check behavior amplifies ChatGPT’s per-person referral rate.
“How would you use this insight for client wins?” (Saeed R.)
Focus on proven revenue drivers first (probably classic SEO). Then, enrich cornerstone pages with entity-dense, value-dense copy, schema, and unique data. Finally, publish supplemental helpful resources that LLMs might cite.
Takeaways
Google still rules reach
ChatGPT rewards depth
Zero-click risk is rising
Data context matters
Open research questions
What share of ChatGPT users rely on the app vs. the browser?
Will citation UX in AI Overviews change click-out rates?
In the spirit of building a more transparent community (Michael King's goal of the Relevance Engineering concept presented at #seoweek), I'm sharing a light version of a report I spent many weeks manually putting together.
The full downloadable report is at the beginning of the post. :)
My intention is to find the answer to what us SEOs/AI Opportunists seem to be arguing about on here. (LLMs as search engines and user adoption rate).
First. My goal is to help take emotion out what's going on at a high level in search to level set us. I'm a community-focused collaborator NOT an opportunist.
Second. I welcome any constructive conversation. Have different data? Let's dialogue. Disagree? Please tell me what I'm not seeing. Have different methods/interpretations? Yes, let's figure this out together. If you wanna tell me to get bent, go for it, but that's mean.
Third. Some of what I've been seeing here on LinkedIn lately have been cherry-picked metrics to defend or dispute a hot take or to promote a tool/service/data provider. But also, please don't bucket me in with influencers/fear-mongers. I am not your opposition.
What The Data Says
Here’s what 13 months of U.S. clickstream data (Mar 2024–Mar 2025) shows for Google/OpenAI. Again, this is data-first, and did my best to exhibit transparency in all 26 pages.
Google's audience is almost 7X the size of ChatGPT. BUT ChatGPT is over 2X more likely to send traffic to external websites than Google.
Important note: to those saying that those websites are all to a handful of websites (and I've seen a few of these posts on LinkedIn over recent days)... it's NOT much different than Google. Check page 21 for where 64% of ChatGPT off-platform clicks go to.
The read is Google delivers the MOST TRAFFIC, but keeps MORE CLICKS within platform. ChatGPT, though notably smaller, sends out web visitors MORE OFTEN.
<div class="post-note-cute">Note of transparency: I am using Similarweb for this data. I am NOT being compensated (actually, I'm a satisfied customer). I moved on from other popular providers early this year after six months of testing different clickstream products against 1st party data I had access to. It was the closest (+/-10% on average, though sometimes up to +/-30%).</div>
Also, a special shoutout for Baruch Toledano for always taking time to have meaningful conversations when we cross paths.
I'll continue to share what I see if there's value in it for the SEO community. Want to contribute? HMU.