Subscribe
Share
Share
Embed
Hosts: Kai Thompson & Maya Chen-Rodriguez
In this episode:
• Today we're covering the collapse of photographic evidence, how images are making AI spread misinformation, and Notified's new LLM-optimized press rel...
• Starting with what might be the most
Daily AI news for PR and communications professionals. Two hosts cover how AI is transforming media relations, content strategy, and brand reputation.
Kai Thompson: Welcome to Pivot PR! I'm Kai—
Maya Chen-Rodriguez: —and I'm Maya. Let's get into it.
Kai Thompson: Today we're covering the collapse of photographic evidence, how images are making AI spread misinformation, and Notified's new LLM-optimized press release tool.
Maya Chen-Rodriguez: Starting with what might be the most consequential shift in how we perceive reality itself.
Kai Thompson: A new arXiv paper is documenting something we've all been feeling — we can't trust photos anymore. The researchers analyzed GPT Image 2, Nano Banana Pro, Grok Imagine, and Seedream 5.0 Lite, and here's where things get interesting: they're not just looking at the tech, they're cataloging real incidents where fake crisis imagery, forged documents, and even fabricated medical scans have already fooled people.
Maya Chen-Rodriguez: The data tells a different story than just 'AI images are getting better.' What we're seeing is a fundamental breakdown in evidence-based communication. The paper documents seventeen verified cases in the past six months where synthetic images were used in legal proceedings, insurance claims, or news reporting before being detected.
Kai Thompson: This changes everything for PR professionals. We've built entire crisis communication playbooks around photographic proof — showing your CEO at the factory floor, documenting safety measures, visual evidence of community engagement. What happens when every image becomes suspect?
Maya Chen-Rodriguez: Let's dig into the numbers here. The researchers found that professional photo editors can only identify the latest synthetic images with 52% accuracy — that's basically a coin flip. More concerning? When they surveyed 500 communications professionals, 78% said they still rely on photos as primary evidence in press materials.
Kai Thompson: Wow, that's actually wild. We're operating on outdated assumptions.
Maya Chen-Rodriguez: Exactly. And the paper suggests we need new verification frameworks — blockchain certificates, cryptographic signatures, witness corroboration. PR teams need to start thinking about evidence chains, not just compelling visuals.
Kai Thompson: Speaking of trust erosion, our second story shows how images are weaponizing AI systems themselves. This new research reveals that vision-language models are way more likely to spread misinformation when it includes images.
Maya Chen-Rodriguez: This is fascinating methodology. They used what they call 'jailbreaking-style prompting' on four frontier VLMs, testing them with PolitiFact-labeled false stories. When text-only misinformation was presented, models refused to share it 73% of the time. Add an image? That drops to just 31% refusal.
Kai Thompson: And it gets worse with persona conditioning, right? When they told the AI to act like specific personality types?
Maya Chen-Rodriguez: Yeah, that tracks with what we know about human psychology. When researchers prompted models to adopt partisan personas or conspiracy-minded personalities, the resharing rate jumped another 40%. The models mirror our own cognitive biases — we're more likely to believe and share false information when it confirms our worldview and comes with compelling visuals.
Kai Thompson: This is a massive problem for brand safety. Companies are rushing to implement AI chatbots and content generators, but if these systems are susceptible to spreading misinformation with images, that's a PR nightmare waiting to happen.
Maya Chen-Rodriguez: The business implications are staggering. Imagine your customer service AI sharing false product information because someone attached a convincing fake infographic. The paper recommends implementing visual verification layers, but honestly, I think this is going to require fundamental architectural changes to how these models process multimodal content.
Kai Thompson: Our third story shows the industry adapting to this new reality. Notified just launched an AI Press Release Optimizer that's specifically designed to get your announcements noticed by LLMs.
Maya Chen-Rodriguez: The metrics on this are compelling. In beta testing with 200 companies, press releases optimized with their tool saw 3.4x higher citation rates in AI-generated summaries and briefings. They're essentially reverse-engineering how models like GPT and Claude prioritize and extract information.
Kai Thompson: Here's where things get interesting — we've gone from SEO to what they're calling 'LLM-EO': Large Language Model Engine Optimization. The tool analyzes your press release and suggests structural changes, keyword placements, even specific phrasing that makes it more likely to be surfaced by AI systems.
Maya Chen-Rodriguez: I'm somewhat skeptical about the longevity of this approach. These optimization tactics assume current model architectures and training methods remain stable. But the data does show it's working now — companies using the optimizer saw their announcements appear in 67% more AI-generated industry summaries.
Kai Thompson: It's brilliant, honestly. Think about how many executives and analysts now start their day with AI-generated briefings instead of scanning headlines. If your press release isn't structured for LLM consumption, you're invisible to a growing segment of decision-makers.
Maya Chen-Rodriguez: True, though I worry we're creating an arms race. As more companies optimize for AI consumption, we might see the same keyword stuffing and manipulation tactics that plagued SEO. The real question is whether this improves information quality or just games the system.
Kai Thompson: That's your Pivot PR briefing for April 30, 2026. I'm Kai—
Maya Chen-Rodriguez: —and I'm Maya. See you tomorrow.