New ad channels bring new budget, and new budget attracts automated traffic. That’s the core finding from TrafficGuard’s review of OpenAI’s ChatGPT ads.
The ad verification platform analyzed 40,021 distinct ChatGPT ad clicks across 47 enterprise advertisers between August 24 and September 14, 2026. Invalid rates ranged from 0.1% on the cleanest account to 34% on the worst.
Machines, not repeat clicks
TrafficGuard applied the same 200-plus per-click signals it uses on Google and Meta. It flagged datacenter or proxy traffic, impossible device configurations, automation signatures and repeat patterns with no conversion intent.
Against 5,594,614 Google Ads clicks from the same advertisers over the same window, ChatGPT showed 3.34% of clicks from hosting, residential proxy or malicious IP space, compared to 0.92% for Google. Impossible device reports were also far higher: 0.86% on ChatGPT versus 0.09% on Google.
One detail flips the usual script. Google’s overall invalid rate was higher at 10.8%, but TrafficGuard says most of that is repeat clicking by real people. On ChatGPT, the invalid activity is machines.
What gets flagged
- Infrastructure signals: IPs tied to hosting providers, commercial proxy or VPN ranges, or recent malicious-activity lists.
- Impossible devices: a phone reporting 32, 64 or 192 processor cores when real phones top out around 12.
- Automation marks: headless browsers, crawler signatures, mismatched user agents and the same user appearing in distant places too quickly.
Four hosting and proxy networks accounted for 47% of flagged clicks. TrafficGuard describes them as infrastructure owners, not necessarily operators, and one VPN egress touched 20 advertisers while most of its clicks passed filters.
Why ChatGPT advertisers should care
ChatGPT ads are young. Self-serve Ads Manager with CPC bidding opened to US businesses on May 5, 2026, and OpenAI’s pilot passed $100 million in annualised revenue within six weeks. Budgets are moving fast.
On CPC campaigns, the click is the unit you buy. If 34% of clicks on the worst account are machines, that’s spend with no chance to convert. The study doesn’t show how many flagged clicks OpenAI actually charged for, and the materials carry no OpenAI response. Treat the figure as a risk signal, not an invoice.
What to do before you scale
Treat ChatGPT like any paid channel: don’t assume new means clean. Pull click-level data, benchmark invalid rates by campaign, and ask OpenAI what reporting or credit process exists for invalid activity.
TrafficGuard sells verification, so the finding is not neutral. But the strategic point stands: low-quality traffic follows spend into any new channel.
Source: PPC Land
