Influence Humans and Agents
AI interfaces are becoming a new decision layer in commerce. Brands need information machines can retrieve and people can trust.
For two decades, most e-commerce marketing was designed around human attention, even as search engines, marketplaces and recommendation systems mediated what people saw.
Now generative interfaces can interpret intent and assemble a shortlist before a customer reaches a product page. Some can compare products, answer service questions or help complete a purchase. A smaller set can act after a customer confirms the action.
This is not one mature, universal “agent channel”. AI search and product discovery are live; integrated checkout is selective; interoperable commerce protocols and autonomous replenishment are still emerging. The shift is uneven, but the operating risk is already concrete: missing, stale or contradictory information can cause a product to be overlooked or represented inaccurately.
Two prongs, one brand
The durable centre of e-commerce remains product, value proposition and human trust. That work continues.
The additional requirement is to make the brand findable, accurate and comparable in the AI systems customers actually use. Influence over agents is not a replacement for influence over humans. It is the discipline of making sure software can retrieve and represent the same commercial promise a person sees.
- Influence humans
- Awareness, desire, trust and brand codes form the emotional and commercial centre.
- Influence agents
- Accurate product and policy information on crawlable pages and in the platform feeds that matter.
Machine legibility can earn consideration. Product, value and brand still earn preference.
What is live — and what is emerging
These capabilities do not share one standard, one data source or one maturity level. A conversational answer, a product carousel, a browser agent and a transacting agent are different systems. Treating them as interchangeable creates false confidence.
Discovery AI search and shopping systems can already research and compare options. Google says its AI search features use the same SEO foundations as Search, with no special AI markup required. Product shopping surfaces can also use current merchant feeds and product metadata. Google’s guidance separates those two mechanisms.
Choice Some comparison and shortlisting can now happen inside an AI interface before — or instead of — a visit to a product page. OpenAI’s shopping guidance says results can consider price, availability, reviews and context, but also warns that not every product will appear and generated labels are not guarantees.
Commerce Selected platforms support AI-assisted or agent-executed purchase flows, usually for eligible merchants, in limited markets and with explicit customer confirmation. Google’s initial rollout, for example, was limited to eligible US merchants. OpenAI has since put more emphasis on discovery and merchant-owned checkout after finding its first Instant Checkout approach too inflexible.
Service AI service agents can already resolve some routine enquiries and escalate others. The human brand still shows up in promise design, accountability and the moments that require empathy or judgement.
Explore the journey
Compare what must be true for people, agents and the shared brand at each point in the journey.
Discovery
- Operating requirement
- Keep the proposition, product truth and availability consistent across every surface.
- When the signal fails
- The promise a person remembers and the facts an agent retrieves begin to diverge.
Choice
- Operating requirement
- Make product attributes, proof points, variants and policies faithful to the same commercial promise.
- When the signal fails
- Contradictory claims create doubt for people and unreliable comparisons for agents.
Commerce
- Operating requirement
- Keep price, availability, fulfilment and policy promises accurate at the moment of action.
- When the signal fails
- The channel says yes while the operation cannot reliably fulfil the promise.
Service
- Operating requirement
- Design clear promises, ownership and escalation paths across automated and human service.
- When the signal fails
- Exceptions move between systems and teams without anyone owning the customer outcome.
Start with honesty
Begin with a repeatable baseline, not a one-off prompt. Choose commercially important questions, run them across the relevant platforms more than once and record the date, location and account state. Score whether the brand appears, which source is used, and whether the product, price, stock, variant and policy details are accurate.
Treat those results as directional. Responses vary by prompt, context, location, inventory and model version. Combine the audit with referral analytics, merchant-platform reporting and Search Console data where available. Google only began rolling out dedicated generative-AI visibility reports to a subset of sites in June 2026, so cross-platform attribution is still incomplete.
Then repair the commercially important gaps: stable product identifiers, variant clarity, current price and availability, shipping and returns, verifiable claims, crawlable policy pages and the specific feeds supported by each priority platform. Google recommends combining on-page product structured data with Merchant Center feeds; OpenAI’s product-feed specification likewise includes price, inventory, variants and return-policy fields.
There is no universal “agent feed”, no guaranteed inclusion and no reliable shortcut around product quality or operational accuracy. Teams inside CognitionHub compare findings with adjacent peers so investment follows customer behaviour rather than vendor hype.
The operating requirement
Dual-audience work is not “add an AI tool and hope.” It requires one governed source of product and policy truth, distributed across product pages, feeds, service knowledge and checkout. Ownership will usually span merchandising or catalogue operations, technology and data, customer service and commercial leadership.
Freshness matters as much as structure. OpenAI recommends publishing a complete catalogue snapshot at least daily. Google says structured data should match visible page content and warns that meeting its technical requirements does not guarantee indexing or inclusion. Machine-readable does not automatically mean accurate, and eligibility does not mean visibility.
Run the work as a repeatable operating loop:
Continue Reading
See All
Visibility in the Age of AI
How commerce brands enter consideration when customers delegate more of the decision to AI.
Feed Completeness Is a Management Discipline
Your weekend sale lives on the product page. The shopping feed still shows full price. Agents shortlist from the catalogue, not the PDP. That is how you lose a comparison you never see.
Comparing Apples with Oranges
A storefront that feels perfect to humans can hide a catalog that is gibberish to agents. That gap is now commercial.
Bring the next decision to the Network.
Compare notes with experienced peers and contribute what you have learned.