I keep coming back to a question Smalk's new DSP makes unavoidable: when a brand buys a native placement on a publisher page an AI system may read before answering a product question, what exactly is it buying: attention, authority, or an attempted shortcut into judgment?

Smalk describes a media product built for this precise transaction. Its platform identifies publisher pages it says conversational systems use when answering product questions, then sells native text placements on those pages. Marketers can target prompts, set budgets, approve inventory, and track changes in model mentions. The advertisement does not appear inside the chatbot. It sits upstream, in material the system may retrieve before it responds.

The ad shelf is now a source context, not a screen placement. Dismissing this as SEO with another acronym misses the commercial invention. Treating a bought placement as a guaranteed recommendation accepts the pitch too literally. The channel is real as a media proposition. Its causal promise remains unsettled.

The ad has moved upstream

Display advertising rents pixels in front of a person. Paid search rents position beside a query. Retail media rents prominence inside a retailer's environment. Source-layer advertising pays to place brand information inside a context that may contribute to an AI-generated answer. It buys eligibility for machine-mediated consideration.

That creates meaningful new work for marketers. Campaigns can be organized around conversational prompts rather than keyword fragments. Native copy can carry specific product facts and category narratives into the information environments surrounding those prompts. The creative task shifts from interrupting attention toward surviving retrieval and synthesis. A person remains the eventual customer, but a machine may decide which material reaches that person first.

Publishers could gain a new class of inventory, too. If their pages become inputs to product recommendations, their authority has media value even when the conversational interface absorbs the click. Source-layer advertising could let publisher economics follow machine readership, provided the placements are actually available to and used by the relevant systems.

A paid page is not a recommendation

Four distinct events sit between the media budget and the sale:

  1. Presence. The brand appears in a paid native placement on a relevant publisher page.
  2. Retrieval. A model, search layer, or connected system finds and processes that page for a particular prompt.
  3. Citation. The resulting answer names the page, the brand, or both.
  4. Recommendation or purchase. The answer favors the product, and the customer may act. A recommendation and a purchase remain separate decisions even at this final stage.

Smalk can directly arrange the first event. It cannot directly arrange the model's subsequent judgments. Crawler access, source weighting, query wording, freshness, competing information, and model behavior all sit between placement and output. A retrieved page may go uncited. A cited page may leave the advertiser unmentioned. A named brand may still lose the recommendation.

The GEO-GEA offering launched by iProspect and Smalk matters because it moves the idea into an agency service layer. Media planners can now package the channel, assign budget, and place it alongside other forms of discovery spending. That signals category formation, not proven efficacy.

Smalk is best read as a market signal rather than a verdict. New inventory is forming because marketers believe conversational answers will influence product consideration. Whether a particular placement changes a particular answer still has to be learned.

Search rank is giving way to recommendation share

Search trained marketers to compete for ranked visibility. Conversational AI compresses the result set into a synthesized answer, often with a short list and an implied judgment. Product facts, reviews, publisher framing, and category language are pulled into the interface. A brand now competes over how it is represented inside the answer, not merely whether its link appears nearby.

Marketing discussion is already moving from rank toward a broader citation footprint. Share of recommendation is more demanding. Inclusion and interpretation happen together. A brand can be cited but presented as the wrong fit. It can be recommended for one narrow use case while losing the category. Presence alone says little about the narrative the customer receives.

The Adaptive Storefront examined what happens after an assistant has already shaped customer intent. Who Owns the Customer Context When Agents Can Orchestrate the Campaign? considered the context inside delegated marketing systems. Smalk adds the layer before both: the public source environment from which an AI system forms its initial view of a product.

That environment is contested and only partly controllable. A paid placement enters alongside product pages, reviews, editorial coverage, community discussion, and other brands' claims. The strategic asset is therefore larger than one mention. It is a coherent product account that a model can understand, compare, and carry forward without distorting it.

The creative brief now has to survive extraction

An AI-readable placement must remain intelligible when the page design falls away. A banner can lean on imagery, motion, familiarity, and mood. Source text is more likely to be processed as a set of claims. Specificity becomes a creative advantage: what the product is, whom it serves, where it differs, and under which conditions the claim holds.

The strongest source-layer placement may resemble a compact product case more than a conventional advertisement. Its narrative still matters, but the narrative must travel through compression. A model may compare it with conflicting information, omit the slogan, preserve one attribute, and reframe the product around the user's question. Paid copy can enter consideration. It cannot compel synthesis.

Brands can engineer presence. They still have to earn interpretive weight.

What the marketer actually controls

There is a clean dichotomy of control here. The marketer controls whether to test the channel, which source contexts to enter, which prompt territories matter, and whether the product facts are accurate, current, and clearly expressed. The marketer can also align a paid placement with the broader narrative carried by product pages, publishers, reviewers, and customer experience.

The marketer does not control whether a system crawls the page today, retrieves it for a specific question, cites the brand, recommends the product, or converts the customer. Those outcomes belong to a chain of machine and customer judgments that no DSP can fully own.

My read is that the first budgets here should be treated as media research and development, not guaranteed recommendation inventory. Early advantage may come from learning which source contexts shape the brand's machine-mediated account and which product claims remain useful after synthesis.

The discipline is simple: move quickly, but name the result accurately. Until evidence shows otherwise, a purchased placement is presence, not a recommendation. That distinction should sharpen experimentation rather than slow it.

The AI-readable shelf is open. Brands can now pay to stand where models may look, and publishers may be able to monetize the role their pages play in machine-mediated discovery. That is a genuine new capability. But presence is rented; authority has to survive retrieval, synthesis, and comparison. So what is the brand building: a reputation a machine can use, or a citation it may discount?

Sources and further reading