Search traffic is falling and conversions aren't. Read that carefully.

In July 2026, traffic arriving at US retail sites from AI assistants converted 60% better than traffic from everything else. That’s Adobe Analytics, measuring across more than a trillion visits. It was the 11th month in a row AI traffic came out ahead. A year earlier it was still converting worse than other channels. The line crossed and kept going.

Meanwhile your sessions graph is going the other way, and someone in your business has drawn a conclusion from it.

For most businesses that sell a product or a service, falling search traffic and flat conversions aren’t a contradiction. They’re the same event described twice. AI answers are absorbing the people who wanted a definition, and the people who wanted to buy something are still arriving. You lost the tourists. Check whether you lost the customers before you spend anything trying to get them back.

The traffic really is going

Pew Research Center tracked the actual browsing of 900 US adults across 68,879 Google searches in March 2025. When an AI summary appeared, users clicked a normal search result 8% of the time, against 15% when there was no summary. Clicks on the links inside the summary itself: 1%. They also ended the browsing session more often, 26% versus 16%.

SparkToro, working from Similarweb clickstream data, put 68.01% of US Google searches ending without a click between January and April 2026, up from 60.45% in 2024 and 49% in 2019. Their read is that AI Overviews now appear on more than 20% of searches and cut click-through by nearly 60% where they show.

So yes. Fewer people are landing. That part is real and it isn’t reversing.

What’s being subtracted matters more than how much

Semrush studied over 10 million keywords across 2025. In January, 91.3% of the queries triggering an AI Overview were informational: definitions, explanations, how-does-this-work. By October that had fallen to 57.1%, with commercial queries up from 8.15% to 18.57% and transactional from 1.98% to 13.94%.

Read that as a sequence rather than a snapshot. The feature ate the top of the funnel first and is now working its way down. A follow-up Semrush study of 600,000 keywords from November 2025 to April 2026 found the share of commercial-intent results carrying an AI Overview grew 71%, while appearances on transactional queries fell 5%.

That’s the whole picture in 2 numbers. Research queries are being answered without you. Purchase queries still send the click.

And the clicks that do arrive are better. Similarweb’s September 2025 report on generative AI had ChatGPT visitors spending 15 minutes on site against Google’s 8, viewing 12 pages against 9 and converting at 7% on transactional sites against Google’s 5%. Adobe’s May 2026 numbers agree: AI-referred shoppers spent 53% more time and viewed 23% more pages per visit. Both organisations sell analytics products and both are measuring their own panels, so treat the precision with suspicion and the direction as sound.

Who this genuinely hurts

If your revenue comes from the page view rather than through it, none of the above applies to you and I’m not going to pretend otherwise.

Ozone, which handles programmatic supply for premium publishers including the Guardian, News UK and the Wall Street Journal, reported to Digiday that between April and June 2026 ad request volumes fell 32% to 37% year on year in the US and 39% to 41% in the UK. Programmatic spend across its US and UK publishers was down 30.6% for the first half of 2026.

That’s not a measurement artefact. That’s an industry losing its inventory. If you’re funded by advertising, or by affiliate volume, or by anything else where a session is the unit sold, the tourists were the customers and losing them is a real loss.

Everyone else is in a different situation and should stop reading the same headlines as if they weren’t.

What to measure instead

Sessions were always a proxy. They were a decent one when traffic and intent moved together. They don’t any more, so retire the metric rather than mourning it.

Stop reporting	Start reporting
Total sessions	Enquiries, bookings, orders
Organic sessions month on month	Conversion rate by landing page
Rankings for terms nobody buys from	Revenue per session, tracked as a trend
Bounce rate sitewide	Branded search volume
Blog pageviews	Assisted conversions from AI referrers

Rankings for the terms people buy from still matter. Step 2 of the SEO check I run before a full audit sorts one kind from the other.

Two practical moves. First, split your referrer report so AI assistants are their own channel rather than sitting in direct or other. Most analytics setups still bucket them wrong, which means the traffic that converts best is the traffic you can’t see. Second, look at revenue per session across the last 18 months. If sessions are down 30% and revenue is flat, you have a cheaper audience, not a smaller one. Say so out loud in the next board meeting before someone approves a budget to fix it.

Then there’s the content. I’ve watched clients spend years publishing definition posts because a keyword tool said 4,000 people a month searched for the term. Those posts converted at nothing. They existed to catch a search that was never going to become a sale, and now they don’t even catch that. Losing them costs you a number on a dashboard and nothing else.

What still earns its keep is anything only you could have written. Pricing with the actual numbers. Comparisons that name the trade-off. Specifics from work you’ve done. Answers a model can’t assemble from the other 9 results, because those are exactly the pages an AI has to cite rather than absorb, and exactly the pages that convince a human once they land.

You didn’t lose your audience. You lost the part of it that was never listening.