Visible, Not Trainable: A Brand’s Guide to Media in the LLM Era

Abstract: As large language models (LLMs) and generative AI tools become new cultural gatekeepers, brands face a new challenge: making sure their media assets are visible in AI-driven environments without inadvertently feeding them into training datasets. Visible, Not Trainable guides marketing and legal teams through this balance of visibility and control. It explains how search and social feeds are giving way to conversational AI as the first brand touchpointedelman.com. It shows how forward-thinking companies like Red Bull turned media into their business model. We then explore risks (copyright exposure, loss of control) and opportunities (LLMs surfacing trusted brand content). Technical solutions – APIs, metadata shields, watermarks, and licensing – are laid out in detail. Finally, we outline strategic paths (syndication via Getty, partnerships with OpenAI/Google, and DIY brand portals) and a practical roadmap. This playbook offers brand managers and legal teams a step-by-step plan to become AI-ready: visible in LLM outputs under their own terms.

Part I: The Shift to AI Visibility

The New Media Gatekeepers

The way consumers find information is rapidly shifting from search engines and social feeds to generative AI chat and voice assistants. In the last year, “consumers have migrated en masse from search engines to generative AI platforms” – for example, 58% of early adopters now use AI for brand discovery versus 25% previouslyedelman.com. One industry observer calls AI “the new gatekeeper” for branding. In practice this means brands must become the content that AI front-ends find and surface. As business strategist Simon Sinek noted, “We used to hunt for information. Now, [information] hunts for us” – meaning that without brand-owned signals, LLMs may simply guess answers based on third-party content.

• Stat: With generative AI referrals skyrocketing (1300% surge in one 2024 study), experts predict search engines could lose half of their share by 2028edelman.com.

• LLM Response Style: Unlike paged search results, AI responses are written as concise, authoritative-sounding answers. If your brand isn’t among the top sources, you may simply be omitted – effectively invisible to customers.

• Implication: Brands can no longer rely on paid search ads (LLMs don’t use PPC) or standard social posts alone. In generative Q&A, earned media and authoritative content are king. An Edelman study found “up to 90% of citations that drive brand visibility in LLMs come from earned media”. This means trusted press releases, news articles, and official guides will dictate how AI sees your brand.

To win in this new environment, experts urge brands to adopt Generative Engine Optimization (GEO) – the AI-era analogue of SEO. The goal is to “transform [your] presence inside LLMs and protect and elevate [your] brand by ensuring the right messages are surfaced prominently”. GEO requires first auditing how AI platforms currently talk about you, then tailoring content and PR so that AI answers reflect your desired positioning. In short, brands must think of AI as the new channel – their content needs to be structured and optimized for conversation, not just campaign pushes.

From Campaigns to Conversations

In a world of generative AI, brand messaging shifts from one-way campaigns to interactive conversation. Instead of broadcasting ads or posts, brands need to prepare to be queried. That means ensuring up-to-date brand facts, images, and stories can be fetched by AI systems in real time, and that they answer consistently with brand voice. For example, a food brand’s nutritional info or a fashion brand’s design ethos might live in an AI-friendly knowledge base or API. When a user asks a chatbot “tell me about Brand X’s sustainability efforts,” the AI can query that source directly, rather than hallucinating or ignoring the brand.

This conversational turn also means brands should create evergreen content and contextual info, not just time-bound campaigns. Conversational AI places premium on contextual relevance: if your brand materials aren’t linked from news or knowledge graph entries, AI may not even “know” they exist. Content creators should craft detailed, factual resources (not just slogans) to seed LLMs’ knowledge. In practice, brands are starting to run their own Q&A chatbots (for sales or support) as experiments in this space, blurring the line between marketing collateral and AI-trained asset. By thinking in terms of conversation flows, brands can turn AI queries into brand-building opportunities.

The Red Bull Blueprint

Red Bull provides a powerful case study of the content-first brand. Rather than focus on product ads, Red Bull built its own media empire: Red Bull Media House. Today Red Bull “sells a feeling – excitement, fearlessness – via adrenaline-fueled content”. Their strategy highlights three lessons for brands in the AI era:

• Owned Media Channels: Red Bull owns platforms like Red Bull TV and The Red Bulletin magazine, along with active social accounts. These channels serve as evergreen entertainment hubs, not just ad slots.

• Original Storytelling: The brand produces full-length documentaries and films (e.g. The Art of Flight, Red Bull Rampage) that go well beyond simple product placement. These high-quality narratives let Red Bull control its story and imagery.

• Authentic Ambassadors: By sponsoring extreme sports athletes and creators aligned with its image, Red Bull generates grassroots content. These influencers create on-brand experiences and share them organically.

Key takeaway: Focus on content and experiences, not just drink promotions. By 2007 Red Bull even launched Red Bull Media House – a full-scale publisher producing sports events, TV, and magazines. This made the company as much a media entity as a beverage maker. For brands today, Red Bull’s approach shows that owning your media means owning your narrative. In the AI context, such owned content (websites, press releases, licensed footage) will become the “knowledge graph” that generative platforms draw from. If a brand controls the content, it can influence what AI says – essentially turning media into IP rather than waste.

Part II: Risks and Opportunities

4. The Double-Edged Sword of AI Training

Generative AI’s thirst for data means it often “ingests and copies content…word-for-word or pixel-for-pixel to learn patterns”. That can be a double-edged sword for brands: if their public images or text are scraped into training, the brand may suddenly gain reach as AI “learns” it – but at the cost of losing control. Content that is just online is not free for AI to use without consequences. Recent court rulings are reminding everyone: publicly available is not public domain. In one high-profile case (Thomson Reuters v. Ross Intelligence), a judge struck down an AI startup’s claim that using proprietary legal summaries for training was fair use. The court held that copying those headnotes to train an AI was infringement.

• Risk – IP infringement: If an AI company uses your copyrighted images or writing without permission, it could be liable for infringement. This includes rightsholders suing AI developers (as Getty Images did against Stability AI). Brands must vigilantly assert their IP rights or risk “transformative” AI reuse without credit. As one report notes, just because content is online “doesn’t mean it’s fair game for commercial AI training.”

• Risk – Reputation damage: Even if not legal, an unvetted AI output can harm brand image. As Edelman warns, AI chat responses sound authoritative. If LLMs base replies on outdated or biased info, customers may take it as fact. A single misleading chatbot answer could undermine months of branding work.

• Opportunity – 24/7 exposure: On the flip side, being included in AI models means your brand can reach users in entirely new ways. If done deliberately, brands can put their story into AI “sensory organs.” For instance, partnering with a chatbot to provide branded answers can significantly boost name recognition. The key is controlling how AI uses your content.

This sword must be wielded carefully: brands want the visibility of being in LLM outputs, but not the vulnerability of unauthorized use. Emerging frameworks (discussed next) aim to give brands the best of both worlds: signals and licenses that keep content visible to AI while forbidding unauthorized training.

5. Owning vs. Losing Control

The heart of the dilemma is intellectual property. Once media is online, large models may suck it up. Unless a brand’s rights are explicitly honored, it could unintentionally “gift” its IP to AI trainers. On one hand, providing access might let AI augment brand presence; on the other, it cedes the ability to dictate how that media is used or remixed.

• Copyright & Licensing: Generative AI companies increasingly recognize this tradeoff. Stock agencies (Getty, Shutterstock, etc.) now license their libraries for AI use, ensuring creators are paid. By contrast, absence of a license has led to legal trouble (the aforementioned Ross case). Brand teams must understand what training means under copyright law and plan accordingly.

• Data & Privacy: Similar issues arise with user data. Models often scrape news, tweets, or forums. Brands need policies on whether customer or employee data should be fed into third-party models.

• Narrative Control: If AI systems (like search or recommendations) start using brand digital assets, the brand effectively co-produces the user experience. Creative collateral could be repurposed (e.g. an AI using a trademarked mascot as inspiration). To maintain control, brands must either restrict their data or explicitly shape its use (via contracts or APIs).

The balance is tricky: brands need to own their media’s destiny. They should consider creating licensed data channels (see Part IV) so that AI can use the content on agreed terms. Meanwhile, defensive strategies (below) help shield core assets until rights are ironed out.

6. Opportunities for Brand Storytelling in LLMs

Despite the risks, AI also opens creative avenues. If a brand can get inside the model’s recommendations, it can tell deeper stories. For example, instead of a one-off ad, a brand might seed long-form content or a knowledge base that a chatbot can draw on. Imagine an AI assistant giving a customer a narrative about the brand’s heritage, latest innovations, or sustainability practices—all derived from approved brand sources. This conversational branding can feel more personalized and authoritative than generic ads.

Key point: earned and authoritative content drives AI answers. Edelman notes that in generative search, “earned media is the single most important driver” of visibility. So PR managers and content strategists should proactively create stories (articles, whitepapers, case studies) that highlight the brand. These become the citations AI will use. For instance, a press release on a product launch can end up as a source snippet in an AI’s answer. Similarly, high-quality how-to guides or influencer partnerships can be cited by chatbots.

In short, brands that imagine their media as part of the AI ecosystem will find new storytelling roles. Newsletters become FAQs, videos become answer fodder, and branded 3D models or voices may be requested by future multimodal models. The opportunity is to transcend the limitations of broadcast; an engaged consumer might “ask” for more about your brand and get a detailed, on-brand response.

Part III: Technical Pathways

7. APIs for Visibility, Not Training

One practical way to share brand content with AI without training loss is to use controlled APIs or plugins. Instead of having a model swallow all your data upfront, it can fetch answers on demand. For example, Microsoft’s “Azure OpenAI On Your Data” lets companies hook up their documents (PDFs, web pages, etc.) so that GPT-4 or GPT-3.5 queries the latest company info via a secure API. The AI system then cites the brand’s sources at response time rather than having them baked into its neural net.

Similarly, GPT-4 and other models now support “plugins” or connectors. A brand can create a knowledge API (for product specs, policies, archives, etc.) and register it as a plugin. When a user asks a chatbot about the brand, the model routes the query to that API. The result: up-to-date, accurate info powered by LLM reasoning, but no need to trust an entire copy of your database to a third party.

This approach offers high precision: the brand controls exactly what data the AI sees. If Apple publishes its code of conduct and plugs it into an LLM, the AI can give answers with Apple’s voice, without Apple needing to train the AI on all its docs. The enterprise version of ChatGPT even promises such control: it guarantees “We do not train on your business data or conversations”, so companies can safely build internal AI bots. In practice, brands should work with IT to expose only the needed assets via secure APIs, ensuring AI visibility without relinquishing raw training rights.

8. The Metadata Shield

To assert “do not train” preferences and protect content, brands can turn to metadata standards and web protocols:

• Robots & AI.txt: Just as websites use robots.txt to forbid search engine crawling, new proposals (like ai.txt) let content owners signal that their material is off-limits for AI training. The Transparency Coalition and industry groups advocate treating AI training directives like copyright notices. For example, one could put a snippet on pages saying “Do Not Train: All rights including AI training are reserved,” analogous to a copyright statement. Major AI developers (Stability AI, Hugging Face) have pledged to honor registries like Spawning.ai’s “Do Not Train” database.

• Schema.org & IPTC Tags: Embedding machine-readable rights info into the HTML can help. The IPTC recommends using schema.org metadata or IPTC properties to flag copyrighted content. For instance, adding <meta name="dcterms.rights" content="Copyright © 2025 Brand. All rights reserved, including for AI training."> on key pages makes explicit that this content isn’t to be copied. Similarly, image files and videos should carry IPTC copyright notices or C2PA content credentials (see below).

• Content Credentials (C2PA): The Coalition for Content Provenance and Authenticity (C2PA) provides a way to cryptographically sign media with provenance. By embedding C2PA “content credentials” into images and videos, brands can let platforms trace whether a file is original or edited/AI-generated. Google, for example, uses this to power its “About this image” feature, warning users if an image was AI-altered. Brands can adopt such standards so that AI systems (and consumers) know which media is authentic.

• Watermarks & SynthID: Invisible watermarks are emerging too. Google DeepMind’s SynthID is a technology that embeds imperceptible marks in AI-generated visuals, distinguishing them from real photos. While SynthID targets AI content, the same idea could apply in reverse: brands could embed identifiable markers or steganographic watermarks in their own images to assert origin. This doesn’t prevent training ingestion, but it provides evidence of unauthorized use if AI-generated imagery appears using that watermark.

By taking these steps, brands build a “metadata shield.” They make it harder for random crawlers and scrapers to swallow their assets. And if content does get used improperly, there’s an audit trail and a legal argument for enforcement. Crucially, these measures are most effective when combined: blocking (robots/ai.txt), labeling (schema/IPTC), and verifying (C2PA/watermarks) together form a multilayer defense.

9. Watermarks, Resolutions, and Rights

Another facet of technical protection is controlling the quality and visibility of publicly available assets:

• Visible Watermarks: Overlaying logos or text on images discourages direct reuse. Stock agencies have long done this, and brands might mark press photos or low-res website images with a translucent logo. This ensures any AI dataset scraped from your site carries a visible badge of origin. The downside is it can detract from user experience, so it’s typically applied to images that aren’t meant for high-quality display.

• Invisible Watermarks: As mentioned, tools like SynthID can embed data at the pixel level. If widely adopted, an AI model could be trained to ignore images flagged as watermark-containing (legal or technical measures could demand it). Alternatively, AI models might inject these marks into any images they output, signaling “this came from an AI” – turning watermarking into a two-way street.

• Resolution Control: Research suggests LLMs and vision models learn best from high-resolution, clear images. Some brands consider only posting lower-res versions of media online. This makes it harder for training algorithms to latch onto fine details. For example, news publishers might limit image size on their public RSS feeds or API endpoints. It’s not foolproof, but it increases the effort required to repurpose the asset.

• Rights and Licensing Metadata: Wherever possible, clearly attach a license to content. Creative Commons or stock licenses can specify non-commercial use only. While not technically enforced by AI, it communicates legal boundaries. A brand might use schema.org/license tags on images to indicate “Not for AI training” alongside the copyright notice.

These tactics must be balanced with accessibility. Brands still want their images seen (the core argument of the book!). Overzealous restrictions (tiny images, aggressive DRM) could make content worthless in marketing terms. The art is to degrade “trainability” without killing visibility. For instance, offering high-quality images on approved partner portals, while having only modest web versions for general consumption. Always pair rights metadata with user education: terms of service can explicitly forbid AI scraping. All together, these technical levers help a brand signal its preferences: “Yes, display me – but please don’t use me to train your AI.”

Part IV: Strategic Models

10. The Syndication Route

One pragmatic path is to license brand assets through established syndicators. Think of Getty, Reuters, Shutterstock, or AP – each controls large media libraries and already sells usage rights. These agencies are now moving into AI-specific licensing. For example, Getty Images launched a “Generative AI” solution: its AI generator is trained only on Getty’s own licensed imagery, and customers get indemnification and assurance of content safety. In effect, Getty lets brands tap into generative tools without giving up copyright, because Getty holds the licenses and compensates creators.

Similarly, Shutterstock signed a major six-year deal with OpenAI in 2023. OpenAI obtained a license to train on Shutterstock’s image, video, and music libraries (plus metadata), gaining high-quality data. In return, Shutterstock contributors receive royalties for their content’s use in AI. Brands could follow suit by assigning their images to such libraries. If a brand’s photos are part of a Shutterstock or Getty collection, and those collections license to AI labs, then indirectly the brand’s visuals are “visible” to generative AI under controlled terms.

Key features of the syndication model:

• Scale & Expertise: Agencies have the technology and legal frameworks to handle massive datasets. Brands piggyback on this infrastructure.

• Revenue Sharing: Creators get paid whenever AI tools use licensed images, aligning incentives.

• Intermediaries: Instead of negotiating with dozens of AI companies, a brand deals with one agency. That agency then brokers or sells its library to AI developers.

• Examples: The Associated Press (AP) even has a deal with OpenAI to license its news archive (from 1985 onward). Any brand-owned photo in the AP wire could be used by ChatGPT with attribution. These partnerships show how syndication can bring brand media into AI in a legally safe way.

In practice, a brand using syndication should review its contracts: ensure the stock or news agency can sublicense content for AI, and that the brand (and its photographers) get fair compensation or control in that context.

11. The Partnership Route

Beyond agencies, some brands might cut direct deals with AI platforms. High-profile media companies have done so. In April 2024 the Financial Times announced a partnership with OpenAI: the FT will license its content to train ChatGPT, and ChatGPT will answer queries with FT summaries and backlinks. Similar deals were inked with publishers like Axel Springer and Le Monde. While those are news brands, the model could extend to any major content owner. Imagine a technology company or retailer negotiating an agreement with Google or Amazon: in return for giving their latest product specs to an AI assistant, the assistant could cite and link those specs, driving traffic and ensuring accuracy.

Why partner directly? Brands that have valuable data (historical archives, proprietary images, product databases) become strategic content providers. In negotiations, a brand might ask for:

• Guaranteed attribution (AI outputs mention the brand).

• Quality controls (model training on the brand’s sanitized data).

• Customization: perhaps the AI model includes brand-specific personas (e.g. “Ask BrandBot”).

• Auditing rights: to verify the brand’s data is used appropriately.

These deals can be complex, but they offer full control and benefits:
e.g., the FT-OpenAI deal explicitly ties ChatGPT answers back to FT, enhancing FT’s authority and traffic. For a consumer brand, a similar tie-up could make an AI respond with an official product summary instead of a random description.

Brands should watch developments: in some cases, tech companies will likely approach key brands to deepen their AI’s knowledge base. Being proactive (via licensing or strategic partnership teams) could position a brand as a preferred content partner rather than a passive data source.

12. The DIY Route

Not all solutions require big partnerships. Brands can take a do-it-yourself approach by building their own controlled media portals and AI assistants. For instance:

• Brand Media Portal: Create a gated website or API that houses your entire brand media library (high-res images, videos, press releases). Use strong access controls. Then allow AI companies to connect only to this portal under terms (like ChatGPT plugins). This way, whenever the AI needs brand content, it fetches from your servers in real time. You get logging of accesses and ensure the data is fresh.

• Custom GPTs/Chatbots: Use tools like ChatGPT Enterprise or open-source LLM frameworks to host a private chatbot trained only on your data. Employees or customers can query it for brand information, and you can extend such a bot to the public if desired. Notably, ChatGPT Enterprise “does not train on your business data or conversations”, offering a model where the AI’s knowledge is your data at usage time. Tech teams can fine-tune an LLM on a company’s product manuals, branding guidelines, and marketing collateral to create a brand voice assistant.

• Controlled Syndication Hubs: Instead of public web crawling, provide AI companies with an official data feed (similar to how RSS feeds work). For example, publish an XML feed of brand assets with metadata tags (per IPTC guidelines) that explicitly describe usage rights. AI crawlers or partners can then subscribe to just that feed.

This DIY route is about sovereignty. The brand alone controls the content pipeline to AI. It requires building some infrastructure and setting policy, but it sidesteps reliance on external deals. It’s especially viable for large brands with enough resources. Internally, it demands new roles (AI-data managers, compliance teams) but yields maximal control.

Part V: Roadmap for Action

13. 12 Months to AI-Ready Media

Month 1-3: Audit & Policy. Inventory all brand media (images, videos, text). Catalog licenses, rights, and whether assets are online. Audit how your brand currently appears in AI tools (e.g. ask ChatGPT “who is [Brand]?” and note the sources). Simultaneously, assemble a cross-functional taskforce (marketing, legal, IT). Establish a clear policy: which content can be shared, which must be restricted.

Month 4-6: Build Defenses. Implement metadata and robots protocols. For example, add “ai: no-train” flags and copyright notices on your highest-value content. Publish a policy statement (perhaps in footer or TOS) like “All rights including AI training are reserved”. Set up an ‘ai.txt’ or equivalent indicating do-not-train. Deploy digital watermarks or C2PA content credentials on key images. Also, consider lowering resolution of open-access images or using visible watermarks on non-critical media.

Month 7-9: Establish Channels. Decide your exposure vs control strategy. If using syndication, reach out to Getty/Shutterstock/AP and negotiate inclusion of your brand assets in their AI-licensed programs. If going direct, open talks with LLM vendors about content licensing or plugins. Internally, set up APIs or data connectors so that approved AI services can fetch brand content securely. Update your website SEO and schema metadata for “Generative Engine Optimization” – e.g. ensure product pages have rich descriptions and structured data so LLMs can parse them.

Month 10-12: Educate & Iterate. Train PR and marketing teams on generative AI. Make sure any new content (press releases, blog posts) follows SEO/GEO best practices for AI indexing. Adjust legal agreements: include “AI use” clauses in photographer and agency contracts (see Appendix C). Pilot a branded chatbot (maybe using GPT-4 with your API) to test how AI cites your materials. Monitor the outcomes: use tools to track how often your brand is mentioned in AI answers, and revise tactics accordingly. By year’s end, you should have a documented AI media strategy, cleared legal frameworks, and a routine process for publishing AI-ready content.

14. Governance in the Age of Generative AI

Effective execution requires organizational alignment. Assign clear ownership: PR or Marketing leads should coordinate with Legal to enforce IP and contracts, while IT/TAM (Technology Asset Management) handles the technical hooks (APIs, metadata). Establish a cross-functional AI committee to oversee generative initiatives. This group ensures your LLM-visibility efforts align with brand values and legal constraints.

Key governance steps:

• Policy & Training: Develop internal guidelines on how company data and media can be used by AI. Provide training for employees (e.g. “What can you share on ChatGPT?”) and partners on AI compliance.

• Contract Updates: Work with legal to add AI clauses in vendor/content agreements (see Appendix C). For example, require any agency or freelancer to affirm AI-usage restrictions on submitted work.

• Monitoring & Audits: Regularly audit AI citations (use generic queries, see what sources pop up). If your content is being misrepresented, have escalation procedures.

• Executive Oversight: Keep C-suite informed – generative AI can pose brand and compliance risks. Frame AI visibility as a strategic project, not just a tech experiment.

As Edelman’s AI team notes, shaping brand presence in AI “requires a truly cross-functional approach”. Marketing, IT, legal, and even HR (for training) all play roles. Encourage communication across teams: for example, let legal review AI licensing deals, let marketers craft the needed content, and let tech professionals implement it. With this governance, brands can stride forward confidently rather than play catch-up.

15. The Future of Brand Media

Looking ahead, generative AI will only get more multimodal. Brands must prepare for a world where AI doesn’t just chat but sees and hears. Think interactive video assistants, virtual brand environments, and voice-generated ads. In that future:

• 3D & AR Assets: Companies should consider creating 3D models of their logos or products and publishing them (with rights metadata) so AR/VR generators can use them correctly.

• Brand Voices: Voice cloning technology means a brand’s “tone” might be modeled. It may become standard to register voice trademarks or release official AI voice banks.

• Synthetic Influencers: The line between AI-generated and human content will blur. Brands might adopt or license synthetic influencers. They should decide if and how to authenticate them (e.g. “AI-created content” labels).

• Augmented Reality Campaigns: As AR heads-up displays become common, brands may overlay virtual content in physical spaces. Ensuring these assets are tagged and governed will be critical.

In essence, every new channel will be an AI channel. Brands that build infrastructure now—ranging from solid metadata to creative AI-ready content—will be best positioned. They’ll not only survive in the LLM era but thrive, turning AI from a risk into a high-fidelity broadcast of their identity.

Appendices

A. Glossary of AI and Media Licensing Terms

• LLM (Large Language Model): A type of AI model (like GPT-4) that generates text (and sometimes images) based on training data.

• Generative AI: AI that creates new content (text, images, audio, video) rather than just analyzing it.

• Training Data: The collection of text/images/etc. used to train an AI model.

• API (Application Programming Interface): A controlled channel that allows software (like an LLM) to fetch specific data on demand.

• Robot Exclusion (robots.txt / ai.txt): A file on a website telling crawlers which areas to crawl or avoid. 'ai.txt' is a proposed standard to opt out of AI scraping.

• Schema.org Metadata: A standardized way to annotate web content for search engines (e.g. licensing info, product specs).

• C2PA (Content Provenance & Authenticity): An industry standard that embeds cryptographic metadata (content credentials) in media to record its origin and edits.

• Do Not Train (DNT) Registry: A list of domains or assets that AI developers agree not to use for training.

• Watermark (Visible/Invisible): A marker (logo or embedded data) in an image to assert ownership or trace origins.

• License (AI Context): A legal agreement specifying how AI developers may use content (e.g. "non-commercial", "no AI training", etc.).

• Indemnification: A legal provision where one party compensates another for potential losses (e.g. if AI misuses content).

B. Checklist: Making Brand Media Visible but Not Trainable

1. Asset Inventory: Catalog all media assets (creative, images, text, audio). Note copyright status.

2. Rights Marking: Add clear copyrights and usage notes on each asset and webpage (e.g. “© Brand, All rights including AI training reserved”).

3. Meta & Robots: Implement robots.txt, ai.txt, and in-page metadata (schema.org/ IPTC tags) to signal non-train status.

4. Content Credentials: Embed C2PA content credentials in key images/videos to enable provenance tracking.

5. Watermarks: Decide on visible logos or invisible marks for premium images. Use SynthID or similar for AI-generated brand content.

6. API/Plugin Setup: Prepare data APIs or ChatGPT plugins for real-time content access. Test with Azure OpenAI or GPT tools to ensure data flows correctly.

7. License Agreements: Review/update any third-party licenses to cover AI usage. Negotiate new deals with stock agencies or AI firms as needed (see Part IV).

8. Content Strategy: Produce high-quality earned media (press, thought leadership) optimized for generative search.

9. Team Alignment: Form a cross-team AI task force and educate them on policies. Ensure legal and compliance review for all AI initiatives.

10. Monitoring: Use GEO reports or simple queries to monitor how your brand appears in AI outputs. Adjust as needed.

C. Sample Contract Clauses for AI Licensing Agreements

• License Scope: “Content is licensed for use by [AI Company] only as part of the [specified model/feature]. The license does not grant permission for perpetual inclusion in training data.”

• Data Use Restrictions: “Licensed content may not be copied beyond the AI system’s internal processing. [AI Company] agrees to honor ‘Do Not Train’ designations and robots/ai.txt directives.”

• Attribution & Linking: “Any generative outputs referencing the content must attribute the Brand and, where digital, include a link or citation to the original source.”

• Indemnification: “[AI Company] will defend and indemnify the Brand against third-party claims arising from unauthorized use of the Brand’s licensed content.”

• Data Deletion/Opt-Out: “Upon request, [AI Company] shall remove all Brand content from its active training datasets and cease using it in future model updates.”

• Audit Rights: “Brand shall have the right to audit [AI Company]’s records to verify compliance with content use and deletion.”

Each clause should be tailored with legal counsel to fit the jurisdiction and the specific use case (image vs. text data, etc.). These serve as starting points to ensure brands control how their media is used in AI systems.