Agentic Search: How AI Agents Search for Shoppers

Agentic search is an AI agent breaking a shopper’s goal into searches, reading the results and deciding what to check next. It can turn a request about a product, budget and delivery deadline into a shortlist with reasons for each choice.

For retailers, this means a product can be judged against several requirements before the shopper reaches your site. Specific product facts and clear delivery information give the agent something to compare.

Agentic Search Defined

Agentic search works through a goal over several steps. The agent chooses searches, inspects what comes back and changes its next action when it finds missing or conflicting information.

A conventional search returns links for a query such as “kids’ rain jacket”. The shopper opens those links and works out which jacket meets their requirements. An AI agent search can do some of that checking, then explain its shortlist in a conversation.

Google describes a query fan-out process in AI Mode. It breaks a question into subtopics and searches for them on the user’s behalf. OpenAI’s shopping research similarly reads sources across the internet and adapts its research when the shopper gives feedback.

The agent’s next action might be another search or a question about the child’s size. Finding a product doesn’t mean the agent can purchase it. Search is one part of the wider agentic commerce journey.

How an AI Agent Searches Step by Step

An agent can search separately for the requirements in a shopping request, then compare the evidence for each candidate. Consider this hypothetical request: “Find a waterproof kids’ jacket under $80, with delivery to Perth by Friday.” The budget is in Australian dollars.

1. The agent identifies the constraints. It needs a children’s jacket, waterproofing evidence and a price below the budget. Delivery must reach the shopper’s Perth address by Friday. It asks for the child’s size and postcode if they’re missing.
2. It finds candidate products. Product feeds can supply price, variant and stock information. Store pages supply specifications such as fabric, seam construction and the manufacturer’s waterproof rating. A water-resistant jacket needs different treatment from one described as waterproof.
3. It checks the required variant. The right size must be available, even if another size appears in stock. Reviews can add accounts of rain performance or comfort, but they don’t establish a technical waterproof rating.
4. It reads delivery information. The agent checks dispatch time, order cut-offs and the delivery estimate for the postcode. “Express shipping” alone doesn’t establish arrival by Friday. Delivery charges also affect the answer if $80 is the shopper’s total budget.
5. It compares the remaining options. The shortlist explains which requirements each jacket meets and where a delivery estimate is uncertain. If none meets every condition, the agent asks whether the shopper can change the budget or deadline.

In this example, a cheap jacket with the wrong size available fails the search. A suitable jacket with an unconfirmed arrival time needs that uncertainty stated. The useful result is a shortlist tied to the request, with product links that let the shopper inspect the details.

Agentic Search Across Public Assistants and Store Sites

Public assistants search across retailers, while an agent on a retailer’s site can work within that store’s range. The available information and actions depend on the service and its connections.

Google’s AI Mode is available in Australia in English and combines web information with shopping data. Google also makes Merchant Center AI performance insights available to Australian businesses. These reports show brand and product visibility across AI Mode and AI Overviews. Availability of search features doesn’t establish availability of every checkout feature. Our guide to Google’s AI shopping features covers that distinction.

OpenAI’s ChatGPT helps shoppers compare products and refine their requirements. Its product discovery announcement explains that merchants can use their own checkout experiences after the initial Instant Checkout approach. OpenAI’s help guidance still describes Instant Checkout for eligible products and merchants. Neither statement makes payment available for every Australian store.

On your own site, shoppa is an AI shopping agent that helps customers find, compare and buy products in the chat. It uses your catalogue and store documents and supports customers after the sale. A two-line embed adds it to your site.

For search, shoppa turns a free-text request into catalogue filters and carries the customer’s requirements into follow-up messages. Someone can ask for cheaper options without starting again. Its answers stay within your range and policies, so the shortlist leads to products your store sells. It also knows which page and product the customer is viewing.

The distinction is where the search happens. Public assistants can introduce a shopper to a store. A store-owned agent helps that shopper choose within the store and continue to buying there.

What Retailers Can Do Now

Give agents the information needed to decide whether your product meets a shopper’s request. Start with the facts that would change a shortlist.

• Make product details specific. Include materials and dimensions, with size information at variant level. A jacket’s waterproof rating is more useful for this search than “ready for adventure”.
• Keep price and stock consistent across feeds and product pages. Identify which variant each value describes. OpenAI notes that changes to prices and shipping terms can take time to appear in shopping results.
• Explain delivery by destination. State dispatch times and order cut-offs, with postcode restrictions and the difference between an estimate and a guarantee.
• Keep returns and sizing information accessible. Separate your change-of-mind policy from consumer guarantee rights. Put exclusions beside the policy they qualify.
• Review what a shopper actually sees. Choose a product with several sizes and compare the feed with its page. Correct a mismatch at its source so other channels receive the same facts.

Agents can check a stated price or size against another source. A vague benefit gives them less to work with. Complete information won’t guarantee inclusion or a particular ranking, but it reduces missing facts when a product is compared. Use the AI-ready product data guide to organise those details.

Open Questions for Australian Stores

Paid placement and responsibility for an agent’s decisions need attention as more shopping moves into conversations. A recommendation can look personalised even when a commercial arrangement affects what the shopper sees.

OpenAI describes ChatGPT product results as independent of ads and partnerships. Its statement applies to its own product results. Retailers need to understand how each channel labels paid placement and explains why a product is recommended.

Australian consumer protections already provide a baseline. The Australian Competition and Consumer Commission (ACCC) sets requirements for product and delivery claims, including accuracy. It also explains that customers receive consumer guarantees automatically when they buy goods and services.

Agent-made purchases raise further questions about what the customer authorised and who made a disputed promise. Responsibility for a wrong product, an unauthorised purchase or an incorrect delivery claim depends on the circumstances. Search and payment also involve different permissions, even when they appear in one conversation.

For your store, keep product claims supportable and make delivery conditions visible before a customer decides. Preserve the recommendation and order information needed to resolve a dispute. A search result that explains its trade-offs gives the shopper a better basis for choosing than an unexplained “best” label.