The Three Roles of an Agent in Your Buyer's Journey: Researcher, Comparison Shopper, Closer

“AI shopping” is too coarse a category to do useful work with.

When a CMO sits down to allocate budget against agentic commerce, the question they actually need to answer is more specific than “are agents important.” The question is: where in the buyer’s journey is an agent doing the work, what does the agent need from us at that point, and how do we measure whether we’re winning that part of the journey?

This post is a framework for that. We split the agent’s role into three: the agent as researcher, the agent as comparison shopper, and the agent as closer. Each role has different content needs. Each rewards different signals. Each is measured by different metrics. A merchant who treats them as one thing will over-invest in some and under-invest in others. A merchant who treats them as three different jobs will allocate cleanly against each.

This is the framework we use internally when we think about merchant catalogs. It works.

Role 1: The Researcher

The agent as researcher is the top of the funnel. The buyer hands the agent a vague or open-ended goal (“I need a winter coat,” “I want to try sourdough baking,” “help me figure out skincare for my skin type”), and the agent does broad-spectrum research to build a starting understanding.

The agent in this role is not yet evaluating specific products. It’s building the shortlist of categories, brands, and considerations the buyer should care about. The work happens against the open web, against editorial coverage, against brand-level structured data, and against llms.txt files of plausible candidates.

What the agent needs from you in this role:

• Brand-level discoverability. The agent has to be able to find you when it does its research pass. This means clean editorial coverage, brand-level structured data on your homepage and category pages, and presence in the open-web sources the agent indexes from.
• A clear declaration of what you sell. Your llms.txt should answer “what is this brand actually about” in the first paragraph. Vague brand stories lose to specific category declarations.
• Topical authority signals. Pages that demonstrate real expertise on the category (guides, comparison articles, technical deep dives) get cited as the agent builds context, even if they don’t sell directly.

What you don’t need at this stage:

• Aggressive conversion optimization. The agent isn’t trying to buy yet. CRO patterns (popups, exit-intent modals, scarcity countdowns) actively hurt you at this stage, because the agent flags them as friction and downweights the source.
• Pricing prominence. The buyer is still building their map. Whether your jacket costs $389 or $489 isn’t the question yet. Pages that lead with price miss what the agent is looking for.

How to measure success here:

The metric for researcher-stage agent visibility is mention rate. When agents are asked the broad open-ended questions in your category, how often does your brand appear in the answer? Most merchants don’t measure this today. The ones who do are running structured query sets weekly against the major agents and tracking their mention rate across queries. This is the agent-era equivalent of brand-search volume tracking. Build the measurement now or be guessing at this part of your funnel through 2026.

Role 2: The Comparison Shopper

The agent as comparison shopper is the middle of the funnel. The buyer has narrowed their goal (“a synthetic insulated jacket, around $400, for Vermont skiing”) and the agent is now evaluating specific candidates against specific criteria.

The agent in this role is doing the structured comparison work the buyer used to do themselves on a SERP. It’s pulling structured data from each candidate, evaluating each one against the buyer’s stated constraints, weighing the trade-offs, and assembling the recommendation.

What the agent needs from you in this role:

• Specific, verifiable structured data. The agent is now making numeric comparisons. Temperature rating, weight, dimensions, AggregateRating, price, return window. The merchant whose data is specific and verifiable wins on comparison. The merchant whose data is vague or inconsistent loses on comparison.
• /agent/query endpoints that answer constraint-specific questions fast. The buyer’s constraints don’t always map cleanly to existing PDP content. An agent issuing “what’s the lowest temperature this jacket is rated for” benefits from a structured response, not a paragraph the agent has to parse.
• Honest review distributions. AggregateRating with the real distribution. Specific written reviews that name real attributes. The merchant whose reviews are aggregated honestly wins the trust signal that drives the agent’s recommendation confidence.
• Real-time price and availability truth. The single most expensive lie at the comparison stage is being recommended for a product that’s actually out of stock. The agent that gets burned once stops trusting your catalog.

What you don’t need at this stage:

• Brand storytelling. The agent has already decided you’re a candidate. The story has already been read or skipped. Pages that lead with brand story instead of specific comparison data lose this round.
• Lifestyle photography. Useful for the buyer who’s going to land on the page, irrelevant to the agent’s comparison logic.
• Trust-badge widgets, “as seen in” press logos, founder photos. These are signals optimized for human visitors. Agents ignore them.

How to measure success here:

The metric for comparison-stage agent visibility is shortlist rate. When agents are asked the specific multi-constraint version of your category’s queries, how often is your brand in the final 3-5 recommendation set? This is measurable today by running constraint-specific query sets weekly. It’s a leading indicator of conversion volume.

Role 3: The Closer

The agent as closer is the bottom of the funnel. The buyer has decided to purchase, and the agent is executing the transaction. This might happen entirely inside the conversation (ACP, Instant Checkout, Apple Intelligence’s commerce path), or it might happen via a click that takes the buyer to the merchant’s checkout flow.

The agent in this role isn’t deciding. It’s executing. The merchant’s job is to make the execution clean.

What the agent needs from you in this role:

• A working ACP integration if the agent is closing inside the conversation. The merchants who have ACP wired up capture buyers who never load the storefront. The merchants who don’t lose them at the moment of intent.
• A fast, clean checkout flow if the agent is handing off to your storefront. The page that loads has to convert the buyer who arrived already committed. Slow loads, complicated forms, account-creation requirements all bleed off the conversion.
• Accurate, real-time inventory and pricing. The buyer who completes a purchase based on a price the agent quoted and then sees a different price at the storefront has a worse experience. The buyer who completes a purchase for an item that’s actually out of stock has a worse experience. Both reduce the agent’s confidence in recommending you next time.
• Clear policy disclosure to the agent. The buyer who’s about to commit may want to confirm the return policy, the warranty, the shipping cutoff. The agent should be able to answer from your llms.txt and structured data without the buyer having to leave the conversation.

What you don’t need at this stage:

• Cross-sells, upsells, “frequently bought together” widgets. These are optimized for human-browsing carts. Agents either ignore them or actively flag them as friction.
• Email-capture popups. The buyer is closing. Don’t put a popup in the way.

How to measure success here:

The metric for closer-stage agent visibility is conversion completion rate. When the agent reaches the close step on a session where you were the recommended product, how often does the transaction complete? Where ACP is wired, this is measurable directly. Where the agent hands off to your storefront, this is measurable as the conversion rate of agent-referred traffic. Both should be tracked separately from your overall site conversion rate.

The composite buyer

The reality of most buyer journeys is that one agent plays all three roles in one session. The buyer types in a vague goal. The agent researches. The buyer adds a constraint. The agent comparison-shops. The buyer says “the second one,” the agent closes.

This means a merchant can’t be excellent at only one role. The buyer who runs the full sequence will drop off if any role isn’t met. The brand the buyer never saw at research stage won’t be in the comparison shortlist. The brand that didn’t survive the comparison stage won’t reach the close stage. The brand that botched the close stage loses the transaction even if the first two stages went perfectly.

This is the reason “agentic commerce” can’t be solved by optimizing for any single stage in isolation. It’s an end-to-end discipline. The merchant who’s good at structured data but bad at ACP integration loses the close. The merchant who’s good at ACP but bad at llms.txt discoverability loses the research. The math doesn’t average; it multiplies.

The CMO version

The version of this we use when explaining to a CMO why their existing P&L allocation might not match where the agentic value lives:

Most marketing budgets are built around a “TOFU / MOFU / BOFU” model that maps roughly to research / comparison / close. The shift with agents is that the agent is doing all three jobs faster than your existing funnel does them, and your money is going to be split between the existing funnel and the agent funnel for the next several years.

The allocation question isn’t “should we invest in agentic.” The allocation question is what percentage of buyers in your category are running their TOFU / MOFU / BOFU through an agent today, what that share will be in 12 and 24 months, and how your current spend is split between supporting the human funnel and the agent funnel. Most brands we’ve audited are 90/10 human/agent in spend allocation when their actual buyer mix is already 75/25 or 65/35 in some categories.

The corrective math is uncomfortable. The brands that do the math early get the leverage. The brands that wait for the obvious version of the data will reallocate later, after losing market share to the brands that moved first.

Three roles. Three different sets of content, signals, and metrics. One end-to-end multiplied outcome. Worth treating it that way.