Consumers accept AI as a shopping advisor, but not a buyer
Seventy-five percent of U.S. consumers say they are comfortable with AI helping them decide what to buy. That figure reflects how quickly the behavior has spread in only a few years. The same research found that the more people use AI for shopping, the more comfortable they become with it.
There is a clear boundary, though. Consumers separate agentic shopping from agentic purchasing. They accept AI that researches products, compares options, finds better deals and narrows the choice. They resist AI that completes the transaction without a sign-off. Sixty-four percent favor AI suggesting brands they would not normally consider. Only 20% are comfortable with AI acting independently. Making a purchase without approval is the single biggest trust-breaker, cited by 55% of consumers.
The same pattern shows up on the buy side. Most advertisers and commerce media network operators want agentic AI to do the majority of the work, with human approvals built in.
What agentic AI advertising is
Agentic AI advertising is the use of AI systems that can carry out multi-step advertising and commerce tasks on their own, such as researching products, comparing options, optimizing campaigns, or proposing media-buying actions, within limits set by a human operator.
It differs from older automation in one way. Traditional automation follows fixed rules. An agentic system takes a goal, decides the steps, and adjusts as new information arrives, a distinction IBM draws between agentic and generative AI. In most current deployments, it stops short of the final decision and hands that to a person.
Two related terms are worth defining here:
- AI media buying agent: a system that selects inventory, sets bids, and reallocates budget toward a stated goal, then surfaces the results for human review.
- Generative engine optimization (GEO): the practice of structuring brand content and data so AI assistants cite or recommend it when a user asks for a product or category.
Why this is a problem worth acting on
For a decade, the rules of commerce media were stable. Success came from winning visible ad placements, buying the right audience segments, and measuring clicks. Agentic AI changes where the advantage sits.
When a shopper asks an AI assistant to find the best running shoe under a set price, the assistant returns a shortlist. Products that are not on that shortlist are effectively invisible to that shopper. The ad position on the page stops being the deciding factor. Inclusion in the AI’s recommendation becomes the deciding factor.
The industry has noticed. In the Koddi research, 84% of commerce media leaders said they would invest in opportunities designed to increase their visibility inside AI-generated answers and recommendations. Sixty-one percent are already moving that money out of performance and paid-search budgets. The funding is shifting before the standards have settled.
That creates three concrete operational problems:
- A new gatekeeper. The AI shortlist sits between brand and consumer. If your products and data feeds are not structured for AI retrieval, you lose presence at the exact moment a purchase decision forms.
- Fragmented measurement. Competing AI ecosystems use different feed structures, checkout flows, and attribution models. A brand operating across several of them is solving multiple measurement problems at once. In the study, 33% of respondents said current gaps in measurement and attribution were holding back investment.
- Budget committed before proof. A third of advertisers plan to allocate between USD 250,000 and USD 1 million to agentic AI products over the next 12 months. At that level, finance teams expect demonstrated return, and the attribution to prove it often does not yet exist.
What brands can do: options for the agentic shift
There is no single fix. The response is a set of moves, chosen against where a brand currently has gaps. Here are the main options, with the operational trade-offs.
1. Structure data and content for AI inclusion (GEO)
The first move is making sure AI systems can find, read, and trust your product information. That means clean product feeds, structured data, explicit specifications, and content written so individual passages make sense on their own. The foundational GEO study by Aggarwal and colleagues found that content edits such as adding citations, quotations and statistics measurably raised how often generative engines selected a source.
Trade-off: this is low-cost and within a brand’s control, but it competes for attention with paid placements and takes weeks to show effect.
2. Buy paid inclusion and sponsored recommendation slots
Commerce media networks are building paid formats for the agentic layer: paid inclusion in shortlists and sponsored recommendation positions. These are the direct equivalent of the old sponsored placement, moved into the AI answer.
Trade-off: this buys presence quickly, but standards and pricing are immature, and over-reliance on paid inclusion does nothing to build the organic trust signals AI systems weight.
3. Build agent-specific measurement and attribution
Because attribution is the stated blocker, measurement is where serious budget is going. In the research, 92% of respondents plan to invest in agent-specific measurement and diagnostics, and 80% would invest in measurement tools tied to AI-mediated journeys. The brands that solve attribution across fragmented ecosystems first will have the evidence to keep scaling while competitors stall.
Trade-off: high effort and cross-functional, but it is the capability that unlocks every other investment, because nothing scales without proof of return.
4. Keep humans in the decision loop by design
The market is not heading toward full autonomy. In the Koddi study, only 3% of commerce media leaders favored environments where AI runs with an almost free hand and staff are limited to oversight. The brands pulling ahead are not the ones automating the most. They are the ones building infrastructure where human and AI collaboration works reliably at scale, across different networks and data environments.
Trade-off: designing approval gates adds process, but it matches both consumer trust limits and internal risk tolerance, which reduces the chance of a costly autonomous error.
5. Run campaign optimization on a self-learning engine
This is where the engine behind the campaign matters. Adello’s mobile DSP optimizes at the level of the individual ad impression rather than broad audience segments. Its deep-learning prediction system predicts the conversion probability of every single impression, follows users through the full funnel (interactions, landings, sessions, conversions) rather than clicks alone, and carries learnings from one campaign into the next instead of starting from zero each time. The system runs automatically, while an operations team monitors for outliers.
This is the practical shape of “agentic, with a human in the loop” that the research describes. The optimization runs on its own at impression scale; people set the goals and watch the edges. For fraud and traffic quality, AdCTRL layers on top of buying to detect suspicious patterns in real time, delivering 99%+ fraud-free traffic verified by third parties. That addresses the measurement-trust gap directly, because budget committed to agentic products needs clean traffic to produce attribution anyone will believe.
Trade-off: a managed, self-learning approach reduces manual optimization load and the cold-start penalty on new campaigns, but it requires sharing goals and data with a platform partner rather than running everything in-house.
6. Concentrate effort at research and recommendation, not checkout
The data points to where consumer acceptance is highest. People are open to AI-mediated discovery and comparison. They are not open to AI spending their money unprompted. The opportunity in agentic commerce is concentrated at the research and recommendation level, where acceptance is high and brands can genuinely be discovered.
Trade-off: this narrows focus and may feel like leaving the transaction on the table, but it aligns spend with the behavior consumers actually permit, which protects against trust damage.
A short comparison
| Option | Speed to effect | Cost / effort | What it fixes |
| GEO / data structuring | Weeks | Low | AI invisibility |
| Paid inclusion slots | Immediate | Medium | Short-term presence |
| Agent-specific measurement | Months | High | Attribution gap |
| Human-in-loop design | Immediate | Medium | Trust + risk |
| Self-learning optimization engine | Ongoing | Medium | Performance + fraud |
| Focus on research/recommendation | Immediate | Low | Acceptance mismatch |
Why now
The behavior is already mainstream, and the budget is already moving. Three-quarters of U.S. consumers accept AI as a shopping advisor, 84% of commerce media leaders are investing in AI visibility, and a third of advertisers are committing up to USD 1 million each to agentic products within the year. The next 18 months will bring new monetization models and advertiser formats built specifically for agentic environments.
The work that creates an advantage, clean data feeds, structured content, working attribution, and a self-learning optimization layer, takes time to build. Brands that start now will have evidence and presence when the standards settle. Brands that wait will be buying inclusion into shortlists they had no hand in shaping.
Commerce media is moving from surfaces to systems, and from impressions to outcomes. The brands that do well in this shift are not the ones that hand everything to the machine. They are the ones that let AI do the research, optimization, and heavy lifting, keep people on the decisions that matter, and build the infrastructure that makes the two work together. The distance between the companies that understand this and those that do not is widening month by month.
Created with help of AI.
