Agentic Commerce

What is agentic commerce?

Agentic commerce is a fancy name for letting an AI assistant do some of the shopping legwork: hunting for products, comparing them, checking price and stock, now and then even placing the order. Nobody opens ten tabs any more. They ask ChatGPT, Gemini, Copilot or Alexa and get a shortlist back. So for a seller, software now reads and judges your products before any person sees them.

Say a family in Pune needs a new mixer grinder. A few years ago, they’d search Google, open Amazon and Flipkart, read reviews and watch a YouTube comparison. Now they might ask an assistant: “best 750W mixer grinder under ₹5,000 with a good warranty, available for delivery by Saturday”. Behind the scenes the assistant reads feeds, specs, reviews and stock, picks three, and links to where you can buy. The brands with full, accurate, machine-readable data made the cut. The rest were invisible.

• Jan 2026Google and Shopify announced the Universal Commerce Protocol with 20+ backers.
• Mar 2026ChatGPT’s Instant Checkout was scaled back after a few months.
• 1thing every agent depends on: accurate product data.
• ~2xnew-customer rate Walmart reported from assistant referrals compared with search.

What actually shipped, and what didn’t

Agentic commerce has had a lot of announcements and a few quiet retreats. Knowing the difference saves merchants from spending money on things that don’t matter yet.

The pattern that settled is simple: discover in the assistant, buy on the site. Assistants have taken over much of the research that used to happen on category pages and comparison blogs. People still mostly pay where they already trust the checkout. Walmart said openly that buying inside the chat converted far worse than sending shoppers to walmart.com.

What this means in practice

For most merchants, the priority isn’t building an in-chat checkout. It’s making sure agents can find, understand and trust your products, and that the handoff to your site is smooth when they send a buyer over. That’s less exciting than the headlines and much more useful.

What you can check yourself this week

You don’t need an agency to start. These checks take an afternoon and show where you stand.

1. Ask the assistants

   Open ChatGPT, Gemini and Copilot. Ask five questions your buyers would ask, budget and city included. Note which brands come up and whether you’re one of them.

2. Open Merchant Center

   Look at the diagnostics page. How many products are disapproved or have warnings? What are the top three reasons?

3. Compare ten products

   Pick ten products and compare the feed price and stock with your live site. Any difference is a problem worth fixing first.

4. Read your titles aloud

   Would a stranger know what each product is from the title alone? If not, agents won’t either.

5. Test a variant link

   Copy the link for one size or colour and open it in an incognito window. Does it show that exact variant?

6. Check robots.txt and bot protection

   Make sure Googlebot, bingbot and AI search crawlers aren’t blocked from product pages.

If most of these look fine, you’re ahead of many stores. If several fail, those are your first fixes.

Product feed readiness

A product feed is basically a big spreadsheet of your products: title, price, stock, pictures, brand, barcodes, shipping and so on. Google and Microsoft Merchant Center read it, so does Meta, and so do most AI shopping tools. Feed accuracy used to be a hygiene task for the ads team. With agents, it decides whether you’re eligible to be recommended at all.

The fields that matter most to agents

Keeping it fresh

Stale data is the most common problem we find. A feed that updates once a night can’t keep up with a festive sale where stock moves every few minutes. Options include more frequent scheduled fetches, Merchant Center’s API for instant updates, and structured data on product pages that Google can read when it crawls. Shopify and most newer platforms sync on their own. Custom builds and ERPs are usually the ones lagging behind.

Checking feed health

• Merchant Center diagnostics: disapproved items, warnings, price and availability mismatches.
• A sample of 20 products checked by hand against the live site every month.
• Price and stock spot checks during sales, when mismatches cause the most damage.

Our product feed readiness service does this audit and the fixes. Our e-commerce SEO work covers the site side.

Structured data on product pages

Structured data is a bit of hidden code, usually JSON-LD, that spells out for machines what’s on the page. For products, Schema.org’s Product and Offer types cover the name, brand, price, currency, stock, shipping, returns and reviews.

Google uses it for star ratings and prices in results, and to double-check your feed. Assistants that read pages directly get clean facts from it too. The rules are simple:

• Mark up only what’s visible on the page. Mark up things that aren’t on the page, or inflate them, and Google can penalise you.
• Prices and stock in the code must match the page and the feed.
• Include shipping and return details where Google supports them.
• Test with Google’s Rich Results Test after changes.

Most store platforms add basic product schema for you. What’s usually missing is shipping, returns, barcodes and variants, and that’s where we dig in.

Using AI to improve product data at scale

A store with 5,000 products can’t rewrite every title by hand in a week. AI tools help, as long as they work from real data and people check the output.

Here’s a sensible way to do it:

1. Start from facts. Hand the AI the supplier’s spec sheet, current attributes and photos. A product name alone isn’t enough. Its job is to tidy facts, never to invent them.
2. Use templates per category. A title pattern for sarees differs from one for phone cases. Agree the pattern first.
3. Fill attributes from descriptions. It’s very good at fishing “fabric: cotton” or “capacity: 1.5 L” out of a messy paragraph and dropping it into the right field.
4. Flag gaps instead of guessing. If a spec is missing, the output should say “unknown” so someone can check, not make up a number.
5. Review a sample. Check at least one in twenty products by hand, and every product in regulated categories.
6. Roll out in batches. Update the top-selling products first, watch Merchant Center for problems, then continue.

Done this way, a large catalogue can be cleaned up in weeks rather than months, without filling it with errors that agents will repeat to buyers.

Images and video for agents

Assistants show your pictures in their answers, and some compare the image with the title to catch mismatches. Poor photos hurt twice. The agent trusts you less, and the shopper scrolls past anyway.

• A clean main image on a plain background, showing the actual product, with no text or watermarks.
• Extra images showing scale, details, packaging and the product in use.
• One image per variant, so the green option actually shows green.
• Fast-loading formats such as WebP on your site, while keeping high-resolution originals for feeds.
• Short demo videos for products where use matters, like appliances and tools.
• Descriptive alt text that says what’s in the image, which also helps accessibility.

Marketplaces and agents

A big slice of Indian online shopping happens on Amazon and Flipkart, and both are building their own AI shopping helpers. Those helpers only read their own catalogue, reviews and seller data. Your listing quality on each marketplace decides whether their assistant recommends you.

What marketplace assistants look at

• Complete listings: every attribute filled, not just the mandatory ones.
• Images and video: clear, multiple angles, with lifestyle shots where they help.
• Reviews and ratings: volume, recency and how the seller handles complaints.
• Questions answered: customer Q&A on the listing is read by the assistant.
• Delivery promise: fulfilment by the marketplace usually means faster, more reliable delivery dates.
• Price and stock stability: frequent stock-outs hurt ranking and recommendations.

Amazon’s assistant and buy-for-me features

Amazon has expanded its shopping assistant with features that can buy on a user’s behalf, including from some brands’ own websites in certain markets. The transaction mostly stays in Amazon’s world. For brands, this means Amazon listings, brand store content and A+ content matter as much as their own site.

ONDC and open networks

India’s Open Network for Digital Commerce lets sellers list once and be found through many buyer apps. As agents grow, open catalogues like this could become another source assistants read. It’s worth keeping ONDC listings accurate if you’re on the network.

Our marketplace agent readiness service and our marketplace marketing team work on listings together.

Reviews and reputation across agents

Agents don’t only read your product pages. They read what people say about you all over: Google reviews, Flipkart stars, Reddit threads, YouTube comparisons, the odd news piece. Brilliant specs won’t save a product whose reviews keep mentioning late delivery or rude service.

What helps:

• Ask every buyer for a review at the right time, after they’ve used the product, not the day it ships.
• Reply to negative reviews calmly and fix the underlying issue where you can.
• Put warranty and service details somewhere obvious. Agents look for them.
• Send products to honest reviewers and creators, without paying for positive verdicts.
• Keep seller ratings healthy on every marketplace, because they feed the marketplace’s own assistant.

Reputation can’t be faked for long. Since agents read so widely, the surest way to look good to them is boringly simple: treat customers well.

How we run it

There’s no standard agentic commerce package, because every store’s starting point differs. The order of work, though, stays roughly the same.

1. Week 1

   Shopping prompt baseline

   We ask the major assistants the questions your buyers ask and record which products and stores they suggest, at what prices, and with what reasons.

2. Week 1–2

   Feed health audit

   Merchant Center diagnostics, field completeness, price and stock accuracy, update frequency, shipping and returns setup.

3. Week 2

   Site and structured data check

   Product schema, variant handling, page speed, crawl access for shopping agents, deep-link behaviour.

4. Week 2–3

   Platform and protocol review

   What your platform supports now (Shopify Agentic Storefronts, UCP, ACP options, payment provider features) and what’s worth turning on.

5. Month 1–3

   Fixes and content

   Feed fixes first, then titles, descriptions and attributes for priority products, then checkout and handoff improvements.

6. Ongoing

   Measure and repeat

   Monthly prompt runs, attribution reports, feed health checks, and updates as platforms change.

What gets handed over

• A feed that passes diagnostics.
• Priority products with complete attributes.
• Product schema matching feed and site.
• A shopping prompt set and baseline.
• Attribution set up for agent traffic.
• A plain-language note on which protocols apply to you.

What it cannot do

We can’t make an assistant suggest a product that’s overpriced, poorly reviewed or often out of stock. Agents are designed to protect buyers, and that’s a good thing. The work makes strong products easy to find and understand. It doesn’t hide weak ones.

How it varies by product category

For regulated categories like supplements, medicines and baby food, assistants are more cautious, and labelling rules apply. Product claims must match what’s legally allowed and printed on the pack.

Agents in B2B buying

Office buyers use assistants as well, usually Copilot or ChatGPT at their desk, to shortlist suppliers, compare specs and draft purchase requests. A purchase manager in Vapi might type:

For B2B sellers, agent readiness looks a little different:

• Product and capability pages with real specs, certifications, minimum order quantities and lead times.
• Accurate listings on IndiaMART, TradeIndia and industry directories, which assistants often cite.
• Downloadable spec sheets backed by the same facts as text on the page.
• Clear “request a quote” paths that work when a buyer arrives from an assistant.

Our generative engine optimization service covers how B2B companies get cited in these research answers.

Common myths

Mistakes we see often

1. Chasing protocols before fixing data. A fancy integration on top of a broken feed sends agents wrong details faster.
2. Feed prices that lag the site. The most common reason products get disapproved or recommended at the wrong price.
3. Ignoring shipping and returns fields. Agents answer “delivered by Friday?” and “easy returns?” with these.
4. Creative brand titles. Clever names mean nothing to an agent matching a specific need.
5. One URL for all variants. Agents send buyers to the wrong colour or size.
6. Blocking shopping bots. Aggressive bot protection can stop agents from reading product pages.
7. Judging the channel by last-click data. Agent influence often shows up as “direct”, so it gets undervalued and cut.

What’s changing next

A few developments are worth watching over the next year or two.

• More agents with permission to buy. As payment standards mature, assistants will complete more purchases within limits users set, such as “reorder detergent when it runs low, up to ₹500”.
• Agent-to-agent negotiation. Some businesses are testing their own agents that answer questions from shopping agents about stock, bulk pricing or custom orders.
• Ads inside AI shopping answers. Google already shows ads in AI Mode and AI Overviews, and others are testing. Paid and organic will sit side by side.
• Voice shopping in Indian languages. As assistants handle Hindi and regional languages better, more first-time online buyers may shop by voice.
• Stronger consumer protection. Regulators will likely set rules on who’s responsible when an agent buys the wrong thing. Clear policies now will make that transition easier.

The basics will stay the same through all of this: accurate data, honest content, a trusted checkout and good treatment of buyers.

Common questions about agentic commerce

Do we need to support checkout inside the assistant?

Not right now for most merchants. In-chat checkout was tried and largely scaled back in 2026, because buyers prefer to pay where they already trust the checkout. Focus on accurate data and a smooth handoff to your site. Turn on in-assistant buying when your platform supports it and your buyers use it.

We’re on Shopify. Are we already covered?

Partly. Shopify turned on agent features for eligible stores, which makes your products available to assistants. But they still rely on your product data. Missing attributes, weak titles and wrong stock will hold you back even with the features on.

What about WooCommerce or a custom store?

You’ll rely more on your Merchant Center feed, structured data and plugins. Support for UCP and ACP depends on your payment provider and plugins. We check what’s available for your setup.

How is this different from Google Shopping ads?

Shopping ads are paid placements. Agentic commerce is about being recommended in AI answers, mostly organically. They use much of the same feed, so fixing data helps both.

Can a small brand compete with big retailers in AI answers?

Often, yes, especially on specific questions. An agent asked for “handwoven cotton dhurrie 5×7 for a living room” looks for the best match, not the biggest brand. Clear data and good reviews matter more than size.

How do we know if agents are sending us sales?

Track referrals from assistant domains, watch direct traffic to deep product pages, tag feed links where possible and ask buyers how they found you. Expect the numbers to be undercounted.

Is buying through an AI agent safe for customers?

The new payment standards are designed so agents never see card details and every purchase has a record of user approval. Merchants should still send clear confirmations and make cancellations easy, in case an agent misunderstood.

How does this work in India with UPI and cash on delivery?

Most agent journeys still end on the merchant’s site or app, where UPI, cards and COD work as usual. Agent-initiated UPI payments have been piloted, and wider options will depend on NPCI and RBI rules. We keep checking what’s live.

How long until we see results?

Feed fixes can show in Merchant Center within days and in AI answers within weeks. Content improvements for large catalogues take longer, usually one to three months for priority products.

What does it cost?

It depends on catalogue size, platform and how much data needs fixing. A small store may need a one-time audit and fixes. Large catalogues need ongoing feed management. See our pricing page for how we scope work.

Should we block AI agents from our store?

Blocking AI search and shopping agents removes your products from their answers. Some stores block training crawlers while allowing search agents. For most merchants who want sales, allowing shopping and search agents makes sense.

Will agents replace our website?

No. Most purchases still finish on merchant sites and apps, and buyers often visit a site to check details before paying. Your site becomes the place where agent-referred buyers confirm and pay, so it needs to be fast and trustworthy.

Do marketplaces and our own site need different work?

The principles are the same: complete, accurate data and good reviews. The execution differs. Marketplaces have their own listing fields and assistant, while your site depends on Merchant Center, structured data and open agent protocols.

Can you help if our catalogue lives in an ERP?

Yes. Many feed problems start in ERPs that weren’t designed for e-commerce. We work with your IT team or ERP partner to map fields, set update frequency and fill gaps in a layer between the ERP and the feed.

Where should a small store start?

Start with Merchant Center. Fix disapprovals, make sure prices and stock match your site, and set shipping and returns. Then rewrite titles and fill attributes for your 20 best-selling products. Those two steps cover most of what agents need, and they cost little beyond time.