The AI Search Losses Are Real. They Are Not Landing Where You Think.

Two sets of numbers are circulating about AI and search traffic, and they appear to contradict each other.

The first says the disruption is overstated. As of August 2026, Google still accounts for 87.6% of search referrals. ChatGPT sent 0.913% in July, roughly one ninety-sixth of Google's share, and ChatGPT referrals amount to about 0.02% of total publisher referral traffic. Perplexity is at 0.002%.

The second says the disruption is severe. Global Google search referral traffic fell around 33% in the year to November 2025. AI Overviews now trigger on roughly 30% of US desktop keywords and appear on up to 47% of Google queries, reaching about 2 billion people a month. News publishers expect a further 43% decline in search traffic by 2029 at the median, with around a fifth expecting drops greater than 75%.

Both are accurate. Reconciling them changes what you should do.

The traffic did not move. It stopped.

The common assumption is substitution: buyers left Google for assistants, so the answer is to court the assistants. The referral data does not support that. Assistants are not sending meaningful traffic to anyone. ChatGPT referrals grew more than 200% and remain under 1% of all referrals, which tells you the growth rate is impressive and the base is negligible.

What happened instead is that a large volume of queries now resolve without a click at all. The user reads the answer and leaves. That traffic did not go somewhere else. It stopped existing.

This matters because the two readings produce opposite strategies. If traffic moved, you chase it. If traffic ceased, chasing is wasted effort and the question becomes what replaces the function that traffic served.

“The traffic did not move to assistants. It stopped existing. If traffic moved, you chase it. If traffic ceased, chasing is wasted effort.”

Who absorbs the loss

Not evenly. Smaller sites lost roughly 60% of their search referrals over two years, against 22% for large publishers.

That gap is the most actionable number in the whole set. A large publisher has brand queries, direct traffic, apps, newsletters, and enough authority to be the source an AI Overview cites. A small site had one channel, and it was the one that contracted.

The traffic did not move to assistants.

So the sector-level averages materially understate the damage for exactly the kind of operation most people reading this run. If your plan was to publish consistently and accumulate organic traffic over eighteen months, the historical returns on that plan were calculated in a market that no longer exists.

What is worth doing instead

Being cited is worth pursuing, but understand what it pays. A citation does not deliver a visitor. It delivers a mention inside an answer, which is closer to being recommended in a conversation than to being ranked. That is genuinely valuable and it does not show up in analytics, which means you cannot manage it the way you managed search.

The channel that behaves like search used to is one you own. An email list does not depend on an intermediary's decision about whether to show your page, cannot be reduced by an interface change, and produces a measurable response. It was always the recommendation. The difference now is that the alternative has stopped working, which removes the argument for deferring it.

For distribution beyond that, the places where your buyers already gather have not been disintermediated at all. A conversation in a community where the people who need your product are already talking is unaffected by any of these figures.

The check worth running

Pull your own referral trend for the past twenty-four months and compare it to the 60% and 22% figures. If you are closer to 60%, you are in the group the averages hide, and the correct response is to reduce dependence rather than to optimise harder within it.

One further detail worth noting for anyone tracking assistant visibility: ChatGPT's share of AI referral traffic fell from about 89% to 63%, with Claude climbing to 18.5% and becoming the second largest source. Checking only one assistant no longer tells you much.

See ranking first and invisible in AI answers for the visibility audit, generative engine optimisation for solo founders for the practice, and building the email list before the product for the channel that survives this. The marketer track covers the wider shift.