The number that should worry you this week is 9.4 percent. That is how much weekly traditional search dropped once ChatGPT Search opened to everyone, according to a Bocconi University study of US desktop browsing.
A recent piece on Search Engine Journal walked through the data. The first-order story is obvious, people ask ChatGPT instead of Google. The second-order story is the one that costs you money.
It Takes More Than It Gives Back
Google sends people to external sites in 31.1 percent of sessions. ChatGPT does it in 5.2 percent. So every query that moves from one to the other is not a clean swap, it is a net loss of outbound traffic to the open web.
The access timeline makes the effect visible. ChatGPT Search went from paid subscribers in October 2024, to free users by December, to anonymous users by February 2025. Each time the door opened wider, traditional search queries fell further.
The decline was not a one-week blip either. It reached 17 percent after twenty weeks. That is not a novelty spike settling back to normal, that is a habit forming and holding.
Here is the part most dashboards miss. The traffic ChatGPT removes from search does not show up as ChatGPT referral traffic on the other side, because ChatGPT barely refers at all. The visit simply ends inside the answer.
Look at who does get ChatGPT traffic and the picture sharpens. The referrals it sends favor reference, academic, developer, and SaaS sites. Ad-supported sites, the ones whose whole business model depends on the click, receive 27.6 percentage points less traffic from ChatGPT than from Google. If your revenue rides on people arriving and seeing something, you are on the wrong side of that split.
ChatGPT own referral rate did climb over the period, from roughly 2.5 percent to nearly 6.5 percent of sessions monthly. Encouraging, until you remember it is still a fraction of Google rate. The channel that is growing sends a trickle, the channel that is shrinking sent a river.
Informational Content Is the First Casualty
The drop was not spread evenly. Transactional and recreational searches barely moved. Informational searches, the how-does-this-work and what-is-the-difference queries, took the hit.
Academic referrals fell 32.8 percent. Reference material fell 26.5 percent. If your content strategy is built on being the helpful explainer that ranks for informational queries, that is precisely the ground shifting under you.
This connects to something I wrote about in your metrics broke, not your content. The pages are still doing their job of informing the buyer. The session counter just stopped recording it, because the buyer got informed inside a chat window and never clicked.
There is a strategic read here for anyone selling into a considered purchase. Your explainer content still shapes the decision, ChatGPT is now reading it and repeating it, but you no longer get the visit that used to prove the content worked. The influence survived, the measurement did not.
What to Do Before the Curve Steepens
Stop treating this as a search problem to be fixed and start treating it as a channel that is being resized. Traditional search is not vanishing, it is getting smaller for informational intent while ChatGPT absorbs the top of that funnel.
First, find out whether the models even mention you. Being read by ChatGPT is worthless if the model never names your brand as the source. This is the actual battleground now, and I built GEOflux (geoflux.ai) specifically to measure where a brand stands inside AI-generated answers and why.
Second, move budget toward the intents that still travel. Transactional and recreational queries held steady in the data, which means bottom-funnel and branded search are more durable than the informational middle. That is where a click still means something.
Third, give people a reason to leave the answer box. A model can repeat your sentence, it cannot hand the reader your calculator, your data set, or your community. The content that survives is the content the summary cannot fully reproduce, a point I made in survival in the zero-click era.
One practical experiment I run with teams. Take your top ten informational pages by past traffic and ask ChatGPT the exact questions they used to rank for. Watch whether it names you, paraphrases you without credit, or ignores you entirely. That single hour reframes the whole conversation from why is traffic down to are we even in the answer.
The mistake I see brands making is waiting for this to stabilize before reacting. It is not stabilizing, it is compounding. A curve that goes from 9 percent to 17 percent in twenty weeks is not asking for a wait-and-see meeting.
The click was never the point. The influence was. The brands that understand the difference will keep showing up inside the answers their buyers actually read, while everyone else keeps optimizing for a search box that fewer people open.