Social Rocket
Automated web traffic is increasingly becoming a measurement problem for advertisers. According to a recent analysis by Digiday, agencies and e-commerce brands report that bots and AI agents are contaminating retargeting signals, attribution and audience models. At the same time, many retailers do not want to block legitimate AI agents, because these can already bring qualified traffic and purchases.
Bots no longer only affect publishers
Cloudflare now describes more than half of all web requests as automated, including classic bots, crawlers and AI agents. For marketers, the crucial point is that automated does not automatically mean worthless. AI assistants can research products, visit pages or prepare purchase processes on behalf of real users.
The British retailer John Lewis illustrates this: according to Reuters, the share of agentic searches on its own website rose from 0,3 to 2,5 percent within a year. The company is therefore deliberately investing in content that both people and AI systems can evaluate better.
It becomes problematic where automated traffic is treated like genuine user interest.
Retargeting can receive false signals
The media agency GoFish, interviewed by Digiday, reports that for its e-commerce clients bot traffic is on average around 80 percent higher than the previous year. Retargeting in particular is affected: bots can visit product pages, fill shopping carts or trigger newsletter forms, and thereby land in audiences that are later targeted again with paid advertising.
For affected clients, GoFish responded with narrower retargeting criteria. According to the agency, this raised the average CPM by around 20 percent. Some advertisers shifted budgets away from classic programmatic retargeting toward platforms such as Google, Amazon or Meta.
These figures are agency observations and not a general market benchmark. They do, however, point to a structural problem: the better bots imitate human intent signals, the less a simple page visit suffices as a basis for media buying.
For attribution, signal quality will matter
For marketers, the question is therefore shifting from "How many users interacted?" to "Which interactions likely come from genuine prospects?". Signals with higher purchase proximity become more meaningful, such as verified logins, recurring sessions, completed qualified form steps or actual transactions.
First-party data also keeps gaining importance. It can be linked more closely to your own business events than anonymous website signals from open retargeting pools. At the same time, teams need cleaner bot classification instead of blocking all automated traffic indiscriminately.
This fits a broader shift in the digital advertising market: AI in media buying increases the need for independent control, while platforms simultaneously keep automating their own measurement logic. At Meta, for example, attribution between clicks and engagement was recently reorganised[](/blog/meta-ads-ki-creatives-attribution-2026/).
Good AI agents and bad bots are not the same
This creates a conflict of goals for retailers. On the one hand, scrapers, fraud bots and automated sessions can degrade advertising data. On the other hand, agentic commerce is growing: users let AI systems research, compare and in some cases already carry out actions on websites.
Cloudflare itself is therefore pursuing a more classification-based approach and wants to give website operators more control over which agents crawl content and for what purpose. For marketing teams, the same logic should make sense: not "bot or human" as the only category, but distinguishing between search crawlers, training crawlers, genuine user agents and suspicious automated traffic.
The case also reminds us how little transparency advertisers sometimes have about the mechanics behind digital advertising systems. Only a few days ago, the FTC sued Amazon over allegedly hidden surcharges in advertising auctions.
What marketers should check now
If you rely heavily on retargeting, evaluate unusual traffic spikes, extremely short sessions, repeated technical patterns and unusual conversion paths separately. It is especially important to ask whether audiences are based on mere page views or on events that reflect genuine intent far better.
The broader trend is clear: with more AI agents on the open web, traffic does not automatically become more valuable. For performance marketing, the ability to distinguish genuine demand from automated activity is becoming a distinct part of measurement strategy.
Sources: Digiday, Cloudflare, Reuters