Marketers are confronting a traffic-quality problem that no longer fits the old human-versus-bot distinction. AI assistants can arrive as useful discovery and shopping intermediaries, while automated bots can imitate the same high-intent behaviors that advertising systems use to identify prospects.
That ambiguity is beginning to affect retargeting, attribution and media allocation. Agencies say bot activity can inflate performance signals, contaminate lookalike audiences and consume frequency caps, while retailers such as John Lewis are trying to attract more agent-led visits. The challenge is not simply keeping machines out. It is deciding which machines create value.
Table of contents
Jump to each section:
- AI traffic is no longer one category
- Retargeting signals are losing reliability
- The new segmentation problem
- What marketers should know about AI traffic
AI traffic is no longer one category
The familiar assumption is that automated traffic is unwanted traffic. The contrasting reality is that AI agents can now research products, compare options and arrive at a brand site as part of a legitimate customer journey. A blanket block may protect analytics while quietly removing a new source of demand.
Mellow Sleep illustrates the upside. The direct-to-consumer pillow brand says traffic sent by AI assistants converts several times better than its site average, so it focuses on classifying automated visitors instead of excluding all of them.
John Lewis is making a similar distinction from the retailer side.
AI agentic searches rose from 0.3% to 2.5% of all web visits in one year. The retailer is positioning its web properties to attract more agent-led discovery in support of AI-enabled commerce.
The strategic implication is clear: traffic quality can no longer be inferred from whether the visitor is human.
That creates a classification problem for marketing teams. An AI shopping agent and a malicious bot may both crawl product pages or trigger events, but their intent and commercial value are entirely different. The useful boundary is shifting from person versus machine to valuable versus distorting behavior.
Retargeting signals are losing reliability
Retargeting depends on observed behavior. A product view, newsletter signup or cart action tells an ad platform that someone may be worth reaching again. AI agents complicate that logic because they can perform these same actions without representing a person who should receive an ad.
GoFish says its ecommerce clients have seen the problem move from analytics noise into media cost.
Bot traffic increased by an average 80% year over year across the agency’s ecommerce clients. The added activity contaminated retargeting pools and contributed to weaker campaign performance.
The agency narrowed retargeting parameters in response.
Average CPM increased by 20% after clients tightened retargeting rules. Cleaner audience signals came with a higher price for reaching the remaining pool.
This exposes an uncomfortable tradeoff. Broad audiences may preserve scale while wasting spend on non-buyers. Tighter filters may improve confidence while raising delivery costs and excluding useful agent activity.
Bad data does not stay inside analytics. It becomes a media-buying decision.
The effects can spread beyond retargeting. Automated visits may inflate impressions or conversions, distort lookalike modeling and consume frequency caps that were intended to manage human exposure. Attribution then becomes harder to trust because apparent engagement can be generated without a corresponding customer.
The new segmentation problem
Cloudflare describes a web in which automated traffic has crossed from edge case to operating condition.
Automated agents and bots account for more than half of all web requests. The scale makes machine traffic a core analytics and customer-journey issue rather than a narrow security concern.
For marketers, the more interesting question is not how much traffic is automated. It is whether the measurement stack can distinguish a useful agent, a scraper, a fraud attempt and an ordinary crawler before those visits shape audience models.
Several agencies are already changing the inputs they trust. Collective Measures has directed some clients toward programmatic targeting based on first-party data or retail media networks. Mediassociates points to pre-bid verification and inclusion lists, while Lighthouse Creative recommends putting more weight on higher-intent actions.
The website is becoming both a customer interface and a machine interface. Measurement systems need to recognize which role is active.
That shift places new pressure on teams that have treated bot management as an IT or security responsibility. Marketing owns the consequences when polluted traffic changes bidding, suppresses performance or redirects budget. Security may identify the visitor, but marketing still has to define what counts as commercial value.
What marketers should know about AI traffic
The immediate task is to protect decision quality without making legitimate AI-mediated journeys invisible.
Separate access from measurement. A brand may choose to let an AI agent reach product information while excluding its activity from retargeting and conversion audiences. Visibility policy and campaign eligibility do not have to be the same rule.
Raise the bar for intent. Page views and shallow events become less useful when machines can reproduce them at scale. Cart progression, authenticated activity and verified downstream outcomes may offer stronger evidence, provided teams assess the privacy and measurement implications carefully.
Audit audience contamination. Sudden changes in retargeting pool size, cost per acquisition, frequency delivery or lookalike quality may reflect automated activity rather than a shift in customer demand. GoFish’s experience shows why traffic diagnostics belong beside campaign reporting.
Give classification shared ownership. Marketing, analytics and security teams need a common definition of beneficial agents, neutral crawlers and harmful bots. Without that agreement, one team may block traffic that another is trying to cultivate.
AI agents are turning traffic governance into a marketing design choice. Brands will need to decide not only who can access their digital properties, but also which machine behaviors should influence budgets, audiences and performance narratives.
The broader change is that customer journeys may include non-human participants without becoming commercially irrelevant. Marketing systems built around human clicks will need to accommodate delegated discovery and shopping while resisting synthetic signals that merely look like demand.
The strongest measurement programs will not claim to remove every machine from the data. They will make the differences legible enough that marketers can act without confusing automation for attention.