OpenAI is giving ChatGPT advertisers a richer creative canvas and more conversion plumbing. But a new verification report is asking the harder question: who is actually doing the clicking?
Visual ads come to image generation
OpenAI confirmed it will begin testing visual ad formats in the U.S. later this month with a group of advertisers. The formats will appear during image generation and can show products, how they are used, or experiences associated with a service. OpenAI says the ads will remain separate from the image being created and will not influence ChatGPT’s answers.
The visual format is a meaningful step beyond the existing ChatGPT ad unit, which has largely been a sponsored placement with an advertiser name, logo and product link. OpenAI says ChatGPT now reaches 1.2 billion people each week, and visual ads are a way to turn more of that attention into inventory.
The measurement stack expands
At the same time, OpenAI is expanding conversion measurement through integrations with Hightouch, Tealium and LiveRamp. Marketers will be able to pass conversion events into ChatGPT Ads, which should make performance easier to evaluate against lower-funnel actions.
But some clicks are not coming from people
TrafficGuard’s three-week analysis of ChatGPT ad traffic from 47 advertisers found invalid click rates ranging from 0.1% to 34%. Some campaigns were nearly clean; others paid for traffic that had little chance of converting.
- ChatGPT ad clicks were 3.6 times more likely than Google Ads clicks to come from datacenters or proxy networks.
- ChatGPT ad clicks were 10 times more likely to come from devices with configurations that could not exist in the real world.
- Just 15 servers at one hosting provider clicked ads from five advertisers 318 times during the period.
“Even though ChatGPT advertising is in its early days, automated traffic has already found it. It is coordinated, it is persistent, and it is clicking every day,” TrafficGuard co-founder and CEO Mathew Ratty said.
Why the spread matters
The 0.1% to 34% range is the detail performance teams should focus on. Platform-level averages won’t protect a single campaign, because one account’s traffic can be nearly clean while another’s is dominated by non-human clicks. Segmentation by placement, geography and device type becomes essential.
The image-generation moment is contextually interesting for advertisers. A user visualising a product mockup, room design or travel scene is already in a commercial mindset, so nearby ads could feel less interruptive. But relevance only matters if the click behind it is human.
How to test ChatGPT ads without leaking budget
AI-assistant inventory is compelling because it is early and less crowded. But the TrafficGuard data suggests ChatGPT Ads should not be treated as a set-and-forget channel.
A practical guardrail is to start with a small, time-boxed test, use the new conversion integrations from day one, and compare verified conversion quality against Google or Meta baselines before increasing spend. Ad verification should be non-negotiable while invalid-click rates vary this widely.
OpenAI’s visual ads may mark the next step for AI assistant advertising, but budgets should follow evidence, not novelty.
Source: MarTech
