A client called me in a mild panic last month. Organic traffic down roughly a third year over year. No site outage. No penalty. No redesign. Nothing in the logs.
His developer told him it was probably a Google algorithm update and it would recover. It won’t. Nothing broke. The exchange his website was built on quietly ended, and nobody sent him a notice.
The number that made this real
On August 11, nearly 300 French daily newspapers filed a complaint with France’s competition authority over Google’s AI Overviews, which rolled out in France on July 22. Buried in that filing is the number that should get your attention: Arcom, the French regulator, estimates the traffic loss attributable to AI-generated summaries at 33% to 38% in European markets where the feature is live.
That’s a regulator’s estimate, not a vendor’s press release. It lines up with independent measurement. Pew Research found that only about 1% of users clicked a link inside an AI Overview. SparkToro and Datos put it plainly: out of every 1,000 Google searches in the US, roughly 360 end in a click to the open web.
The publishers are fighting about copyright and compensation. That’s their business model at stake. But the mechanism they’re describing applies to your business too, and you don’t have a trade association filing on your behalf.
The deal you didn’t know you’d signed
For twenty years, the arrangement was simple. You published useful content. Google indexed it, showed a snippet, and sent a person to your site. You paid for that person with the effort of publishing.
That was a referral engine. What Google is becoming is an answer engine. It reads your content, synthesizes it into a response, and the user’s question gets resolved on the results page. The transaction completes without you.
Your website architecture assumes the old deal. Every landing page, every gated PDF, every “read more” was designed to capture a visitor who arrives with a question. Increasingly, they arrive with the answer.
Before you panic, look at the second number
Here’s where most of the coverage stops and most of the advice goes wrong.
Traffic volume is down. Traffic quality is up, sharply. Semrush measured AI-referred visitors converting at roughly 4.4 times the rate of traditional organic. Adobe’s retail data flipped from AI traffic converting worse than other channels to converting materially better inside about a year. Conductor pegs blended AI referral share at only about 1% of total site traffic. Tiny, but that 1% behaves like qualified demand rather than browsing.
The logic isn’t mysterious. Someone who reaches you through an AI answer has already had the basic question resolved. They compared options inside the chat window. They’re following a citation to verify you or to act. They arrive late in the journey, not at the top of it.
So the honest read is not “traffic is collapsing.” It’s this: your top-of-funnel volume is being absorbed, and what’s left is a smaller, harder-to-earn, much more valuable stream. If your dashboard tracks sessions as the headline number, you’re watching the wrong line move and drawing the wrong conclusion.
This is not a marketing problem
I’ve watched a dozen businesses respond to this by calling their SEO agency. That’s the wrong first call, and I say that with respect for good SEO people.
The problem is structural. An answer engine has to be able to parse your content, attribute it, and decide it’s trustworthy enough to cite. That’s a question about how your information is organized, marked up, and served. It’s a data problem wearing a marketing costume.
What machines need that humans didn’t
A human visitor forgives a lot. They’ll scroll past a wall of text, squint at a PDF, and infer from context that the “Pricing” tab applies to the service described three pages earlier.
A model won’t. It needs claims that stand on their own, structured data that says what a page actually is, clean canonical URLs, and content that isn’t trapped inside images or client-side JavaScript that renders after the crawler leaves. Most SMB sites I audit fail on at least two of those, and they’ve been failing quietly for years because human visitors covered for them.
The businesses getting cited right now aren’t the ones with the best copy. They’re the ones whose content is legible to a machine and specific enough to be worth quoting.
You can’t fix what you can’t see
Years ago at Kyndryl, I led an engagement where the client asked for a 25% improvement in incident detection. We delivered 38%, and more than $2 million in documented savings followed. What surprised everyone was the mechanism. We didn’t get meaningfully better at fixing things. We got better at knowing things were broken.
Instrumentation was the whole game. The savings were downstream of visibility.
The same principle applies here, and almost nobody has it in place. Most analytics setups bucket AI referrals into “direct” or generic “referral” traffic, so the one channel that’s growing is invisible in the report where decisions get made. Conductor’s data suggests ChatGPT alone accounts for the large majority of visible AI referrals and a meaningful share of AI-influenced traffic arrives with no referrer at all, which means the visible number understates the real one.
If you do one thing this quarter, do this: build a dedicated AI channel grouping in GA4 that separates traffic from ChatGPT, Perplexity, Gemini, Copilot, and Claude out of the direct bucket. It takes an afternoon. Then track conversion rate by source, not sessions by source. You will almost certainly find that your worst-looking channel is your best-performing one.
Four moves, in order
Instrument first. Separate AI referral traffic in your analytics before you change a single page. Otherwise you’re optimizing blind and you’ll have no way to prove whether anything you did worked.
Change the metric that gets reported. Revenue per visitor and conversion rate by source, not total sessions. If your monthly report leads with traffic volume, it will keep telling you a scary story about a business that’s actually getting healthier.
Make your content machine-legible. Schema markup, server-rendered content, clean URL structure, claims that are specific and self-contained. This is unglamorous infrastructure work, and it’s the part that determines whether you get cited.
Stop writing for the click and start writing to be quoted. Specific numbers, named sources, direct answers near the top of the page. The content that gets cited in AI answers is content that makes a checkable claim. Vague thought-leadership prose gets summarized away and never attributed.
The uncomfortable part
I’m not going to tell you this nets out fine. For a lot of businesses it won’t. If your model depended on high-volume informational traffic and display advertising, the math genuinely got worse and no amount of schema markup fixes that.
But for most of the SMBs I work with where the site exists to generate qualified inquiries, not pageviews the shift is closer to neutral than the traffic chart suggests. Smaller audience, better fit. That’s survivable. It’s arguably better.
What isn’t survivable is spending 2027 optimizing for a referral model that stopped working in 2026 because your dashboard never showed you the switch.
So here’s the question I’d put to your team this week: can you tell me, right now, how much of last month’s traffic came from an AI engine and how it converted compared to everything else? If the answer is no, that’s not a reporting gap. That’s the whole problem, and it’s the cheapest thing on this list to fix.
Not sure whether your infrastructure is legible to the systems that now decide who gets found? That’s exactly what our AI Infrastructure Assessment surfaces. Start the conversation.
Want to see how we approach data readiness and integration? Our services overview walks through the infrastructure-first sequence we use with every client.
Working through this yourself? There’s more on data readiness, AI cost, and vendor evaluation on the Summit AI blog.
Russell Love is the Founder & CEO of Summit AI Business Solutions, based in Browns Summit, NC. With 20+ years of enterprise transformation experience at IBM and Kyndryl, Russell helps businesses build the foundations that make AI actually work.