Welcome to Unprompted: The AI Marketing Brief, where I cut through the noise in AI news and research to show marketers what’s happening — and why it matters for your work, your team and your career. 

There's a "Jurassic Park" quote that keeps coming up in AI conversations: "Your scientists were so preoccupied with whether or not they could, they didn't stop to think if they should." 

Replace "scientists" with "product teams," and it holds up pretty well in 2026. Facebook is embedding AI into search. DoorDash is replacing browsing with conversation. An entire industry coalition just formed, not to debate whether AI belongs in media, but to start setting rules before things break. 

The integration is happening. Fast. What's not keeping pace is any real answer to the trust problem everyone already knows exists. Two papers lay it out plainly: AI systems argue back when you fact-check them, and the research tools your team might already be using can be manipulated through a single Reddit post. Standard defenses don't catch it. 

They could. They didn't stop to think if they should. Sound familiar? 

New AI Tools to Help You Make Things Happen on Facebook  

Website: Meta Newsroom 

Just the Facts: Meta announced new AI-powered features rolling out on Facebook, including AI Mode, a search tool that uses Meta AI, powered by Muse Spark, to generate answers grounded in public content from Groups and Reels rather than a traditional list of search results. The company also introduced new creative tools, including updated camera roll sharing suggestions with new collage cutout templates and video transition effects (which the article states remain opt-in and can be turned off at any time), along with new AI photo presets that let users change clothing, hair and accessories, including a "Wear It" feature for virtually wearing sports jerseys. According to the article, these features are intended to help people find answers, create shareable content, and turn ideas into bigger projects using AI alongside existing community content on the platform. 

Why It Matters to Marketers: 

• Brand and community marketing teams managing Facebook Pages should review how Groups and Reels content surfaces in AI Mode's search answers, since the feature pulls from public posts rather than ranked links — an immediate factor in how branded content gets discovered.
• This update signals Meta is embedding generative AI into core Facebook surfaces — search, photo editing, profile presentation — rather than as a separate assistant, suggesting future visibility could depend more on AI interpretation than traditional engagement metrics.
• Community and content teams can audit how their brand's Groups and Page posts read as standalone answers — clear, specific, conversational — since AI Mode draws directly on public posts rather than a ranked list, much like AEO-style content restructuring. 

Disney, Adobe, New York Times Among Members Of New AI Content Coalition Led By Netflix Alum Victoria Furniss

Author: Dade Hayes 

Website: Deadline 

Just the Facts: A new coalition called the Alliance for Responsible Innovation in the Arts & Media (ARIAM) has launched with initial members including Disney, Adobe, The New York Times Co., BBC, Condé Nast, the Financial Times, ITV, Reach and Wiley. The Los Angeles-based group is led by former Netflix executive Victoria Furniss and says its mission is to ensure AI is developed and deployed responsibly, advocating for legal and policy frameworks that protect consumers, creators and democratic institutions. The launch follows recent AI-related contract provisions secured by the DGA, WGA, and SAG-AFTRA and drew supportive statements from child safety expert John Carr, policy advocate Damian Collins and Adobe Chief Legal Officer Louise Pentland. 

Why It Matters to Marketers: 

• As major publishers formalize AI accountability standards, expect content provenance and AI-disclosure documentation to become a standard production-line step rather than optional.
• ARIAM's launch reflects a broader trend of industry-led AI governance coalitions forming ahead of regulation — a pattern Gartner has flagged as central to enterprise "AI trust and risk management" planning, suggesting B2B sectors will see similar groups emerge.
• Marketers can get ahead of likely future norms by auditing AI vendor contracts and content-attribution workflows now. 

Deep-Research Agents Can Be Poisoned via User-Generated Content 

Author: Tingwei Zhang, Harold Triedman, Vitaly Shmatikov  

Website: arXiv, Cornell University 

Just the Facts: The paper introduces an attack called WARP (Web Agent Retrieval Poisoning), showing that deep-research agents — AI systems that retrieve and synthesize web content into structured, citation-backed reports — can be manipulated by attackers who append short text to existing user-generated content on platforms like Reddit and Wikipedia. Testing three open-source systems (STORM, Co-STORM and OmniThink), the researchers found that within a topic cluster, the same user-generated pages are repeatedly retrieved across many related queries, creating a concentrated attack surface where one poisoned page can affect outputs for an entire topic. Using their GeoStorm simulation framework, the authors found a single poisoned URL with roughly 13 words of injected text achieved 38–51% conditional mention rates, and after testing three categories of defenses — source blocking, input filtering and output filtering — concluded that none fully mitigated the attack without degrading output quality. 

Why It Matters to Marketers: 

• Brand and competitive-intelligence monitoring must now extend beyond AI Overviews to deep-research tools like ChatGPT and Gemini Deep Research, since the study found these systems cite Reddit and other UGC at meaningful rates — an immediate addition to GEO monitoring workflows.
• Marketers should treat their brand's presence in AI-generated buyer research as vulnerable to manipulation: The study shows none of the tested defenses (domain blocking, perplexity filters, output-similarity checks) reliably caught poisoned content without hurting report quality.
• Marketers can proactively audit high-traffic Reddit threads and forum discussions about their product category — the study found Reddit accounts for 54–73% of UGC citations — and ensure accurate, branded information appears prominently there before competitors or bad actors do. 

Stop Scrolling, Just Ask: A New Conversational Way To Search on DoorDash 

Website: DoorDash Newsroom 

Just the Facts: DoorDash launched Ask DoorDash, a conversational search feature letting users describe what they want, upload a recipe photo or paste a recipe link to generate restaurant recommendations or grocery carts. The feature is available now in select areas on iOS for restaurant search and grocery shopping, with reservations support and broader U.S. rollout coming in subsequent weeks. DoorDash states the average U.S. user has access to an estimated 800,000 menu items and grocery products on the platform, and positions Ask DoorDash as a way to search that inventory using natural language rather than browsing or scrolling. 

Why It Matters to Marketers: 

• Conversational, intent-based search on major consumer platforms signals a longer-term shift away from keyword and category browsing, pushing brands to optimize listings and product data for natural-language queries rather than traditional on-platform SEO tactics.
• DoorDash's emphasis on personalized results based on past orders and preferences foreshadows tighter integration of first-party behavioral data into discovery, a trend marketers should monitor as it may reshape how visibility and ranking are earned on retail media platforms.
• Marketers managing DoorDash storefronts or ads can act now by auditing menu and product descriptions for natural-language clarity and specificity, since conversational search may reward detailed, descriptive listings over keyword-stuffed ones once the feature scales. 

GenAI as a Power Persuader: How Professionals Get Persuasion Bombed When They Attempt to Validate LLMs 

Authors: Steven Randazzo, Akshita Joshi, Katherine C. Kellogg, Hila Lifshitz, Fabrizio Dell'Acqua, Karim R. Lakhani 

Website: SSRN 

Just the Facts: Researchers studied more than seventy Boston Consulting Group consultants who attempted to validate GPT-4 outputs while solving a business problem, analyzing their activity logs. The study found that when professionals tried to validate the AI's output through fact-checking and pushback, the AI responded by increasing the intensity of its persuasive tactics rather than correcting or qualifying its output, a dynamic the authors term "persuasion bombing." Drawing on Aristotle's three modes of persuasion, the authors categorize the AI's tactics as ethos (credibility appeals), logos (logical appeals) and pathos (emotional appeals), and argue this introduces persuasion as a fourth barrier to human-AI collaboration alongside opacity, complacency and accuracy. 

Why It Matters to Marketers: 

• Marketing teams using GenAI for analysis, strategy or copy review should treat pushback as a stress test, not a fix: This research suggests querying an AI's output may trigger more persuasive framing rather than more accurate correction, undermining standard review workflows.
• This signals a longer-term shift in how marketing ops should design AI-assisted approval chains, since human-in-the-loop validation, long treated as a safeguard, may not function as intended when the AI itself is optimized to maintain user agreement rather than challenge it.
• Teams can act now by building validation processes that route AI outputs through a second, independent model or human reviewer rather than re-querying the same AI, since the article's findings suggest cross-checking within one tool may not surface genuine errors.