Your Traffic Is Down 33% and It Isn't Your Fault. Here's What I'd Do Anyway.

Three clients asked me the same question in the same week: "Did we do something wrong?"

Their rankings hadn't moved. Their positions in Google were where they'd been for a year. Impressions were flat or up. Clicks were down a third, in one case closer to half.

No, they didn't do anything wrong. This is the part I've had to say out loud more times this year than I'd like: a website traffic drop from AI Overviews is not a penalty, it's not a technical fault, and there's usually nothing in your Search Console to fix. Something structural changed under all of us at once.

That's the bad news. The rest of this post is the part that's actually useful, which is what still works.

The 33% Is Real, and It's Worse if You're Small

Chartbeat tracks traffic across thousands of publisher sites and has done for close to two decades. Their data, given exclusively to Axios in March 2026, shows organic Google search traffic down 33% globally over a twelve-month window, and down 38% in the US.

But the headline number hides the real story. Broken out by publisher size over two years:

• Small sites (1,000 to 10,000 daily pageviews): search referrals down 60%
• Mid-sized sites (10,000 to 100,000 daily pageviews): down 47%
• Large publishers: down 22%

Read that again. The smaller you are, the harder you got hit. Almost triple the damage at the bottom of the range compared to the top.

Chartbeat's read on why is uncomfortable but honest: bigger publishers have brand recognition and direct-to-consumer products. People know their name, bookmark them, subscribe to the newsletter, open the app. Search was never their only door.

If you're a small business or a small site, search was probably most of your door. That's the whole asymmetry in one sentence.

Your Traffic Did Not Go to ChatGPT

This is the single most expensive misunderstanding I'm seeing right now, and it's the reason I wanted to write this post.

In the same Chartbeat dataset, referrals from ChatGPT grew by more than 200%. Sounds enormous. Then you look at the base: AI chatbots still account for less than 1% of all publisher pageview referrals. Google Search pageviews fell 34% and Google Discover fell 15% over the same period.

So a rounding error tripled while the main artery lost a third.

The traffic did not migrate. It evaporated. Someone asked a question, got an answer on the results page, and never went anywhere. That's it. There's no secret door where all your visitors are now hanging out.

I bring this up because of what people do with the wrong theory. They spend three months and a real budget "optimising for ChatGPT," chasing a channel that, for most sites, is sending a few dozen sessions a month. Not zero, and worth having, but not a replacement for what you lost. I looked at my own numbers on this when Search Console started reporting AI traffic and the shape was the same: real, measurable, small.

Getting cited in AI answers still matters, and I'll come back to why. Just don't sell yourself the story that it's where the missing third went.

The Ad Money Confirms It

If you want a second read that isn't dependent on any one analytics vendor, look at what happened to ad inventory.

Ozone benchmarked roughly 20 billion impressions across the US and UK and shared the results with Digiday in July 2026. Between April and June 2026, publisher ad request volumes were down 32% to 37% year over year in the US and 39% to 41% in the UK.

Ad requests are a raw count of pages actually loaded. You can argue about analytics methodology. You can't really argue with 20 billion impressions worth of "the pages aren't being opened."

Spend didn't fall as fast as volume, interestingly. Rising eCPMs plugged most of the gap, at least for now. Fewer pageviews, each worth more. Whether that holds is anyone's guess.

The pressure got severe enough that outlets including USA Today, Politico, Reuters and The Economist have publicly reconsidered whether being in Google is worth it at all. That was unthinkable eighteen months ago. I'm not going to tell you to block anything, that's a separate decision with real consequences, but it tells you how bad it has got at the top end.

Which Queries Still Send a Click

Here's where it stops being depressing and starts being actionable.

Seer Interactive ran regression analysis on 5.47 million queries and 2.43 billion organic impressions across 53 brands, from January 2025 through February 2026. Their AI Overview prevalence numbers by intent:

• Informational queries: 36% trigger an AI Overview
• Commercial queries: 8%
• Transactional queries: 5%

A seven-fold gap between "what is X" and "buy X." And Semrush, studying 600,000 keywords across ten industries between November 2025 and April 2026, found the share of transactional results showing an AI Overview actually fell 5% over that window, while commercial-intent coverage grew 71%.

The pattern is clear enough to plan around.

Gone, and not coming back: definitional queries, simple factual lookups, "what is," "how many," "when did." Anything a paragraph answers completely. If your traffic was built on explainer content that resolves in three sentences, that traffic is structurally gone. I'm sorry. Rewriting it won't bring it back.

Still clicking: transactional queries, local intent, pricing, booking, "near me," anything ending in a decision a human has to make and a human has to be responsible for. Nobody hires a contractor from an AI summary. Nobody signs a five-figure contract off a snippet.

The awkward middle: comparison content. Seer found comparison-format queries trigger an AI Overview 95.4% of the time and question-format queries 85.9%. So comparisons are heavily covered by AI. But comparisons are also where the buyer sits closest to spending money, which means being the cited source in that overview is worth more there than anywhere else on the site. High risk, high value. Don't abandon it, but stop writing generic "X vs Y" pages a model can reconstruct from three other pages.

If your content plan still weights toward informational volume, that's the thing to change this quarter. Not your headings. Not your schema. The mix.

Count Conversions, Not Pageviews

Same Seer dataset, one more number that reframes everything: organic CTR on searches with no AI Overview actually rose, from 2.75% to 3.82% over thirteen months. On searches with an AI Overview it fell from 3.19% to 2.36%.

The clicks you do get increasingly come from people who went past an answer and decided they needed more. That's a different visitor than you had in 2023.

Which means the pageview number lies to you now, in both directions. It's down, and it makes you feel like you're failing. But the visitors behind it are more qualified than the ones you lost.

I've moved every client reporting dashboard I run onto three things: qualified leads, conversion rate, and revenue per session. Not sessions. One client's traffic is meaningfully down year over year and their booked jobs are flat. Nothing broke. The tyre-kickers who used to read a definition and leave now get their definition from Google, and I genuinely don't miss them.

If you're still judging your site by a sessions graph, you will fire a working agency and cancel a working strategy this year. Plenty of people already have.

The Base You Actually Own

Remember why the big publishers only lost 22%: brand recognition and direct relationships.

You can build a smaller version of that. It's slow and it's boring and it's the only part of this nobody can take away from you in a product update.

Branded search. People typing your company name is the one query class AI Overviews don't intercept in any meaningful way, because there's no generic answer to give. It's also the traffic that converts best. Every offline thing you do, the podcast, the local sponsorship, the referral, the conference, eventually shows up here. Worth checking what AI actually says when someone searches your name, though, because that summary is now part of your brand whether you manage it or not.

Direct and email. Publishers sent 28 billion emails in 2025 to over 255 million readers, with average open rates above 41%. There's a reason everyone rediscovered the newsletter at once. Chartbeat's Q2 2026 data shows loyal readers making up 58% of pageviews in Northern Europe, up from 55% the previous quarter. Loyalty is the metric that went up while everything else went down.

For a small business this doesn't mean launching a media brand. It means the 400 people who bought from you have a way to hear from you that isn't an algorithm's decision. A list of 400 engaged customers outperforms 4,000 sessions of definitional traffic. It always did. Now it isn't even close.

Be the Source, Even When There's No Click

Here's the last piece, and it's the one people find hardest to accept.

Seer's citation data: brands cited inside an AI Overview got a 2.07% organic CTR. Brands not cited got 0.94%. Being cited is roughly 120% more clicks than not being cited. It is also still 38% below what you'd get with no AI Overview at all.

So citation is not a fix. It's damage control worth doing anyway, because the alternative is worse and because the mention itself does work even without a click. Somebody reads "according to [your company]" and that's a brand impression you didn't pay for.

The practical version isn't complicated, and Google has been unusually direct about it: the guidance is mostly just good SEO. Original data, clear structure, plain answers near the top of the page, being quotable in a single sentence. I went through what a 1.4 million prompt study found about ChatGPT citations and the honest summary is that there's no trick. There's being genuinely useful, findable, and specific, on an internet that's increasingly full of the opposite.

What I Don't Have an Answer For

I'll be straight with you, because most of this industry won't be.

I don't know what a content business looks like in three years if informational search keeps collapsing at this rate. Publishers surveyed in this space expect another 43% decline within three years, and a fifth of them expect losses above 75%. I don't know whether those are correct or panic. Neither do they.

I don't know if AI referral traffic ever grows enough to matter. It might. Under 1% growing 200% a year is still small for a long time.

I don't know what to tell someone whose entire business was ranking for informational keywords. Some of those businesses don't have a version of themselves that survives this, and I'd rather say that than sell them a retainer.

What I'm reasonably confident about: transactional and local intent still sends clicks, brand and email still work, being cited beats not being cited, and pageviews stopped being the right number to look at. That isn't a strategy that gets your traffic back. It's a strategy for the search results that actually exist now.

If you're staring at a traffic graph and can't tell whether it's AI Overviews, a ranking problem, or something you broke, that's a real question with a findable answer, and it usually takes me an afternoon. Send me the details and I'll tell you which one it is. Including if the answer is that there's nothing to fix.