Dialogue shelf: brands must rewrite qualifications recommended by AI

01 Facts & context

UnBoxed 2026 released something.

2026Year9Month30On Sunday, the Amazon ad was held in San Francisco.2026Publication will be systematized into eleven points. At the meeting, Paul Kotas, Senior Vice-President of the Amazon Advertising, stated that the company wished to use AI to provide better advertising and results and to make the whole process as simple as possible. The first day revolved around products, tools and underlying intelligence, and the next day turned to the reimagina series to discuss what happens when transformative technologies meet people ' s original thinking.

Specific publications include: Branded Navigations, which allows brands to provide differentiated information in Alexa for Shopping ' s shopping dialogues; For the first time, Sponsored Services extended the closed ring to local service providers, with Yelp as anchor partner for home and local services; Amazon Ads Agent merged the former DSP with the Advertising Console into a single purchase entrance and introduced an extended closed test of the dialogue-based AI experience; Full-Funnel Campaigns are organized by AI to deal with channels, creativity, audience and optimization; DVA+ combines presentation, video and audio into one goal, one budget, one advertising campaign; Generating AI creative tools embedded in dialogue experiences; Nearly thousands of indicators can be accessed in natural languages; Amazon Ads MCP Server (Lite) and AMC Supert Tool are available to smart people. Several effects figures were given for the testing period and Martha Stewart was invited to share his judgment and true story with Chaunté Lowe.

02 GSP analysis

From purchase flow to purchase of quoted qualifications

The real change in this round is not the delivery interface but the entry structure of brands into consumer decision-making. (b) The competition unit of the traditional search advertisement is the keyword and the bid, and the brand buys a place on the shelf; In the context described by Branded Navigations, consumers use multiple rounds of dialogue to complete research, comparison and purchase, and brand purchases are qualifications cited, explained and recommended by AI in the dialogue. Positions can be bought by the budget, but eligibility depends on whether brands provide a structured and differentiated fact that can be called.

This gives rise to a counter-intuitive trade-off: brands used to tend to concentrate their budgets on the few key words that transform them to the strongest, leaving brand narratives out of hand; In the dialogue portal, the lack of a brand with clear positioning and product knowledge is likely to be excluded at the consumer question-and-answer stage, and subsequent high bids will not buy back that dialogue. Another mechanism that is often overlooked is the cumulative nature of the dialogue: during multiple rounds of question-and-answer rounds, AI constantly narrows its recommendations based on information already available, and early unincorporated brands are difficult to recall in subsequent rounds. This means that brands need to complete the supply of knowledge prior to the start of the dialogue, rather than relying on intermediate bid interventions.

One hypothetical example is that when a cleaning brand uses high-transformation terms such as “controlled oil-washed milk” and never combines information such as “introduction” with “virtual folding” information into machine-readable assets, the probability that it will be cited will be systematically depressed when consumers ask “to help me set up a routine process”. This is not about creativity, but about the supply of knowledge. The test could be to regularly raise high-frequency issues in this type of product at the dialogue portal, to document whether brands are cited and what facts are cited, and to supplement the knowledge base accordingly. It is suggested that brands should be organized into a branded knowledge base that can be used by smart people and that they should be governed as media assets rather than as market literature.

03 GSP analysis

Value mechanism: closed loop extension and budget redistribution

Looking at Sponsored Services together with Amazon Ads Agent, the Amazon is doing two things: one to extend the closed link from the goods to service performance and the other to reduce the decentralized operation to a single entrance. The value mechanism for service-type advertising is to fill “buyed but unusable” breakpoints — smart taps, home-placed transactions such as this — the goods pages themselves cannot be converted, and the local service provider's access allows advertising to perform a task directly. For brands, this means that the conversion chain for some of the product categories has been changed from “purchase ends” to “purchase starts”, which, in turn, affects repurchase and evaluation after sale and the experience of installation.

The value mechanism for harmonizing entry points is to reduce the cost of coordination. DVA+ combines presentation, video and audio into one goal, one budget, one advertising campaign; Full-Funnel Campaigns allow AI to combine and optimize its processing channels. The economic implication is that the manpower originally used to align across platforms, match the statements and duplicate the build-up was released, but that part of the savings would not automatically translate into profit but would instead be redistributed to creative supply, data governance and experimental design. If brands lack the corresponding carrying capacity, the time saved may be consumed in more detailed day-to-day operations.

Example of scenarios: A medium-sized consumer brand used to be managed by two drop-outs each to search and out-of-station videos, each with approximately one third of the time per week spent on reconciliations and budget changes. This part of the time is theoretically released if the entry is harmonized, but if the team does not define the priority of “new customer acquisition” and “long-term return” in advance, the time saved is likely to be used to increase the number of advertising activities and, instead, to increase the complexity. This could be done through a quarterly comparison of the number of activities, the number of experiments and the change in the long-term return indicator before and after the entry point to the unified portal; If the number of activities rises without an improvement in the return, the target function is not yet clear. Therefore, the key to the trade-off is to determine the target function before deciding which decisions will be taken over by AI and which judgements will be retained by the team.

04 GSP analysis

User travels are compressed into a conversation that can be quoted.

At the launch, the Amazon stressed that shopping was rarely handled by a single tip, and that research, discovery and re-examination constituted a recurring round-trip. The operational meaning of this phrase is that brands need to cover the entire process of dialogue rather than one of the nodes of the journey. In traditional funnel models, different channels and different KPIs are known, considered, converted; In the dialogue interface, the three parties are condensed into the same interactive segment, and any missing link of information immediately terminates the dialogue. Because each user ' s query is screened, when a brand is missing in the dimension of an issue, the dialogue flows to a competitive brand that is willing to provide details, a loss that is often reflected only in a decline in exposure or clicks on the statements, and which is difficult to trace to specific problem gaps.

As a result, brand assets need to be organized according to “what consumers will ask” rather than according to internal product lines. The combination of clean, wet, sun-proof and sensitive muscles and beginners is essentially asking “what I should buy and in what order”. Brands that match these issues to the reality of their products have the opportunity to be cited repeatedly in multiple rounds of dialogue; Conversely, providing a slogan would be tantamount to giving the platform and other brands the power of interpretation. The trade-off is whether to continue to invest the budget in a more loud volume or to devote part of it to sorting out the boring details that determine the continuation of the dialogue.

The GSP suggests that the user's journey be remixed as a question map: who raises questions in what context, what kind of verifiable answers a brand can give, which product facts are the answer, and who reviews its accuracy. This map is a common document linking the Ministry of Markets, the Department of Products and the Department of Compliance. Example of hypothesis: The local service trail of Spoonsored Services may be effectively carried after a brand has “how long it takes to come to the door” with the words “what conditions are required for the installation”; If only “professional installation” is mentioned, the dialogue will be interrupted at the point of detail. The test could be a dialogue-by-article return by question map, counting the number of times each answer is cited and the follow-up follow-up follow-up rate, giving priority to the return to work of items with low reference or high follow-up.

05 GSP analysis

Organizational division of labour: who owns brand knowledge and who is responsible for AI output

When nearly a thousand indicators are accessible in natural languages, when creative tools can directly generate display and video assets, the problem facing the organization is no longer “there are no tools”, but “who is responsible for the output”. AI can generate advertisements, but cannot decide what is worth saying and what cannot be said. The maintenance, vetting and updating of brand knowledge bases must fall on clear duty-bearers, otherwise dialogue-based placements can spread out obsolete selling points and wrong matching messages at a faster rate. The trade-off lies in whether AI should be used as a reason to reduce manpower or as a magnifier requiring higher judgement density.

Rather than creating a large new sector, the rational division of labour identifies the boundaries of three types of actors: products and R & D provide verifiable facts, the Ministry of Markets is responsible for translating facts into consumer languages and answers to questions, and compliance and justice sets an insurmountable red line. The drop-and-agent is responsible for the experimental design and budget allocation and for making the final trade-off on AI ' s recommendations. MCP Server and AMC Supert Tool in the Amazon have lowered the technology access threshold, but this precisely means that people have more, not less, responsibility for auditing. The smoother the tools, the lower the cost of the wrong answer to enter the dialogue, and the shorter the correcting window.

Example scenario: A brand has found an increase in the share of new customers during the Full-Funnel Campaigns test period, but the repurchase has not been synchronized. If no one is responsible for transmitting the “what kind of dialogue scene” back to the product and content team, this signal will be seen as a recurring issue of budget optimization, and the real reason for this is that the commitment to use in the dialogue may not be consistent with the actual product. The more automatic AI is when the duty bearer is not clear, the easier it is for the organization to accelerate in the wrong direction. The test is to establish the sole liability for each branding knowledge, to review the date and expiry conditions and to check periodically whether there is an expired expression in the AI output.

06 GSP analysis

Input structure and alternative interpretation: what is left beyond efficiency

The test figures given by the Amazon at the launch — an increase in long-term advertising returns of 67%, a decrease in the acquisition cost of new guests of 29%, a sale of a pet plate 41% from new customers and an increase in product coverage for creative tool users of up to 71% — all came from the platform calibre and belonged to the manufacturer itself and could not be used as an independent industry benchmark. Brands should use these figures as a test assumption when formulating their budgets, rather than as replicable commitments. The trade-off is whether to put these figures directly into the annual targets or to design an experiment that can justify them.

Another interpretation is equally well established: improved impact may result from selection deviations during the testing period, and the early participants themselves are large and have a better data base; It may also come from short-term increases after budget concentration rather than from institutional improvements. The real test is whether the incremental amounts still exist when the budget returns to normality, when the test group is comparable to the control group, and whether the incremental amounts come from groups of people whose brands could not otherwise be reached. If the increment disappears, it suggests that more of the previously observed resource tilt effect than the structural changes brought about by AI, the continued addition will only push the marginal cost.

Thus, the core issue of the input structure is not “AI or not”, but “where will the time be saved”. If the saved operating time continues to be put into homogenous competition, the return will be quickly spread; The difference between the supply of ideas and the building of branding knowledge can result in a platform that is difficult to replicate. The judgement of the GSP is that efficiency tools are entry vouchers, not barriers; The barriers come from brands that know what to say and insist on more than others. The test could have a line of control: how much of the saved working time was reallocated to problem map maintenance and creative experiments, rather than simply translating into more investments.

07 Suggested actions

Implementation order: knowledge base first, automate

The first step is to take stock and structure it. Brands should first complete a brand knowledge audit: list the twenty to fifty issues most frequently raised by consumers, confirm whether there are proven product facts and indicate which questions are currently unanswered. This is the pre-condition for all follow-up actions and, without it, any automated placement is magnifying the information hole. The responsible person should be the head of the brand or market, and the output should be a collection of brand issues that can be reviewed internally and that can be used externally.

The second step is the experiment of small-scale dialogue. Select a product line that compares a dialogue portal with a search advertisement, and see if the consumer ' s question depth, acquisition behaviour and the ratio of new customers change when brand information is quoted. The objective at this stage is not to be large, but to find out what kind of questions and answers really affect decision-making. The input is a question bank and creative material, and the output is a controlled experimental conclusion. The reason why the experiment was placed before automation was that the full funnel tool and AI creative creation would magnify existing information structures: if there were gaps in the knowledge base itself, automation would only spread the wrong answers in more format and faster.

The third step is to expand automation. When brand knowledge is stable and the results of the experiment are clear, the channels are gradually combined and budget allocations are transferred to the single portal and AI optimized. The order cannot be reversed: automation before knowledge is supplemented will lead to faster and broader errors. For small and medium-sized brands with limited budgetary and human capacity, the first step and the second step could be a simplified version that would concentrate resources on one of the core product lines rather than being fully rolled out.

08 Suggested actions

Indicators and experiments: management of AI releases in a perjured manner

Indicator design should distinguish between three layers. The first tier is process indicators: the frequency with which brand messages are quoted in the dialogue, the breadth of the questions, and the accuracy of the answers. This layer measures the effectiveness of the knowledge base rather than the beauty of advertising. The second tier consists of efficiency indicators: acquisition costs for new visitors, marginal returns under a mix of channels, and creative production cycles. The third tier consists of outcome indicators: the ratio of new arrivals, the repurchase rate and trends in long-term sales. The three layers must be seen at the same time to avoid using efficiency improvements to mask the emptiness of brand assets.

Experiments should be guided by the principle of perjury. (b) Establish clear cross-reference conditions and state in advance how the incremental amount would be adjusted if it did not occur; (b) Separate the baseline figures provided by the Platform from their own comparisons; For the crucial question of whether a new customer is truly new, a test caliber independent of the releasing team is used. With regard to privacy and data use, the specific scope and compliance requirements need to be identified by agreement between the judiciary and the Platform, and this is not done. In order to ensure that the conclusions are rediscoverable, each round of experiments should have a fixed observation window with a attribution caliber and a contrast period before and after the introduction of automation.

Example of the hypothesis: A brand saw a significant increase in the share of new customers in the dialogue portal tests and decided to expand its placement overall. However, if the repurchase and return of these newcomers are not monitored at the same time, it may be a one-time promotion. It would be prudent to divide the experiment into two phases: to verify whether the references to the dialogue changed decision-making and then whether the added population was of long-term value. Only if both questions are answered in the affirmative will there be a basis for greater investment.

09 Suggested actions

Applicable conditions and adjustment judgement

This logic is more appropriate for three categories of brands: complex product information, and the need to explain and match product categories; Dependence on local service performance, purchase of separate product categories; And newcomers get brands with long-term pressure and declining marginal returns from existing investment. Conversely, if product decision-making is extremely simple, consumers rarely do research, or brands have yet to settle even basic product facts, the marginal value of dialogue delivery is limited, and it is more important than catching up with new tools to organize basic information.

There are three types of signals that determine whether adjustments are needed. One is the citation of quality: the extension of automation should be suspended if the quoted content of the brand is frequently wrong or ignored, indicating that the knowledge base and the audit mechanism are not in place. The second is the structure of the population: if there is an increase in the share of new customers but no improvement in the repurchase and retention, it means that the dialogue attracts a price-sensitive population and should go back to the value proposition and product experience itself. The third is the organizational load: if the operating team devotes all the time saved to increasing the number of advertising activities, indicating that the target function is not harmonized, the priority should be redefined before efficiency is discussed.

Finally, there is a need to preserve the judgement of the person. At the launch, Martha Stewart shared the same point with Chaunté Lowe: technology could spread the story further, but it was impossible to decide what story was worth telling. For brands, AI is responsible for bringing the right information to the right person, while choosing what to say, to insist on and to abandon what remains the responsibility of the operator. It is the most consistent division of labour in this cycle of change that technology should take on what it is good at, and that people should keep up with its irreplaceable judgement.