Two advertisers can look at the same traffic report right now and reach completely opposite conclusions. One sees a shiny new distribution channel. The other sees a media bill that no longer makes sense.
Both are looking at the same thing: machines.
The headline number
Digiday research published on September 8, 2026 documents a split inside the buy side over what to do about non-human traffic arriving at brand sites. Cloudflare data cited in the piece puts bots and agents past 50 percent of all web requests. GoFish said clients are seeing bot traffic climb 80 percent year over year.
And the cost of dealing with it is real. David Dweck, president of GoFish, told Digiday that the adjustments made to retargeting in response pushed CPMs up by an average of 20 percent. One client saw CPA spike temporarily in Q4 and into the current year.
Why retargeting breaks first
Retargeting is the most exposed tactic in the stack, and the mechanism is simple.
You record who visited which pages. You buy media against that list. When a growing share of those “visitors” are crawlers, agents and assistants, the list stops describing potential buyers and starts describing requests.
You still pay a real auction price. You just reach a different population than the one you modelled.
The scale is not theoretical. Kinsta measured AI bots hitting WordPress cart pages 3.75 million times in a single day. Cart pages are exactly the pages a commerce retargeting pool is built from.
The optimistic half of the argument
Not everyone is upset. Chad Keller, co-founder of mattress brand Mellow Sleep, told Digiday that AI assistants send traffic converting several times better than his site average. UK retailer John Lewis has watched agentic searches grow from 0.3 percent of visits to 2.5 percent in a year, and wants more.
Small percentages, big multiple. That is the classic early-channel profile.
But there is a catch worth sitting with. Traffic that converts several times better is traffic where intent was formed somewhere else. The assistant did the comparison, the filtering and the shortlist before the click. Invoca has measured calls originating from ChatGPT converting to leads at 49 percent, ahead of every other channel.
Great conversion rate. Also a sign that a referring layer is quietly capturing the part of the funnel your ads used to be paid for.
Where the budget is going
Digiday reported clients shifting spend toward social platforms and retail media networks. That is not a technical outcome, it is a competitive one. Those are closed environments where identity comes from a login, not from an inferred request.
Open-web publishers do not control how many bots hit their pages. They absorb the cost when the resulting data is treated as unreliable. The winners are the platforms that never relied on that data.
The gap nobody has closed
Lunio found only 5.3 percent of marketers use invalid traffic tools, while 75.6 percent say they lose ad budget to bots. The same firm measured 72 percent more invalid traffic in retail search campaigns running on AI Max.
Worth noting: DoubleVerify has reported ad fraud falling 41 percent in North America and 45 percent in EMEA. That is not a contradiction. Fraud detection catches traffic trying to steal money. Most agentic traffic is not stealing anything, which is precisely why it is so hard to filter or invoice back.
What to do this quarter
- Audit your remarketing seeds. Anything built from raw pageviews is suspect. Rebuild pools from logged-in events, add-to-cart, purchase or CRM data.
- Segment agent traffic separately in analytics rather than lumping it into “direct” or “referral”. You cannot value what you cannot see.
- Watch Lookalike seeds. Site-engagement seeds inherit whatever your traffic actually is, then the model multiplies that composition many times over.
- Re-baseline CPM benchmarks. If your open-web CPMs rose ~20 percent, say why in the report before a client says it for you.
- Test AI referral traffic as its own channel with its own conversion targets, not as an anonymous slice of organic.
The core problem this year is not AI. It is the widening gap between a signal and the thing it was supposed to measure, and the growing number of parties with a reason to manufacture the signal instead.
Source: PPC Land
