What is agentic commerce? How AI shopping agents change product discovery
Agentic commerce is shopping where an AI agent does the searching, comparing and shortlisting. Agents read facts, reviews and customer evidence, then decide.
A shopper types "best linen overshirt under £90 that does not wrinkle" into an AI assistant. Three brands come back, with reasons. She never saw your product page, your hero shot or your banner. That is agentic commerce, and here is how the agent decided.
In this article
For twenty years, ecommerce optimisation meant one thing: get a human onto a page and persuade them. Better hero shots, faster load, a tighter add-to-cart. Every playbook assumed a person with eyes, a scroll wheel and a few seconds of patience.
That assumption is quietly breaking. A growing share of product discovery now starts with a question typed into an AI assistant ("best linen overshirt under £90 that does not wrinkle") and ends with a shortlist the assistant assembled. The shopper never saw your homepage. Something read it for them. That is agentic commerce, and it changes what "optimising" even means.
What "agentic commerce" actually means
Agentic commerce is any purchase journey where an AI agent does part of the work a shopper used to do themselves: searching, comparing options, reading reviews, narrowing a shortlist, and in some cases completing the checkout. The person stays in charge of the decision. The agent does the legwork.
It sits on a spectrum. At one end, an assistant simply answers "which of these three is best for me?". At the other, a fully delegated agent reorders a consumable without being asked. Most real journeys today are in the middle: the agent researches and recommends, the human approves. Either way, a non-human reader is now between your catalogue and the buyer.
How an agent shops differently from a person
A person forgives a messy page. They infer, they squint at a photo, they trust a brand because the typography feels expensive. An agent does none of that. It is fast, literal, and comparison-driven, it will happily line your product up against nine others and weigh them on whatever facts it can actually extract.
- It rewards specificity. "Breathable" loses to "linen, 170 gsm, pre-washed".
- It trusts evidence over adjectives. A claim with a review behind it counts; a claim on its own is discounted.
- It is unmoved by visual polish. A beautiful page with thin facts loses to a plain page with rich ones.
- It never sees a page it cannot parse. If the proof is locked inside an image or a slow widget, to the agent it does not exist.
A person shopping against an agent shopping
A person
- Scrolls the page and looks at the hero shot
- Can be won by visual polish
- Reads proof inside images and widgets
An agent
- Reads structured facts: "linen, 170 gsm, pre-washed"
- Counts claims backed by a review; discounts the rest
- Skips any proof it cannot parse
A human shopper asks "do I like this?". An agent asks "can I defend recommending this?". Those are different questions, and they reward different pages.
Rohin Aggarwal · Co-founder & CTO, Idukki
What an agent reads, and what it ignores
When an agent evaluates your product it is pulling from a narrow set of surfaces: the structured data in your page markup, the plain-text description, the review corpus, and whatever third-party coverage it can find. It largely ignores decorative imagery, motion, and anything that needs a click or a hover to reveal. The gap between "what a person experiences on your PDP" and "what an agent can extract from it" is the new conversion gap.
A human shopper
Emotional, forgiving, persuadable in seconds.
Wins at
- Responds to photography, story and brand feel
- Infers quality from polish and confidence
- Will tap to expand, watch a video, scroll
Struggles with
- Easily distracted; bounces on friction
- Hard to reach at all if an agent shortlisted first
An AI shopping agent
Literal, comparison-driven, evidence-hungry.
Wins at
- Reads structured facts and reviews at scale
- Recommends what it can justify with evidence
- Never tired, never skims past your spec
Struggles with
- Cannot see proof trapped in images or slow widgets
- Discounts adjectives with nothing behind them
Why a page tuned only for humans underperforms once an agent is in the loop.
Why customer evidence matters more to agents
There is a reasonable fear that AI flattens brands, that if an agent just reads specs, the brand with the cheapest spec wins. In practice the opposite pressure shows up. Agents are built to be cautious; they look for corroboration before they recommend. The strongest corroboration on the internet is what real customers said and showed: reviews tied to a specific SKU, photos of the product in a real room, a video of someone actually using it.
That customer evidence is the one moat an agent cannot dismiss as marketing. It is also the material most brands leave scattered across social platforms and review tools, unstructured and unreadable. Pulling it onto the product page, tied to the product, in a form an agent can parse, is the highest-leverage move available right now.
What to do this quarter
- 1Audit one high-value product page as an agent would: strip out the imagery and ask what facts survive as plain, structured text.
- 2Make every important claim specific and corroborated, pair each one with a review or a piece of customer media that backs it.
- 3Surface customer photos and videos inline on the PDP, tied to the SKU. A separate social feed buries them.
- 4Make sure that evidence is in the page markup and loads fast, proof an agent cannot reach is proof you do not have.
- 5Re-run the audit monthly. The agents are improving faster than your catalogue is.
Sources & notes
- 1Gartner, research on AI agents in commerce and customer journeys · Direction of travel for delegated, agent-assisted buying.
- 2Adobe Analytics, generative-AI referral traffic to US retail sites · Evidence that AI assistants already sit upstream of the product page.
- 3Baymard Institute, product-page UX research · How shoppers (and, by extension, agents) consume PDP content.
- 4Idukki: ChatGPT can try on your products now. Can it find them? · ChatGPT try-on, October 2026, and what merchants should fix.
- 5Note on framing · This article describes a direction. Idukki did not measure it. Treat the behaviour described as representative of where agent-assisted shopping is heading.
41%
of 18-34s use GenAI to search products
Bazaarvoice 2025 SEI
$2.6T
Global social commerce 2025
eMarketer
+108%
TikTok Shop US growth YoY 2025
eMarketer
Continue reading
2 pieces in this clusterThese long-form pieces on the Idukki blog link back to this article, go deeper on the cluster.
- Agentic commerce
Agentic checkout explained: ACP, Instant Checkout and what merchants should do now
Agentic checkout lets an AI assistant complete the purchase for the shopper. How ACP, Instant Checkout and Google's AP2 work, and what merchants should do now.
- AI search
The Future of eCommerce Is Shoppable Video: 5 Things You Need To Know
Five shifts to prepare for in shoppable video, from AR try-on to AI shopping agents.
More from Rohin Aggarwal
- Agentic commerce
Will AI agents recommend your store? A 2026 readiness read
An agent recommends you only if it can reach, parse, trust and quote your catalogue. If it cannot, you never make the shortlist. Here is how to tell where you stand.
- Agentic commerce
Agentforce Commerce just went GA: what it means for your product pages
Salesforce's Agentforce Commerce is now live, plugged natively into ChatGPT, Google AI Mode and Gemini. Here is what the GA actually changes for brands trying to win a sale through an AI agent.
- Agentic commerce
The Agentic Commerce Stack: a reference architecture for merchants in 2026
Seven layers sit between a shopper and your checkout. A reference architecture plus a 90-day plan for being visible, citable and convertible inside AI shopping agents.