In November 2025, Adobe tracked a 1,200% jump in retail traffic from AI sources. Not AI-assisted traffic. Traffic where an AI system did the browsing. The human never visited the product page. By Cyber Week 2025, AI-influenced purchases had driven $67 billion in online spending. And according to Salesforce’s State of Commerce report, 73% of consumers now use an AI agent or AI-powered assistant somewhere in their purchase journey — a number that sat close to zero three years ago.
Most brand marketing was not built for any of this. The websites, the campaigns, the SEO strategies, the carefully produced product photography — they were built on an assumption so fundamental nobody wrote it down: a human is present at the moment of discovery. That assumption is getting quietly dismantled, faster than most marketing teams have noticed.
What agents actually do — and why it matters
ChatGPT telling you the best running shoes for flat feet is one thing. An agent that checks your size history, compares current availability across five retailers, confirms it against a budget you set three weeks ago, and places the order — that is something else. Google’s Project Mariner, OpenAI’s Operator, Amazon’s Rufus, and Apple’s evolving intelligence layer are all building toward the second version. They browse, compare, transact, and execute.
Gartner projects that by 2028, 33% of enterprise software applications will include agentic AI, enabling 15% of day-to-day work decisions to be made autonomously. By 2027, Bain estimates agentic AI could account for 25% of US e-commerce. Autonomous agents are already projected to handle $150 billion in consumer transactions before the end of 2026.
What agents look for when evaluating a purchase is not what humans look for. They do not respond to emotional advertising, brand story, or visual identity. They parse structured data. They read schema markup. They check pricing consistency across sources. They pull verified reviews. They follow instructions the user encoded previously — sometimes explicitly (‘always prefer sustainable brands’), sometimes implicitly through past behaviour. A brand whose product data is incomplete, whose pricing varies between its own website and third-party listings, or whose product attributes are not machine-readable will simply not appear in many agent-generated shortlists. The creative work never gets a chance to matter.
The SEO parallel — and why it is more disruptive this time
In the mid-2000s, brands that had not invested in SEO discovered that Google had become a gatekeeper they had ignored. The agentic web is a similar moment, but the stakes are higher and the window to act is shorter. With SEO, at least the human still saw the results page and could be swayed by a headline. With agentic commerce, the human may never know what was on the shortlist they were not on.
BCG’s analysis found only an 8% to 12% overlap between traditional search results and AI-generated answers. That gap matters enormously — it means optimising for Google does not optimise you for agents. The emerging discipline is Answer Engine Optimisation: structured data, context-rich content with proper schema markup, consistent presence across the third-party sources AI systems trust most (Reddit and Wikipedia are the most-cited domains in ChatGPT responses), and direct integration with agent platforms through product feeds and API connections.
| CASE STUDY
A Global Skincare Brand — BCG, 2025 Facing a product range of 750 SKUs and a customer base that reported decision paralysis (70% said they felt overwhelmed by choice), the brand worked with BCG to deploy an agentic AI that drew on 150,000 dermatologist-annotated images and real-time inventory to generate personalised recommendations. The agent replaced a static quiz with a dynamic, conversational experience that adapted to user inputs in real time. The brand now has an always-on touchpoint that deepens customer relationships — and captures nuanced preference data from every interaction that feeds back into future personalisation. |
The skincare example points to something the channel-by-channel analysis misses: the most competitive brands are not just optimising to be found by agents. They are building their own agents, deploying them as customer touchpoints, and using those interactions to gather the preference data that makes future personalisation better. There is a compounding dynamic here. Every agent interaction trains the system. Brands that start now are building a data asset; brands that wait are not standing still.
What the data says about channel by channel
Content and discoverability shift more than most marketers expect. AI-sourced traffic to retail sites jumped 527% from January to May 2025 alone. That growth is coming almost entirely through structured data and machine-readable content — not through campaign-driven traffic or social referrals. Schema markup and factual accuracy across every source an agent might consult (the brand’s own site, third-party retailers, review platforms, price aggregators) become the primary discovery assets. Agents deprioritise brands whose information conflicts across sources. A product listed at one price on the brand’s website and a different price on a marketplace will often be filtered out entirely — not flagged, just absent.
Brand and creative investment does not become redundant, but its function in the funnel shifts. Humans still make complex, high-involvement decisions — which car, which agency, which holiday. Agents handle routine and transactional decisions. The creative work that matters most in an agentic environment is the work that shapes consumer preferences before the agent takes over: the campaign that makes someone add ‘prefer brand X’ to their agent configuration, or the influencer post that shapes a stored value the agent will later act on. The influence happens further upstream, and the connection to eventual purchase is harder to trace. Attribution models that are already struggling will struggle more.
Influencer and creator content is harder to evaluate but not irrelevant. Agents act on the preferences consumers have stored — and creators shape those preferences. A consumer who has been influenced by a creator to value a specific brand attribute (sustainability, durability, a particular aesthetic) carries that preference into their agent configuration. The challenge is that this influence now happens further from the purchase moment, making it harder to attribute. Brands that cut creator investment because it does not show up cleanly in agentic conversion data will be making a mistake — they will be de-funding the upstream influence that shapes what agents are instructed to prioritise.
The trust question nobody is answering well
OpenAI introduced ads into ChatGPT in February 2026. Demis Hassabis of Google DeepMind publicly stated Google has ‘no plans’ for ads in Gemini, calling it a trust risk. Two of the largest agent builders taking opposite positions on the same question — whether ads belong in agent recommendations — tells you how unsettled the landscape is. For brands, this creates a genuine strategic uncertainty: the rules governing how agents surface products are still being written, and the companies writing them have divergent interests.
Adobe’s 2026 research found that one-third of consumers will stop interacting with a brand if they discover the content was AI-generated without disclosure. Gartner’s survey of 335 US consumers found 78% called clear labelling of AI-generated content ‘very important’ or ‘the most important factor’ in maintaining trust. Consumers are more sceptical about this than most marketing teams are currently assuming. Brands that are transparent about where AI is operating in their customer interactions — and where it is not — are building a durable asset. The ones treating it as a disclosure risk to be managed are likely misjudging their customers.
Bangladesh: the gap between the two versions of this story
For brands in Bangladesh competing in export markets — garments, leather, pharmaceuticals, agricultural products — the agent-readiness question is already urgent. Global procurement agents evaluating supplier options are operating now. A Bangladeshi manufacturer whose product specifications are incomplete, whose certifications are not machine-readable, or whose pricing is inconsistent across platforms is structurally less visible to the agent-mediated sourcing processes that international buyers increasingly use. This is not an abstract future risk. It is a present competitive disadvantage that compounds with every quarter it goes unaddressed.
The domestic picture is earlier-stage but moving fast. Agentic adoption in Bangladesh will likely arrive first in financial services and travel — categories where the structured data infrastructure exists and consumer benefit is immediate. The brands in those categories that have built machine-readable product and service information, consistent review profiles, and API accessibility will surface. The ones that have not will be invisible to a decision-making layer their customers are already using.
There is also an opportunity most brands are missing. BCG found that only 10% of consumer goods and retail companies globally have successfully integrated AI agents across their teams and workflows — despite the fact that LLMs now directly influence up to 20% of purchasing decisions. The gap between adoption and available capability is widest in markets where competitors have also not moved. Bangladesh’s brand landscape is earlier than the US or UK on agentic investment, but so are its competitors.
| CASE STUDY
Foodora — Intelligent Timing Agent The food delivery platform deployed an agentic workflow that analysed individual customer behaviour to determine optimal message timing — not a scheduled send, but a per-customer decision made in real time by the agent. The result was a 41% conversion rate from messages sent with the timing agent active, and a 26% reduction in unsubscribes. The case is instructive because the agent was not doing anything a human could not theoretically do — it was doing it at a scale and speed no human team could match, and improving with every interaction. |
Three things worth doing before the window narrows
Audit what an agent actually sees when it evaluates your brand. This means checking product data completeness, pricing consistency across platforms, schema markup coverage, and what appears when an agent queries your category. Most marketing teams have never done this audit because the audience it addresses did not exist two years ago. Running it is uncomfortable, because the gaps are usually larger than expected.
Treat product data as a marketing function, not a logistics one. The most common failure mode is that structured product information lives in a supply chain or operations team, managed for internal use, and is too incomplete and inconsistently formatted for agent evaluation. Moving that capability into marketing’s remit — or at minimum making marketing responsible for its quality — is a structural change most organisations have not made.
Do not cut upstream brand investment to fund agent optimisation. The most sophisticated brands in this environment are doing both: building the data infrastructure agents evaluate and investing in the creator and campaign work that shapes the consumer preferences agents then act on. The two budgets are not in competition. One without the other underperforms. An agent-optimised brand with no upstream consumer preference built in its favour will appear on shortlists but lose the final decision to the brand that did the earlier work.
SOURCES
Adobe Digital Economy Index 2025–26 · Salesforce State of Commerce 2026 · Gartner Mainstream Marketing Predicts 2026 (Jan 2026) · BCG — How AI Agents Are Transforming Consumer Goods (Dec 2025) · BCG — Agentic Scenarios Every Marketer Must Prepare For (Apr 2026) · Bain — Agentic AI and US E-commerce by 2030 · Adobe 2026 AI and Digital Trends: Customer Behaviors (Mar 2026) · PwC AI Agent Survey (May 2025) · McKinsey State of AI 2025 · Adweek — 10 AI Marketing Trends for 2026 (Feb 2026) · Rellify — Top 8 Agentic AI Marketing Trends for 2026 · Digital Applied — AI Marketing Statistics 2026 · Credencys — AI Agents Are Changing How Consumers Shop (Dec 2025) · AI Magicx — Agentic Commerce Explained (Mar 2026) · CommercetTools — 7 AI Trends Shaping Agentic Commerce (Apr 2026) · HBR — How Brands Can Adapt When AI Agents Do the Shopping (Feb 2026)
