Key Takeaways
- →ChatGPT ads live in a conversational context where brand adjacency works differently than search or display.
- →Traditional brand safety controls (keyword blocklists, category exclusions) don't map to LLM conversations.
- →AI hallucinations in context can create brand risk that no amount of audience targeting prevents.
- →OpenAI's controls are basic. Brands using ChatGPT ads are early adopters in uncharted risk territory.
OpenAI opened ChatGPT ads to new markets this week. Advertisers can now bid on cost-per-click at $3–5 per click, down from the $50K minimum spend that gated the pilot. The self-serve dashboard is live. The machinery of ad buying is familiar.
But the playbook is not. ChatGPT ads run inside a conversation. Your ad doesn't sit next to content.it sits inside a context created by an LLM. That changes everything about how brand safety works.
Adjacency isn't about the publisher anymore.
It's about what the AI said before your ad showed up.
53%
of media experts cite genAI adjacency as top brand safety risk
83%
of digital experts say brand safety concerns are growing fast
Feb 2026
ChatGPT ads launched as CPM pilot for select advertisers
The Problem With Context
Why traditional rules fail here
On Google Search, brand safety works through page context. If your luxury watch brand's ad shows next to a counterfeit reseller, that's bad adjacency. You can exclude it.
On Facebook, brand safety works through audience context. Your ad goes to people in a certain age range, interest bucket, or lookalike group. You control who sees it.
In ChatGPT, context is conversational. Your ad appears after the user asks a question and the AI responds. The risk isn't the page or the audience.it's what the LLM said in that exact moment. And you can't control that.
A finance brand's ad could appear after ChatGPT makes a factually wrong claim about interest rates. A health brand could appear next to AI-generated medical misinformation. A legal brand could appear after the AI gives incomplete legal advice.
Why Brand Safety Controls Break Down
Three reasons your playbook won't work
Hallucinations aren't blocklistable
You can exclude "misinformation" as a keyword, but you can't exclude the possibility that the LLM will hallucinate misinformation on any query. Blocking keywords after the fact is too late.
Conversation topics are infinite
On search, you can exclude a few thousand bad keywords. In ChatGPT, the user might ask about literally anything. Your "safe" category placements mean nothing in a system where context is generated on the fly.
Compliance claims can't be verified in real time
If your brand makes healthcare claims, you need to know those claims won't appear adjacent to false health information. In a conversational model, that verification happens in milliseconds. You're betting on OpenAI's safety guardrails.
The real risk: you're not just running ads, you're endorsing the AI's outputs.
Every ad placement next to an AI response signals to the user that your brand trusts that response. If the AI is wrong, your brand is associated with that error.
How to Think About It
A new risk framework
If you're testing ChatGPT ads, don't rely on category targeting or keyword exclusions. Those are hygiene, not protection.
Start with small budgets on high-intent queries only. A user asking "best financial advisors" is lower risk than "should I invest in crypto." Narrow your audience to high-intent, high-intent-only queries where the AI's response is likely to be helpful and accurate.
Monitor manually. Set up alerts for any mention of your brand in ChatGPT conversations. Track which queries your ads appear on. Watch for hallucinations.
And here's the hard part: accept that you can't fully control adjacency. You're placing a bet on OpenAI's LLM to not hallucinate. That's a different risk profile than search or social.
The Uncomfortable Part
What this means for brands
Most brands will run ChatGPT ads anyway. The ROI looks good early. CPC is cheaper than Google. Conversion rates are higher because intent is crystal clear. Brands will ignore the risk because the short-term metrics work.
But the first major hallucination incident.where a brand's ad appears next to spectacularly wrong AI output and goes viral.will reset how the industry thinks about this.
Until then, ChatGPT ads are a brand safety frontier with no map. Lawyers haven't caught up. Compliance teams don't have frameworks. Media buyers are using search-era playbooks on a system that works completely differently.
If you're moving budget to ChatGPT ads, you're doing it with open eyes. Which is fine. Just don't pretend the safety controls are the same as they were before.
Related Reading
Brand safety in AI is uncharted.
The playbooks are changing. Stay ahead of the shift.
Back to all posts→