Latest / The CMO Podcast with Fexingo: Marketing Leadership, Budgets, and Executive Strategy / How CMOs Use AI for Real-Time Brand Safety Monitoring
Transcript
- Lucas: Luna, I keep coming back to this number — in 2025, a single Fortune 500 brand ran an average of 1.2 million programmatic ad placements per month. That's roughly 40,000 per day. No human team can manually check even a fraction of those for brand safety. Luna: And that's exactly why we're seeing CMOs hand that job to AI. But the question is — does the AI actually work, or does it just flag everything and create noise? Lucas: That's the tension. Let me give you a specific case. Back in early 2025, a major CPG company — I won't name them, but you'd recognise the brand — had a flagship video campaign running on YouTube's programmatic display network. Their old system used a keyword blocklist: maybe three hundred terms like 'violence', 'hate speech', 'illegal drugs'. Lucas: And within three days of launch, an AI brand safety model they'd just deployed caught a placement on a channel that was ostensibly about 'urban survival skills'. No flagged keywords. The channel's description was clean. But the AI analysed the actual video content, the comments, the channel's other uploads — and flagged it as 94 percent likely to contain extremist rhetoric. The brand pulled the placement within 90 seconds of the alert. Luna: Wow. So the keyword blocklist would have completely missed that. And I bet the channel's name and description were designed to fly under that radar. Lucas: Exactly. The channel called itself 'Wilderness Preparedness'. The video title was 'How to secure your property in a crisis'. The comment section was where the real toxicity lived — and the AI read that too. This is the shift from reactive brand safety to proactive brand safety. Luna: So what's the tech stack here? Is this a separate vendor, or are the big platforms building it in? Lucas: Both. YouTube itself has had brand safety controls for years — but they're largely keyword-based. The newer wave comes from third-party specialists like DoubleVerify, Integral Ad Science, and Zefr. They're using computer vision and natural language processing to analyse the creative context in real time. Not just the metadata, but the actual frames and audio. Lucas: One vendor I spoke with said their model processes 500,000 ad impressions per second. It scores each one on a brand-safety scale from zero to a hundred, and the brand sets the threshold. Some set it at 95 — meaning they only block placements that are 95 percent likely to be unsafe. Others go as low as 80, which of course catches more but also blocks more safe placements. Luna: And that's the cost side. Every false positive means a missed impression. A CMO at a consumer electronics brand told me they saw their reach drop 12 percent when they tightened their AI threshold from 90 to 95. That's real money. Lucas: It is. And that's why the smartest CMOs are moving toward 'brand suitability' rather than just 'brand safety'. It's not about blocking everything that could be controversial — it's about aligning placements with the brand's specific risk appetite. A news outlet might accept placements next to political content. A family-friendly snack brand probably won't. Luna: Right. And context-aware AI can make those nuanced calls. It can differentiate between a documentary about terrorism and a video that glorifies it. Lucas: That's the key. A keyword blocklist would flag both. But an AI model that understands visual context and tone — it can see the difference. That's where the industry is heading. The FTC even issued updated guidance in late 2025 around brand safety disclosures, pushing advertisers to document their monitoring methodology. So it's becoming a compliance issue too. Luna: So if I'm a CMO listening right now, what's the first step? Do I rip out my existing blocklist and replace it with an AI vendor? Lucas: You don't rip it out — you layer. Start with a pilot on one campaign. Run the AI model in monitoring-only mode for two weeks. Compare what it flags against what your existing system flagged. You'll almost certainly see a gap. Then you can start setting automated actions. But test first. Luna: And the cost? I imagine this isn't cheap. Lucas: It's priced per thousand impressions — typically between five and fifteen cents per thousand. For a brand spending a million dollars a month on programmatic, the incremental cost is maybe two to five percent of the media budget. Most CMOs I've talked to say the ROI is obvious because one brand safety crisis can wipe out years of equity. Luna: Speaking of ROI — if these marketing conversations have sparked something you've actually used or shared with your team, that's the kind of thing that keeps this show ad-free. We run no sponsors, no mid-rolls. Listener support is what makes that possible. If you find value here, you can toss a few bucks our way at buy me a coffee dot com slash fexingo. Lucas: Yeah, it's a small gesture that genuinely helps us keep the conversation independent. And it means a lot. Luna: Alright, back to the tech. Lucas, you mentioned the FTC guidance. What specifically did they say about AI and brand safety? Lucas: The FTC's 2025 update said that advertisers using automated brand safety tools must be able to explain how those tools work — at least in principle — and must retain records of the criteria used. Essentially, you can't just say 'our AI handles it' and wash your hands of accountability. You need to document the decision logic. Luna: That's a big lift for a lot of marketing teams. Most CMOs I know aren't machine learning experts. Lucas: No, but they don't need to be. They need to ask the right questions of their vendors. How does your model determine context? What training data did you use? What's your false positive rate at different thresholds? If the vendor can't answer those, that's a red flag. Luna: And the vendors are getting more transparent. I saw that DoubleVerify now publishes an annual transparency report with performance benchmarks. That's useful. Lucas: It is. And the competition is driving innovation. One emerging approach is 'brand safety as a service' — where the AI doesn't just block placements but also suggests safer alternatives in real time. If a video pod is flagged, the system automatically reallocates that spend to a similar but vetted channel. Luna: That's elegant. It solves the reach problem. Instead of just losing impressions, you're redirecting them. Lucas: Exactly. And that's the direction CMOs should be pushing. Not just protection, but intelligent allocation. The AI becomes a decision engine, not just a filter. Luna: So to wrap — what's the one thing you'd say to a CMO who's still relying on a manual blocklist? Lucas: I'd say: test an AI layer on your highest-spend campaign for two weeks. You will see placements your blocklist missed. And you will see false positives that teach you how to calibrate your threshold. The data will speak for itself. Luna: And the cost of not testing? That's the part no one wants to think about. Lucas: Yeah. One brand safety incident — even a brief one — can go viral in hours. The damage is done before your team even wakes up. AI monitoring at least gives you a fighting chance to catch it in seconds.