Semrush Says Amazon Lost 15% of Its Traffic. Similarweb Says It Grew.

Semrush traffic data published by Search Engine Land on September 4 shows ChatGPT.com up 48.38% year over year to 1.09 billion monthly US visits in July 2026, with Bing down 50.43% and Amazon down 15.31%. Similarweb's May 2026 panel shows Amazon growing 4.8% and Bing down 17%. The direction of the story holds across both panels. The magnitudes do not, and that gap is the useful part.

The list everyone is going to screenshot

Greg Jarboe pulled Semrush Traffic Analytics for the top 150 US sites and compared July 2026 against July 2025. The winners, in rough order of how loud the headlines will be:

• ChatGPT.com: +48.38%, now ranked 9th at 1.09 billion monthly visits
• YouTube: +36.6%, 10.3 billion visits, holding 2nd
• Twitch: +25.6%, up five spots to 25th
• LinkedIn: +15.48%, 473.7 million visits, up from 21st to 19th
• USPS.com: +12.04%
• Google: +10.72%, 25.3 billion visits, still 1st by a mile

And the losers:

• Bing: −50.43%
• Temu: −28.55%
• NBCNews.com: −20.2%, down 35 spots to 145th
• ESPN: −16.0%, 366 million visits, down from 19th to 24th
• DuckDuckGo: −15.83%
• Amazon: −15.31%
• Yahoo: −13.82%

The framing you'll see on LinkedIn tomorrow is that AI is reshuffling the web. The number in the same article that argues against that: the top 10 sites took 68.6% of all traffic across the top 150, versus 68.8% a year earlier. The pie got carved differently inside the top 10. The pie itself is exactly as concentrated as it was. Google and YouTube, both owned by the same company, are the two biggest gainers in absolute visits, which is a strange outcome for an "AI is eating Google" year.

Two panels, two different internets

This is the part I'd want a paid social manager to sit with before forwarding the list to a CMO. Semrush and Similarweb are both clickstream panels. Neither one is looking at server logs. They take a sample of browsing behavior, model it up to the whole population, and publish estimates. That is fine for direction. It gets shaky on magnitude, and the two panels show it.

Similarweb's May 2026 top-100 list has YouTube at 5.7 billion US visits, up 2.61% year over year. Semrush has YouTube at 10.3 billion, up 36.6%. Similarweb has Amazon up 4.80%. Semrush has Amazon down 15.31%. Similarweb has Bing down 17.39%. Semrush has Bing down 50.43%. Two months separate the snapshots, so some drift is expected. Not that much. Bing did not lose a third of its remaining traffic between May and July, or if it did, that's a bigger story than the one being reported.

Temu is the cleanest example of why this matters. Semrush says −28.55% year over year. But Sherwood News, citing Similarweb in late May, reported Temu's US visits had broadly returned to pre-tariff levels, around 353 million in April. Meanwhile Digiday's reporting on Temu's ad spend has the company cutting X spend by 95%, YouTube and TikTok by 74% each, over the first five months of 2026. So a 28% traffic drop is entirely plausible as a paid-media hangover. It's also plausible the panel is undercounting app-driven visits from a user base that Sensor Tower says grew. I genuinely don't know which, and neither does anyone quoting a single panel with two decimal places of confidence.

If I were you, I'd adopt a simple rule for competitor traffic claims: trust the direction only when two panels agree, and treat the magnitude as plus or minus 20 points. That sounds sloppy. It's more honest than the alternative, which is building a Q4 forecast on a decimal that another vendor says has the wrong sign.

What the winners have in common (it isn't AI)

Look at the winners again without the AI lens. YouTube, Twitch, LinkedIn, USPS, ChatGPT. Every one of them is a place people go on purpose. You open the app, you type the URL, you have a reason to be there that predates any search. Nobody discovers Twitch through a featured snippet.

Now the losers. Bing, DuckDuckGo, Yahoo are intermediaries, layers between a person and the answer they wanted. NBC News and ESPN are publishers whose traffic model was built on search and social referrals. Those are the two groups getting squeezed: the sites that stood between people and information, and the sites that depended on those intermediaries for their audience.

The sites growing are the ones people would still visit if Google disappeared tomorrow.

That is the whole pattern, and the AI Overview data explains the mechanism. Pew Research tracked 900 US adults across 68,879 Google searches and found people clicked a traditional result 8% of the time when an AI summary appeared, versus 15% without one. Ahrefs measured a 34.5% lower click-through rate on position-one results when an AI Overview was present, using 300,000 keywords, and its early-2026 re-run put the gap closer to 58%. Chartbeat's two-year study, covered by PPC Land, has search referrals down 60% for small publishers, 47% for medium, 22% for large. The same day the Semrush list ran, Search Engine Land also reported that USA Today is overhauling its audience strategy and named search pressure as the reason. Big newsrooms don't rebudget publicly unless the internal numbers are worse than the external ones.

Amazon is the one that doesn't fit, and I'll admit it bugs me. A destination site, down 15%, in the panel that also has ChatGPT up 48%. One reading is that product research is moving into chat interfaces and Google's AI Mode before the buyer ever lands on Amazon. Another reading is that it's panel noise, given Similarweb has Amazon growing. I lean toward noise, but only a little.

And honestly, USPS is the one I keep coming back to. Nobody is optimizing the post office for AI Overviews. People go there because they need to go there. That's the entire strategy, and it's up 12%.

The 10-minute audit: how much of your traffic is borrowed?

Search traffic is a rented apartment where the landlord keeps moving in new roommates and charging you the same rent. AI Overviews are the roommates. Owned traffic is the part of the building you actually hold the deed on. The audit is figuring out which floor you live on.

Open your analytics, set the last 12 months against the prior 12, and look at the default channel grouping. Add up direct, email, SMS, push, and app sessions. Divide by total sessions. That number is your owned share, and from what I've seen it sorts sites into three rough buckets:

• Under 25% owned: you are an intermediary in this dataset's terms. Your audience belongs to Google and Meta, and the trend line above is your trend line.
• 25% to 40%: mixed. You have a real audience, but a bad core update still shows up in revenue.
• Over 40%: you're a destination. The Semrush list is interesting to you but not threatening.

Second check, and this one takes about four minutes. Open Search Console, filter the last 16 months, and split branded from non-branded queries. If non-branded impressions rose while non-branded clicks fell, you are feeding AI Overviews for free. We walked through that exact Search Console filter when Google rounded everyone's clicks to "billions" in July, and the query hasn't changed.

Third, for the paid social people specifically: check what share of your conversions have a direct or branded-search touch anywhere in the path. If it's climbing, your paid campaigns are building a destination, which is the good outcome. If it's flat while spend rises, you're renting.

The ChatGPT number matters here too, but not for the reason the headlines say. 1.09 billion monthly visits in one country is why OpenAI is building an ad stack, and it's why the early-CPM window on those ads is worth a test. It seems less likely to me that ChatGPT sends you meaningful referral traffic anytime soon. Chartbeat had ChatGPT referrals up over 200% and still under 1% of publisher page views. Growth from nothing is still nearly nothing.

Where you are drifting on the map

My prediction, and I'll put a number on it: by the time the July 2027 comparison runs, at least two AI assistants sit in the US top 15, and at least four of the news sites currently in Semrush's top 150 are off the list entirely. Bing stays in the top 20 only because Windows keeps shipping with it.

The instinct is to read a winners-and-losers list as a scoreboard and figure out which team you're on. I think it's more useful as a map. The sites losing are the ones that lived in the space between a person and what they wanted. The sites gaining are the ones people wanted in the first place. Where you sit on that map is the thing to figure out this week, and the panel data, wobbly as it is, will at least tell you which direction you're drifting.

I don't think the sites that survive this are the ones with the best AI Overview strategy. They're probably just the ones somebody would type into a browser without being told to.

Notice Me Senpai Editorial