Should we check whether AI agent sessions inflate the checkout and lead events we send to Meta and Google, and how do we filter them?
Our answer
Yes, check whether AI agent sessions inflate your checkout and lead events. Filter agents that identify themselves before events reach Meta and Google, and bid on backend-confirmed purchases and qualified leads rather than checkout starts or raw form fills.
- Confidence
- Medium
- Strongest proof
- 3 official rules
- 3 datasets, 3 opinions or claims
- Sources
- 8 sources
- 9 statements
- G
MCWHSVFC
New here? Take the 1-minute tour
Why this is the answer
3 reasons and our judgmentAgent sessions fire more checkout events. Microsoft Clarity found checkout events in 25.21% of sessions involving AI agents against 14.65% of human sessions, and form submits at 19.85% against 18.56%. It cannot split agent from human actions, and events are not orders.
Standard bot filters may miss agents. Google's GA4 bot filter relies on bot lists and doesn't mention agents. HUMAN Security says agent browsers look like ordinary browser visits. Cloudflare lets rules act on agents that sign their requests.
Completed agent sales can go missing instead. When an agent checks out on Google's surfaces, your site tags never fire (Tony Adam of Visible Factors, Google). Those sales go uncounted rather than inflating counts.
Our judgment
- Read agents' extra checkout events as a risk to checkout-started and lead events, not purchases. Order success does not appear in Clarity's agent top 10, whose last entry is 1.69%, while it appears in 1.79% of human sessions. HUMAN Security put 2.4% of agent activity on checkout and payment routes in May 2026 and 1.9% in October.
- Filter at the CDN or server-side tag layer rather than rely on GA4. GA4's known-bot filter can't be inspected, while Cloudflare classes agents that identify themselves by behavior, including Agent and Transact, and lets every plan allow or block its Agent preset. HUMAN Security says Comet and Atlas present as browsers and the Claude extension as Chrome, so filtering covers only part of agent traffic.
- Make backend-confirmed purchases and qualified leads the bidding events. This applies the rule in Which lead stage to optimize for and the server setup in Server-only conversion tracking to agent traffic. WeltPixel notes, about staff test orders, that no IP rule can remove a purchase, and Webtures advises a server-side purchase event labeled by order source.
- Tony Adam's view that agents break attribution for the purchases they complete doesn't change the answer for checkout and lead events. Google's help page confirms that checkout under its Universal Commerce Protocol (UCP) happens on Google's surfaces, while its agentic checkout buys on the merchant's site.
Our judgment, built from Microsoft Clarity, HUMAN Security, Google, Cloudflare, Webtures, WeltPixel and Tony Adam: our reading of these sources, not a finding from any one of them.
What the sources say
9 statements from 8 sourcesGoogle Analytics automatically excludes traffic from known bots and spiders, using Google research and the IAB list; you can't turn this off or see how much was removed. AI agents aren't mentioned.
“At this time, you cannot disable known bot traffic exclusion or see how much known bot traffic was excluded.”
Google's UCP checkout lets its agent complete purchases on Google's surfaces with you as merchant of record, for select US, Canadian and Australian merchants. It doesn't say how these orders show in Ads reports.
“With the new checkout experience enabled, the checkout happens directly on Google’s surfaces while keeping you the merchant of record.”
HUMAN Security, an agent detection vendor, says Meta's Muse agent made 59% of the AI agent requests it detected in October 2026, 3.7 times ChatGPT Agent. It gives no absolute volumes.
Cloudflare's docs say bots and AI agents that identify themselves can be verified and labeled by behavior, such as acting for a user or checking out, and acted on in all plans. Unsigned agents aren't covered.
“User-directed agents visiting a page on behalf of a human”
Microsoft Clarity's study of 5.2 million AI agent sessions found checkout events in 25.21% of them against 14.65% of human sessions, though it can't separate agent from human actions within a session.
“agentic-session volume increased by 88% in just three weeks.”
WeltPixel, which sells Shopify tracking apps, says Shopify's September 21, 2026 bot counting change was not applied to past data, so mark it on conversion charts, and keep staff test orders out at the source.
“No IP rule can remove a purchase, so keep staff test orders from being sent in the first place.”
Webtures, an analytics agency, says sales inside AI assistants never reach your site's tag, so label each order's channel in your order system and send purchases from the server; decide which agents count.
“Write a channel label on every order in your order system: web, app, ACP, UCP, Buy for Me, marketplace”
Visible Factors' Tony Adam says agent purchases lose campaign credit rather than inflate it: with UCP checkout your site pixels see nothing, so lean on Merchant Center data, server-side tracking and blended efficiency.
“What breaks is the link back to the campaign that earned the sale”
HUMAN Security, which sells bot detection, found 76.4% of AI agent activity on product and search pages and 2.4% on checkout in May 2026; agent sessions show up in analytics as ordinary browser traffic.
“agentic sessions appear in analytics dashboards as browser traffic, often without clear attribution.”
In the full answer
For readers who sign in- What to do this week3 steps
- The numbers behind it3 figures from cited sources
- Where each source standsA chart placing each source on the question
- The strongest disagreementThe best case against the answer, from a cited source
- How sure we areWhat its confidence rests on, and what would change the answer