How Large Language Models Are Reorganizing Media, Marketing and the Infrastructure of Trade
By Nthanda Manduwi
I’ve spent the past few weeks deeply curious about Apple products, and more specifically Apple TV: the elegance of it all, and also, the business model.
I recently asked my AI how Apple TV makes money. The first answer was reasonable. Apple earns subscription revenue, sells and rents films, distributes third-party channels, sells Apple TV hardware and uses entertainment to strengthen its wider services business. None of this was incorrect. It simply was not the answer I was trying to find.
I kept asking.
The question beneath my question was why Apple continued to finance television at all. What economic job did Apple TV perform inside a company whose principal advantages had been built through devices, operating systems, distribution, payments and a tightly integrated services environment? I was not looking for a list of places where money entered. I wanted to know why the unit existed, which behavior it was designed to produce, what it made more valuable elsewhere and whether it needed to be profitable on its own.
Apple does not report Apple TV as a separate business. It reports a broader Services category that generated $109.2 billion in fiscal 2025, alongside an installed base of devices through which those services can be distributed. Apple TV therefore cannot be understood only as a small Netflix. It also helps populate a bundle, maintain a billing relationship, make Apple hardware more useful and extend the company into another part of its customers’ lives. Whether every show independently earns back its production budget is a different question from whether television contributes to the economics of Apple’s larger system. In both cases, the visible media product is only one layer of the system. The more important question is what function the media performs and where the value it creates is captured.
This distinction has become increasingly important to me because it reflects a change in how I understand my own work. For most of my career, people have told me that I am a very good marketer. I have generally responded that I am a media entrepreneur.
I did not think they were insulting me. I simply thought they were identifying the wrong profession.
Media was how I expressed my creativity.
I wrote, blogged, photographed, produced video, built websites, organized events, developed campaigns and translated ideas across different public formats. I did not feel like only a writer, a blogger, a videographer or a communications professional. No single medium contained what I was doing. “Media entrepreneur” was the broadest available language for a creative practice that moved according to the idea rather than according to a fixed format.
The people calling me a marketer were looking at the work from the outside. They saw what happened after the media existed. I could take a complicated institution, project or idea and create a public frame through which other people could understand it. I could identify what mattered to an audience, find the language that made it legible, position it within a larger conversation and create some form of movement around it. They were observing function. I was describing form.
Both descriptions were accurate. It took me longer to understand why.
My understanding of media was not formed inside an American advertising agency. I grew up in Malawi and worked across public institutions, development, cultural media, digital education and entrepreneurship. At UNDP’s Independent Evaluation Office in New York, including through the Global SDG Synthesis Coalition, communication was inseparable from evidence use. A synthesis that remained technically sound but could not be understood by decision-makers had not completed its institutional function. Media helped evidence travel across departments, organizations and levels of authority. Its value appeared through coordination, learning, accountability, policy influence and the legitimacy of better-informed decisions.
Earlier work in Malawi had placed media alongside public participation, gender and development, culture, technology access and education. Through arts and cultural programming, I saw media create recognition and community. Through digital-development work, I saw information systems affect who could participate in an economy or institution. In these environments, the most important return was often public rather than privately captured.
The move into business development at Microsoft’s ID@Xbox placed cultural products inside another language. Games still contained story, artistry and identity, but they also existed in portfolios. Their trajectories were shaped by distribution rights, platform incentives, regional markets, developer relationships, customer acquisition, engagement data and the commercial logic of the wider gaming business. A game could be excellent and still struggle to find an audience. It could attract a devoted audience and still produce weak economics. It could also perform a strategic function for a platform that was not visible from the game’s direct sales alone.
My Master of Science in Entrepreneurship had already given me a framework for studying digital adoption and information-management systems. The MBA at Michigan State University, particularly the combination of marketing and business analytics, gave me a sharper language for the mechanisms underneath the media I had spent years creating. I began to ask about acquisition cost, conversion, retention, lifetime value, market structure, pricing, distribution and value capture. These were not replacements for questions about meaning or public value. They were additional questions about how a system survived and who benefited from what it produced.
That was why the Apple TV answer did not satisfy me. I had begun refusing to confuse a revenue stream with a business model.
Media and Marketing Are NOT the Same Thing
Media and marketing overlap so frequently that people often use the words as though they describe the same activity. They do not.
Media is the creation and organization of meaning, knowledge, information, culture, entertainment or expression. Distribution is the infrastructure through which that media reaches people. Marketing begins with a market: a group of people, institutions or buyers whose needs, incentives and behaviors can be understood and influenced. Advertising is a paid mechanism for purchasing access to attention or intent. Monetization is the method through which value is captured. A business model connects the entire system of creation, delivery and capture.
Separating these functions does not require treating one as nobler than the others. A poem does not become a product merely because it is published. A public report may have substantial institutional value even if nobody pays to read it. A documentary can alter public understanding while losing money. A campaign can generate sales without contributing much cultural value. A film can do all of these things at once.
The separation matters because “successful” means something different at each level. Media can succeed artistically, culturally, intellectually or politically while failing commercially. Marketing can successfully generate demand for a weak product. A business can own valuable distribution while producing very little media itself. A creator can generate significant commercial demand while capturing almost none of the resulting income.
This was visible in my earlier tourism work. A beautiful destination video might be judged by its cinematography, narrative coherence and emotional force. The tourism system surrounding it eventually needs a different set of answers. Did it increase awareness of the destination? Did viewers search for flights, book rooms, visit an attraction, attend an event or change what they believed about the country? Did the hotel, airline, tour operator, destination agency or creator receive any measurable value from that movement?
The request to think about a vlog “from a marketing perspective” was not necessarily an instruction to make it less creative. It was an instruction to connect creative work to an intended market response.
This does not mean that every piece of media must be designed to sell something. It means that when media is being financed as an enterprise, someone must understand the relationship between the work and the resources required to continue producing it. The relationship may be direct: a reader buys a book, a viewer buys a cinema ticket, or a listener pays for a subscription. It may involve selling access to an audience through sponsorship or advertising. It may be indirect: a podcast creates authority that generates consulting work, a film strengthens a streaming subscription, or a founder story increases trust in a consumer product.
The error is not making media that does not directly earn money. The error is assuming that attention, influence and economic value automatically accrue to the person who created them.
Different Languages of Value
The public sector and private enterprise often describe the same underlying process in different terms. An international organization may speak about dissemination, uptake, influence and institutional learning. A company may speak about reach, engagement, conversion and retention. Both are trying to understand whether an intervention moved through a system and changed something that mattered.
The difference is most visible at the point of capture. Public institutions are generally meant to create value that is widely distributed: better policy, stronger accountability, improved services or more capable institutions. A private company must capture enough of the value it creates to finance operations, reward capital and continue competing. The first model can become vague about sustainability; the second can become excessively narrow about what counts as value.
Working across both environments has made me suspicious of the weaknesses in each language. A communications team can report impressive reach without showing whether anybody understood or used the information. A marketing team can report conversions without asking whether the product produced durable customer value. An organization can create substantial public benefit without building the financial capacity to continue. A company can capture substantial private value while transferring costs to workers, consumers or society.
Systems analysis requires keeping all of these outcomes visible.
It also requires asking who owns the route between them. Media does not move directly from a creator’s mind into a market. It passes through publishers, platforms, broadcasters, retailers, search engines, social networks, app stores, payment systems and recommendation algorithms. Each intermediary can shape what is visible, who encounters it and where the resulting value accumulates.
That is why distribution is never a merely technical layer. It is an economic and political one.
Media as the Infrastructure of Trade
The Audience Was Never the Whole Product
Trade has a persistent coordination problem. Sellers need to find buyers. Buyers need to identify possible sellers, distinguish among their claims and decide whom to trust. Media has repeatedly helped solve that problem by gathering people around information, entertainment, identity, aspiration, fear, desire and community.
Once an identifiable group has gathered, access to that group becomes valuable to somebody trying to sell.
Economists describe many advertising-supported media businesses as two-sided markets. A platform serves at least two groups whose participation affects one another. A newspaper needs readers to attract advertisers; advertiser revenue can then reduce the amount charged to readers. A search engine needs users to attract businesses, and a large inventory of businesses can make the search engine more useful. A social network offers people communication and entertainment while offering advertisers access to those people’s attention. The platform manages not only a total price but the distribution of costs and benefits across the different sides. s therefore more than a banner pasted onto media after the creative work is complete. It is one of the mechanisms through which the cost of media has repeatedly been shifted away from its immediate audience and toward sellers seeking to influence that audience. Economic research on digital advertising similarly treats advertising as a means of financing media goods that consumers can access without paying their full production and distribution cost directly. nt is powerful because it can dramatically expand access. A household does not need to pay the full cost of producing a broadcast schedule before watching television. A user does not need to compensate every engineer, moderator and data-center operator before opening a social application. An advertiser pays because the platform has assembled people in a context where commercial communication may produce a return.
The arrangement is also unstable. The user often experiences advertising as a cost rather than a benefit. More ads may raise platform revenue while making the product less attractive. More intrusive targeting may increase advertiser performance while reducing privacy. A platform may favor engagement that produces inventory rather than engagement that improves users’ lives. The party financing the media may acquire influence over what is produced, emphasized or avoided.
Advertising is neither an accidental contamination of otherwise pure media nor an innocent public subsidy. It is a financing arrangement with its own incentives.
Its importance to the current economy is difficult to overstate. Global advertising revenue crossed $1 trillion in 2025. U.S. internet advertising revenue reached $294.6 billion, almost exactly equal to Alphabet’s $294.7 billion in global advertising revenue for the same year. Meta generated $196.2 billion from advertising, accounting for nearly all its Family of Apps revenue, while Amazon’s advertising-services business reached $68.6 billion. These companies are not peripheral media firms selling occasional sponsorships. They operate some of the largest systems ever built for converting attention and intent into trade. ustry’s repeated return to advertising is therefore not evidence that executives have failed to invent another model. Consumers regularly demonstrate that they value access but resist paying the full price of every piece of digital media they consume. Advertising supplies a third payer.
Subscription services have not eliminated that bargain. They have increasingly reproduced it in hybrid form. Deloitte’s 2026 U.S. media survey found that 68% of streaming subscribers were paying for an ad-supported plan, a rise of more than 20 percentage points from 2024. Consumers did not suddenly discover a love of interruptions. They were making a price trade-off. e-off is moving into games.
Breakfast and Games
On the morning of the 29th of July, 2026, my friend Neil asked me if I was up for grabbing an early breakfast, which lucky for him, I was. Neil was one of my first friends in Detroit. I knew [of] him through my work with DAP at Xbox, as he happens to create games as Aerial_Knight.
He told me, biting into his french toast, that he had received a message from Xbox informing him about a strategy involving advertising around games. His first reaction was that he did not want ads in his games.
I understood the reaction immediately. As [media] creators, we think about rhythm, atmosphere and the integrity of the experience. Advertising enters from another logic. It risks treating the work as a container whose empty spaces can be sold.
Neil’s position became more complicated as he kept talking. He wanted people to play his games. He also wanted to be paid for making them. A player might prefer an uninterrupted experience but still decline to purchase the game or pay for another subscription. If the same player accepted a short advertisement in exchange for access, advertising could become the economic bridge between the desire to play and the creator’s need for compensation.
He did not become enthusiastic about ads. He recognized the contradiction.
The timing made the conversation especially useful. Xbox publicly began testing ad-supported game streaming on July 23, 2026. Under the test, Xbox Insiders could stream selected games they already owned without an additional streaming purchase, with advertisements appearing before a session rather than interrupting gameplay. Xbox described advertising as a way to lower access costs while insisting that the core game experience should remain unchanged. aming and Neil’s private ambivalence described the same economic triangle. The creator wants compensation. The player wants access. The advertiser pays for the opportunity to address the player.
There are many ways for this bargain to become unfair. The platform may make the decision without meaningful creator participation. The creator may receive little of the advertising revenue. Ads may become more frequent after users have been trained to accept them. The advertised product may conflict with the creator’s values. The player may technically have a choice while every practical alternative becomes more expensive. Data collected for ad targeting may exceed what the exchange reasonably requires.
Yet refusing to examine the bargain does not protect creators. It merely leaves its design to platforms.
The relevant question is not whether advertisements are aesthetically desirable. It is what financing structure allows a work to exist, who chooses that structure, how the revenue is divided and what it does to the experience.
Creators frequently focus on the first production problem: how to make the work. The platform economy imposes a second: how the work enters a system in which access, discovery and payment are controlled by other parties. The person who made the game may not control the store, subscription, advertising inventory, recommendation system or customer relationship.
That problem extends well beyond games.
U.S. creator advertising expenditure was projected to reach $37 billion in 2025 and $44 billion in 2026, according to the Interactive Advertising Bureau. Brands increasingly treat creators as a formal media channel rather than an experimental addition to social campaigns. The economic value of creator attention is no longer in doubt. The unsettled issue is how much of that value creators retain after platforms, agencies, technology providers and intermediaries take their shares. iscomfort with advertising can therefore coexist with a more mature demand: if the work is generating advertising inventory, the creator should understand and participate in the economics around it.
From Attention to Intent
The major digital advertising companies did not become powerful by gathering the same kind of information.
Google organized explicit queries. A person searching for a plumber, graduate degree or pair of shoes was disclosing a need in language that could be matched with a seller.
Meta organized social and behavioral signals. Likes, follows, relationships, viewing time and repeated interactions helped infer identity, interest and aspiration even when a user had not articulated a purchase.
Amazon organized commercial behavior. It could observe searches, product pages, reviews, baskets, purchases, returns and price sensitivity within the same retail environment.
YouTube organized sustained topical attention. A person who repeatedly watched reviews, tutorials or travel videos revealed interest over time.
Large language models can combine elements of each system while adding something different: articulated intent inside a continuing conversation.
The difference became visible to me at Newlab in Detroit through an interaction so ordinary that it would have been easy to miss.
My videographer and creative collaborator, Donn, was hungry. He was also trying to gain mass and wanted a high-calorie meal. He mentioned a fast-food option, left to collect it and did not present the choice as technologically interesting. After he left, I reached toward his laptop because I wanted to borrow the charger. Gemini was open on the screen. He had asked it what he should eat to support his goal.
Donn had not begun with a restaurant. He had begun with an objective.
He gave the model a problem involving his body, appetite and nutritional preference. The model translated the problem into a commercially actionable suggestion. He then left Newlab and bought the food.
I tend to use AI differently. I use it to synthesize, interrogate, structure, research and test. I still discover products through the habits I developed on the open web. I search. I use Pinterest. I read Amazon reviews. I navigate websites and compare sources. I do not ordinarily ask an LLM to make every consumer decision on my behalf.
Donn used it as the front door to a transaction.
One anecdote cannot sustain a sweeping generational claim. Pew’s evidence is more cautious and more useful: 64% of U.S. teenagers reported using AI chatbots in a 2025 survey, with 57% using them to search for information. Among U.S. adults, 49% reported using chatbots by February 2026, including 24% who used them daily. Search remained one of the most common uses. The evidence shows broad adoption, not the disappearance of older discovery methods. e still matters because it reveals what adoption looks like when it becomes normal. He did not think he was replacing Google. He did not describe himself as an early adopter of conversational commerce. He had a need and used the interface closest to him.
For marketers, the commercially significant difference lies in the information available before the recommendation. A search query might say “high-calorie fast food.” A conversation can reveal that the user is skinny, trying to gain mass, currently hungry, located near a specific neighborhood, unwilling to cook and ready to purchase immediately. It can incorporate previous preferences, dietary constraints, budget and rejected options.
This does not mean every conversation should become advertising inventory. Many conversations are personal, sensitive or entirely unrelated to commerce. It means that when commercial intent is present, the model can understand it with greater context than a traditional keyword auction.
The platform is no longer only selling access to attention. It is approaching the moment when a person’s objective becomes a decision.
LLMs as the New Front End
The [World Wide] Web vs the [Large Language] Model
The World Wide Web trained my generation to navigate.
We learned to break a problem into searches, inspect a list of links, open multiple tabs, recognize the visual grammar of a website, compare claims and move between publishers, marketplaces and reviews. That process was never neutral. Search engines ranked the pages, advertisers purchased prominent positions and interface design influenced what we clicked. Yet the user still experienced the web as a collection of destinations.
Large language models reorganize the experience around an answer.
The conventional sequence is familiar:
A need becomes a query. The query produces a ranked page. The user visits websites, gathers evidence, compares options and eventually transacts.
The conversational sequence can be shorter:
A need becomes a description. The model interprets it, asks questions, formulates criteria, synthesizes information, narrows the options and presents a route to action.
The websites do not vanish. Their functions move behind the interface. Product pages supply prices and specifications. publishers supply analysis. Review sites supply consumer experience. Databases supply structured facts. Payment providers complete the transaction. The user may encounter all of this through a single conversational layer.
That is what I mean by the new front end.
In software, the front end is the part of a system with which the user directly interacts. The back end contains the databases, services, logic and infrastructure required to produce the experience. The web has traditionally exposed many front ends: each publisher, retailer, institution and platform built its own interface.
An LLM can place a new interface over all of them.
This does not make the underlying web less important. It can make the web more structurally important while reducing its visibility. Businesses may invest in websites that fewer people consciously browse because those websites provide the structured information, authority and transactional infrastructure that models require.
The transition is already affecting outbound traffic. Pew’s analysis of 68,879 Google searches found that users clicked a conventional search result in 8% of visits when an AI summary appeared, compared with 15% when it did not. They clicked a source cited inside the AI summary in only 1% of visits. They were also more likely to end the browsing session after encountering a summary. h has found similar substitution effects. A 2026 study using matched Wikipedia pages estimated that exposure to Google AI Overviews reduced daily traffic to English-language articles by approximately 15%, although effects differed by subject. Another large-scale measurement study found that AI Overviews were especially likely to appear for question-form searches and that 11% of the claims it examined were not supported by the cited pages. These are early studies of fast-changing products, but they show why publishers are concerned: the answer layer can retain the user while drawing upon material produced elsewhere. in between websites and search engines was imperfect but intelligible. Publishers allowed crawling and indexing because search could send human visitors. Those visitors could see advertisements, subscribe, buy products or deepen a direct relationship with the publisher. AI synthesis changes the exchange when a platform can crawl information, answer the question itself and send fewer people onward.
The front end is therefore not a decorative interface. It determines where attention stops.
The Commercial Decision Layer Is Already Here
It would be easy to frame conversational advertising as a future possibility. By July 2026, that framing is already outdated.
OpenAI began testing ads in ChatGPT in the United States on February 9, 2026 and subsequently expanded the pilot. Ads may appear below responses for eligible Free and Go users. OpenAI says the organic answer remains independent, while a separate advertising system considers the current conversation’s context and intent, ad content, advertiser-supplied hints and, where personalization is enabled, selected signals from a user’s broader ChatGPT experience. vague experiment involving a few hand-selected sponsorships. ChatGPT Ads supports CPM and CPC buying, relevance-weighted auctions, advertiser bids, conversion pixels and a Conversions API. Advertisers can supply descriptions of the conversations, topics or problems in which their offerings may be relevant. OpenAI’s advertiser documentation recommends initial maximum CPC bids of $3 to $5 and explicitly describes the product as a way to reach people while they “explore, compare, and decide.” reveals the asset being commercialized. The advertiser is not purchasing a generic banner at the edge of a website. It is purchasing possible adjacency to a decision process.
ChatGPT’s shopping interface already shows product recommendations, prices, review summaries and merchant options. For eligible products and merchants, users can complete checkout without leaving ChatGPT. OpenAI describes those organic product results as independent from ads, while acknowledging that product selection depends on structured metadata, availability, price, reviews, user context and information from third parties. from a different position. It already owns one of the most sophisticated advertising and commercial-discovery systems in history. In 2025, Google Search and related properties generated $224.5 billion in revenue, within a broader advertising business of $294.7 billion. Gemini does not need to invent auctions, advertiser relationships, merchant feeds, local business data or conversion measurement. Google needs to adapt that infrastructure to a more conversational interface. ppear in AI Overviews and AI Mode. Google has tested formats in which Gemini generates an independent explainer beside sponsored creative, allowing an advertisement to answer why a product may fit a user’s situation. It is also building Direct Offers, branded business agents and checkout into AI Mode and the Gemini app. Merchant Center is adding attributes intended specifically for conversational discovery, including answers to common product questions and information about compatible products or substitutes. l an important distinction. Google is extensively monetizing AI-mediated Search, but its official materials do not establish that every standalone Gemini conversation has become a conventional advertising surface. Google’s advantage is broader than putting ads directly into Gemini. Its AI products are being integrated with an existing discovery, advertising, mapping, merchant and payment system.
Google is bringing AI into an advertising empire.
OpenAI is bringing advertising into an AI relationship.
Microsoft has already combined both paths. Ads in Copilot have operated across multiple markets, with formats designed around conversational context. Microsoft has also introduced Copilot Checkout, through which users can compare products and transact inside the conversation while the retailer remains the merchant of record. Microsoft reports that shopping journeys involving Copilot were more likely to result in purchases, although these performance figures come from Microsoft’s own first-party analysis and should be treated accordingly. even closer to the transaction. Its AI shopping assistant, formerly Rufus and now presented as Alexa for Shopping, can interpret product questions, remember preferences, compare options and take actions. Amazon introduced Sponsored Products and Brand Prompts into the assistant in 2026. Brand Prompts operate as conversational product experts, allowing the user to continue asking questions about a sponsored brand. Amazon reported that nearly 20% of shoppers who interacted with one of these prompts continued the brand conversation. Again, this is a company-reported performance metric, but it demonstrates how advertising is moving from display toward dialogue. s developed an answer-engine commerce model with product cards and Instant Buy. Its current policy says organic product listings cannot be purchased and are ranked through relevance, ratings and product information. The system nevertheless brings recommendation, comparison and purchase into a single interface, and Perplexity has separately experimented with sponsored questions. t path is less about inserting an ad beneath every Meta AI answer and more about using conversational interaction as a signal across an existing advertising system. Meta says interactions with its AI products can inform the content and advertisements shown across Facebook and Instagram, subject to restrictions on the use of sensitive topics. A discussion with Meta AI about hiking can therefore contribute to seeing hiking-related groups, posts and advertisements elsewhere in Meta’s applications. es are approaching the same commercial layer from different starting points.
OpenAI begins with a trusted conversational product and adds advertising and commerce.
Google begins with search, merchant infrastructure and advertising, then makes the search journey conversational.
Microsoft combines search advertising, enterprise relationships and an AI assistant.
Amazon begins with retail transactions and adds a conversational adviser.
Meta begins with identity, social graphs and advertising, then incorporates AI conversations as another signal.
Perplexity begins with answer-oriented search and adds purchasing.
The common destination is an interface capable of understanding a need, narrowing the market and participating in the transaction.
Paid Visibility Follows the Organic Answer
The most interesting feature of LLM advertising is not that a sponsored card appears below some text. It is what the text has already accomplished before the card arrives.
A person may begin by describing hair breakage, protective styling, scalp dryness and a reluctance to use heavy products. The model can explain possible causes, separate scalp care from hair care, identify the characteristics of an appropriate routine and help the person understand what to avoid.
By the time an advertisement appears, the model may already have created the category, clarified the criteria and increased readiness to purchase.
The organic answer educates. The paid placement offers an action.
OpenAI says the two systems are separate and that advertisers cannot shape or rank the answer. Its advertising rules also exclude sensitive contexts and restrict categories such as political advertising, gambling, tobacco and many health claims. Those safeguards are meaningful. They recognize that a conversational interface is not equivalent to a search-results page, particularly when users discuss emotional, medical or financially vulnerable situations. n does not remove every commercial question. A platform decides which conversations are eligible for advertising, where ads appear, what product categories are supported, which outcomes its auction optimizes and how much visual attention is given to paid material. It decides whether the user remains inside its interface or visits the advertiser. It can alter these choices over time.
There is also a distinction between answer independence and product independence. Even when no advertiser directly alters a response, the platform’s broader business model can influence product priorities. An interface designed to maximize subscriptions may remain deliberately ad-light. An interface seeking performance-advertising revenue may expand the number of commercially interpretable conversations, invest heavily in attribution and make shopping actions more prominent.
Economic incentives do not need to rewrite an individual sentence to shape a system.
Publishers confront a related problem. Their reporting, reviews and analysis may contribute to an AI-generated answer. The platform can then place its own advertising beside that synthesis while the publisher receives neither the visit nor the advertising impression. One 2026 study found that more than half of pages cited by Google AI Overviews carried display advertising, meaning that the answer interface could reduce the publisher’s opportunity to monetize even as Google continued to show sponsored placements on the search page. the old media bargain. The publisher once created information, the search engine helped the user find it, and the publisher had an opportunity to convert the visit into revenue or loyalty. In the new structure, the model can become the publisher, distributor, adviser and advertising platform simultaneously.
For the user, this may be wonderfully efficient.
For the original producer of the information, it can be economically devastating.
For the platform, it is an opportunity to capture more of the journey.
The argument against this model cannot simply be that advertising is impure. Advertising has long financed media that audiences would not fully fund themselves. The more demanding question is whether the platform is compensating the parties whose information makes the answer useful, maintaining a credible boundary between evidence and sponsorship, and allowing users to understand why particular options were shown.
Epistemic integrity and commercial sustainability have to be designed together. Neither will appear automatically.
What Marketers and Creators Can Do
From SEO to Machine Legibility
The arrival of LLM interfaces has produced a new marketing vocabulary: Generative Engine Optimization, Answer Engine Optimization, LLM optimization and AI-search optimization. Some of this terminology is useful. Some of it is an industry attempting to sell certainty before the evidence has stabilized.
Traditional SEO is not disappearing. Search engines still crawl and rank pages. LLM products frequently rely on web search, retrieval systems, product feeds and established authority signals. A company with a technically poor website, contradictory product information and no independent references will not become more discoverable merely because it adds a page titled “Best answer for ChatGPT.”
The larger requirement is machine legibility.
A business must make it possible for systems to determine what the company is, what it sells, where it operates, how its products differ, whether its claims are credible and whether its information remains current. This requires clear product descriptions, structured data, prices, sizes, ingredients, specifications, availability, policies, FAQs, reviews and corroboration beyond the company’s own pages.
Google’s new Merchant Center attributes make the direction explicit. Retailers are being encouraged to supply information that exceeds conventional keywords, including answers to expected questions, product compatibility and substitute relationships. Perplexity says merchants providing deeper details about availability, pricing, reviews and specifications are more likely to appear in relevant product results. ChatGPT’s shopping documentation similarly identifies structured metadata, price, reviews and merchant information as inputs into product selection. ft from writing only for a ranking page toward maintaining a trustworthy public data environment.
I recognized an early version of this problem through my own public identity. I wanted to correct and strengthen what search engines and AI systems could find about me, so I worked on the information environment, including Wikipedia.
It would be inaccurate to claim that editing Wikipedia allowed me to control what an LLM said. No single page can do that. Models and retrieval systems may use search indexes, licensed sources, knowledge graphs, public databases, recent articles and other material. They can also make mistakes despite excellent source information.
What I could do was reduce inconsistency. A coherent public record makes it easier for both humans and machines to distinguish the actual person from incomplete, outdated or conflated information.
For brands, the same principle applies. Machine legibility is not manipulation of a model. It is the disciplined publication of information that can be retrieved, verified and reconciled.
The early GEO research suggests that authoritative third-party coverage may matter greatly, but the evidence remains unstable across platforms, languages and repeated queries. A 2026 review of 45 studies concluded that no examined technique had yet demonstrated stable, longitudinal, cross-platform control over organic discoverability or downstream behavior. The most defensible strategy is therefore not a magical formatting trick. It is clear first-party information combined with earned credibility, technical accessibility and repeated measurement. uld also resist assuming that visibility means citation. A brand can influence the substance of an answer without being named. It can be named without receiving a click. It can receive a click without producing a sale. It can produce a sale while losing the direct customer relationship to the platform.
The relevant measurement system will need to distinguish among discoverability, mention, citation, referral, conversion and retention.
That complexity is another reason not to abandon conventional marketing infrastructure. Businesses still need direct websites, email lists, customer data, reliable analytics, retailer relationships and recognizable brands. The objective is not to replace every channel with an LLM strategy. It is to ensure that the business remains intelligible when an LLM mediates the customer’s route to it.
From Keyword Ownership to Problem Ownership
Traditional search advertising often begins with a keyword. The advertiser attempts to predict which terms a customer will type and how much the resulting click is worth.
Conversational advertising begins with a more complete problem.
For Effortless Natural, a conventional campaign might target “natural hair products,” “wigs for Black women” or “repair shampoo.” These phrases matter, but they contain little of the context that determines whether a product is actually suitable.
A customer may instead explain that she wears wigs most days, uses heat to blend her leave-out, has noticed breakage around the front of her hair, dislikes heavy oils and wants a routine she can complete quickly before work.
The opportunity is not merely to bid on “hair breakage.” It is to understand the system surrounding the problem: protective styling, texture matching, scalp access, product buildup, heat exposure, time constraints and the emotional desire to change one’s look without sacrificing the hair underneath.
That is problem ownership.
It requires a different kind of marketing content. Product pages must answer questions customers actually ask. Comparison pages must explain where a product is appropriate and where it is not. Ingredient claims must be specific enough to verify. A brand must publish enough information for the model to distinguish it from hundreds of superficially similar options.
Conversational systems may reward companies that understand customer language at a deeper level than a keyword list. They may also reward companies that can supply structured, current and independently supported evidence for their claims.
Paid advertising will become one route into these conversations. It should not be the only one.
A sensible marketing portfolio still includes technically sound websites, conventional search optimization, creator relationships, reviews, customer communities, email, retail distribution and direct brand building. Emerging LLM advertising should be tested against measurable questions: Which types of conversations produce qualified traffic? Do users arriving from an AI interface understand the product better? Does conversational context improve conversion? Does the platform return enough data to learn without compromising customer privacy?
No company should hand its entire acquisition strategy to an interface whose auction, policies and organic recommendation system it does not control.
This is especially important for smaller firms. OpenAI’s advertising auction combines relevance with bids. Google’s systems combine extensive advertiser data, merchant feeds and campaign history. Large companies will enter with more content, more reviews, more technical integrations and more money.
A smaller brand’s advantage has to come from specificity, authority and a sharper understanding of the customer problem. Even then, platform design will determine how much that advantage matters.
The Creator Must Follow the Money
My changing view of media does not lead me to believe that every video should carry an advertisement, every podcast should be interrupted by a sponsor or every essay should end in a sale.
It leads me to ask what economic activity the media enables.
A podcast can sell books without charging the listener. A book can establish intellectual authority that produces speaking engagements or institutional partnerships. A founder story can increase trust in a consumer brand. A technology demonstration can support enterprise contracts. A philanthropic documentary can attract resources and participation toward a public mission.
The revenue does not need to sit beside the media. The relationship should still be understood.
My own work increasingly makes sense to me as a portfolio of media functions. Essays and podcasts create public reasoning around systems, power and development. Books turn that reasoning into intellectual property I own. Q2 translates systems analysis into technology, simulation, robotics and potential institutional contracts. Effortless Natural connects cultural understanding, product design and commerce. My philanthropic institutions use communication to mobilize legitimacy, resources and participation around social outcomes.
These are not all the same business. Media moves among them because media is part of how I build public meaning around each system.
The danger is assuming that because the portfolio produces attention, value will naturally return to it. Platforms may monetize the podcast. Retailers may monetize product discovery. Event organizers may monetize the authority of a speaker. Technology companies may use public writing to improve models. A destination may benefit from tourism media without compensating the creator whose audience generated the demand.
Creators need to trace the chain.
Who owns the audience account? Who controls distribution? Who processes payment? Who receives the customer data? Who can change the rules? Who earns if the media succeeds? Who loses if the platform changes its ranking system?
The commercial discipline is not “put ads on everything.” It is to design the value-capture architecture as intentionally as the creative architecture.
This may involve direct sales, subscriptions, sponsorship, licensing, intellectual property, products, events, consulting, education or institutional partnerships. Advertising is one option. It becomes particularly important when audiences resist paying and a third party is willing to subsidize access.
Neil’s breakfast problem is therefore not confined to games. It is the creator economy’s central negotiation. The creator wants broad cultural participation without giving away the economic basis of future work.
The Global South Cannot Remain an Invisible Back End
You know I will ALWAYS include a global south angle!
The transition to LLM-mediated discovery is often discussed as a competition among American technology companies. From an African perspective, another question appears immediately: which people, businesses and institutions will be legible to the systems they build?
A small business can be excellent in the physical world and nearly invisible in the digital one. It may operate through WhatsApp, Facebook, phone calls and informal customer networks. Its website may be outdated or nonexistent. Prices may not be published. Product information may be embedded in images. Reviews may live inside private conversations. The company may have little coverage from publications that international models treat as authoritative.
The International Finance Corporation estimates that more than 600,000 formal African firms and 40 million microbusinesses could benefit from greater digitalization. Although 86% of surveyed African firms had access to basic digital tools such as mobile phones and the internet, only a small fraction used more sophisticated technologies intensively, with underuse particularly pronounced among smaller firms. nal local market, a business can remain discoverable through geography, reputation and personal networks. In an LLM-mediated global market, invisibility can become computational.
The model cannot recommend what it cannot identify. It cannot compare a price that has not been published, verify a business that has no consistent record or interpret a product whose attributes are trapped inside an unlabelled image. A company may not merely rank lower. It may fail to enter the consideration set at all.
This creates both risk and opportunity.
A small African firm that develops strong digital records, structured catalogues, credible reviews and reliable fulfillment could become accessible to customers far beyond its immediate geography. Conversational interfaces may reduce the consumer’s need to know the right marketplace, terminology or local search method. The model can translate an unfamiliar business into a relevant answer.
The same system can intensify concentration. Large brands possess more data, more third-party coverage, more reviews, larger advertising budgets and direct platform relationships. If AI systems interpret digital abundance as authority, firms from under-documented markets will begin at a disadvantage.
The digital divide will therefore concern more than connectivity. It will concern representational capacity: the ability to describe a business in formats machines can retrieve, the institutional credibility to have those descriptions corroborated and the commercial infrastructure to fulfill a transaction once discovered.
This is why machine legibility should not be treated solely as a new marketing technique. It is becoming part of market access.
Governments, development institutions and business-support organizations will need to think beyond teaching entrepreneurs how to create social posts. Firms may need support with product information systems, digital identity, interoperable payments, merchant feeds, logistics, verification and consumer protection. These are the back-end conditions required to participate in a front-end economy.
The political question is who establishes those standards. If a few platforms determine what information counts as credible, which merchants can integrate and which regions are economically attractive enough to support, they acquire influence over the geography of digital trade.
An African business may become visible to the world through an American interface while remaining unable to influence the rules governing that visibility.
The Front End Is Power
The front end has always contained power because it organizes the user’s field of vision.
A supermarket shelf influences which products are noticed. A newspaper front page influences which events appear important. A search engine results page determines which sources seem authoritative. A social feed selects which relationships, emotions and products receive attention.
A conversational interface goes further because it can define the problem and the answer space in the same interaction.
The user may not know which products were excluded, which sources were not retrieved, how evidence was weighted or whether a different phrasing would have produced a different recommendation. The model’s fluency can make a contingent selection appear complete.
This matters commercially. A seller outside the answer may never be considered. A publisher outside the citations may receive no traffic. A creator whose work informs the model may receive no attribution. An advertiser may purchase entry into the user’s attention after the model has already shaped the buying criteria.
It matters politically because the same interface increasingly mediates questions about health, education, work, government, identity and public affairs. OpenAI’s decision to exclude ads from vulnerable and politically sensitive contexts reflects an understanding that not every form of intent should be monetized. Other platforms will make different choices, and all of them will face pressure to expand the inventory from which revenue can be earned. mpetitively because the strongest firms in this transition already control adjacent infrastructure. Google controls search, advertising, maps, browsers, video, merchant systems and payments. Amazon controls retail data, fulfillment, advertising and a marketplace. Microsoft controls software, enterprise distribution, search advertising and payment partnerships. Meta controls social graphs and one of the world’s largest advertising systems. OpenAI controls a conversational relationship used by hundreds of millions of people and is rapidly building the commercial systems around it.
The user experiences convenience. The platform accumulates scope.
The strongest version of the “new front end” thesis is therefore not that Google Search will disappear, or that every consumer will ask an LLM what to buy. It is that a growing share of digital activity will be mediated by systems capable of interpreting intention before retrieving information, and increasingly capable of carrying that intention through to a transaction.
The shift changes the unit of marketing.
The old unit was often the audience impression, demographic segment or keyword.
The new unit is the situated problem.
It also changes the creator’s unit of analysis.
The relevant question is no longer only how many people saw the work. It is what the work caused, which market activity formed around it and whether the creator participated in the resulting value.
This brings me back to the apparent disagreement that followed me through my career.
I was right to call myself a media entrepreneur because I experienced the work as creation across forms. Media allowed me to think publicly, connect institutions with people and move between culture, evidence, technology and enterprise.
The people calling me a marketer were also right. They recognized that my work did not simply express. It framed, positioned, distributed and moved.
My business education and private-sector experience did not force me to choose between those identities. They allowed me to see the system connecting them.
Media produces and organizes meaning. Distribution determines who encounters it. Marketing connects that meaning to a market and an intended response. Advertising allows one party to purchase access to the attention or intent gathered by another. Commerce converts some of the resulting action into exchange. The business model determines who captures enough value to continue.
Large language models are beginning to mediate all of these layers at once. They can interpret the need, retrieve the evidence, narrow the choices, place the advertisement and complete the purchase. The websites, publishers, merchants and databases behind the interaction will remain indispensable, but they may increasingly function as infrastructure behind somebody else’s interface.
That is the opportunity marketers are beginning to see. It is also the problem creators, publishers, smaller businesses and governments will have to solve.
The front end is where the user enters the system. It is also where the system decides what the user can see. As LLMs become a principal interface between intention and trade, the central economic contest will concern more than whose model gives the best answer. It will concern who controls the passage from a question to a market, whose knowledge makes that passage possible and who is paid when the answer becomes a transaction.
Read my Published Works:
If you’d like to go deeper into my journey — from Malawi, through the United Nations and Microsoft to now building my own companies in Detroit, you can find it in my books.
Read the Lessons Book Series:
Read my Previously Published Works:
CONNECT WITH NTHANDA ONLINE:
About the Author
