Automated agents and bots now account for more than half of all web requests, per Cloudflare, and buy-side teams are splitting into two camps: block the traffic before it poisons retargeting pools, or classify it and let the highest-converting bots — actual AI shopping agents — through. Neither answer is settled, and the split shows up inside the same agency roster.
01What’s actually changing
- AI-agentic search is a measurable, fast-moving share of traffic. At John Lewis, agentic searches rose from 0.3% to 2.5% of all web visits in a single year, per Reuters reporting — small in absolute terms, large in growth rate.
- GoFish’s e-commerce clients saw bot traffic rise 80% year-on-year, president David Dweck says, with degradation starting in Q4 of last year and dragging into this one: “Those actions got muddied, or poisoned, with a lot of bot traffic.”
- The agency’s fix cost money to apply. Narrowing retargeting parameters to filter out the noise pushed average CPM up 20% for those same clients — the cleanup has a real price, not just an engineering cost.
- Not every operator is blocking. Mellow Sleep co-founder Chad Keller: “We’re less interested in blocking automated traffic than in classifying it. AI assistants are already sending us traffic that converts several times better than our site average.” His view: blunt blocking “would cost us more than it saves.”
- Bots don’t just add noise — they corrupt the signal. Lighthouse Creative’s Mallory Beck says bot traffic inflates impressions, muddies lookalike-audience models, and compromises frequency caps, the mechanics several agencies said were pushing client budget out of programmatic and into social and retail media.
02Why an operator should read this differently than a marketer would
A brand-side team can route around bad retargeting data by shifting budget to another channel. A publisher can’t route around the same traffic sitting inside its own GAM inventory: undetected bot volume inflates impression counts, distorts viewability and invalid-traffic (IVT) rates, and — because MCM and PPID matching both depend on distinguishing a real user from an automated one — degrades the exact signal buyers are now telling agencies they can’t trust. The GoFish number is the tell: a 20% CPM increase wasn’t the market repricing the inventory, it was the buy side pricing in doubt about what the inventory actually is.
That makes this the sell side’s problem before it’s a filtering problem. A publisher whose GAM invalid-traffic reporting isn’t catching the same agentic-search growth John Lewis is measuring is sitting on inventory that buyers are already discounting in practice, even if the reported numbers still look clean.
03What this changes for buyers and publishers
- Publishers should treat AI-agent traffic as a classification problem, not a binary block/allow toggle — the same distinction Mellow Sleep is drawing on the buy side. A shopping agent that converts is a different line item than a scraper inflating impressions, and GAM’s IVT tooling needs to tell them apart.
- Buyers auditing retargeting performance should ask publishers for bot/agent segmentation, not just an aggregate IVT rate. The 80% year-on-year swing GoFish saw is exactly the kind of drift that hides inside a headline IVT number that still passes muster.
- Expect more budget to keep moving toward first-party data and retail media, the same rotation several agencies already described — a hedge against a traffic-quality problem neither side has fully instrumented yet.
04The bottom line
Cloudflare’s own infrastructure now carries more automated requests than human ones, and the agencies closest to the problem can’t agree on whether that’s a threat or a channel: one client base saw bot traffic up 80% year-on-year and paid a 20% CPM premium to filter it out, while another is deliberately routing AI agents through because they convert several times better than average. Both are describing the same shift from opposite sides of it — which means the publishers supplying the inventory are the one party who has to get the classification right for everyone.
Separating bot visits from human ones is the first step; getting paid for the bot visits is the second, and it is what Archon Genesis is being built to do.