Agentic AI in Shopping — A Merchant Primer

This is a primer for operators — heads of ecommerce, CMOs, founders — who have been hearing the phrase "agentic AI" for nine months, have read the explainer posts, and want to know what to actually do on Monday. No hype. No 2030 predictions. Just the work that ships value this quarter.

The one-liner

Agentic AI is software that takes a goal in natural language and completes multi-step tasks on a user's behalf. In commerce, it is the layer that increasingly does the discovery + comparison work for shoppers before they ever see your store.

What you should ship this quarter

Five things. Total dev cost: roughly two engineer-weeks across the list. Total marketing cost: a single resource hub refresh.

1. Open AI crawlers

Edit /robots.txt to explicitly allow GPTBot, ClaudeBot, PerplexityBot, Google-Extended, Bytespider. This is fifteen minutes of work and most stores are still blocking by accident.

# robots.txt
User-agent: GPTBot
Allow: /

User-agent: ClaudeBot
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: Google-Extended
Allow: /

2. Ship an /llms.txt

Publish a single text file at /llms.txt that lists your 12-24 most citable pages — PDPs, category guides, FAQs — with one-line descriptions. Acts as a table-of-contents for AI crawlers. Idukki generates this from your sitemap + JSON-LD automatically.

3. Audit JSON-LD

Run Google's Rich Results validator across your top 20 PDPs and your top 10 resource pages. Fix anything red. Most stores have at least one broken AggregateRating or stale Product schema. Agents lean heavily on this; broken schema means invisibility.

4. Restructure your resource hub

Pick the top 8 questions buyers ask in support tickets. Write a single resource page per question with a clear "question → 2-sentence answer → evidence" structure. This is the shape AI agents quote.

5. Stand up citation monitoring

Read your server logs for referer headers from chatgpt.com, perplexity.ai, claude.ai. Count them weekly. Tag the URLs that appear most often. This is a 30-line scheduled job, no SaaS subscription required.

What to ignore until 2027

Three things have hype-to-impact ratios that are too high to chase yet.

• Building your own branded agent on the storefront. Wait for a clear winner among the agent-orchestration SaaS vendors. The cost of switching after committing is too high.
• Agent-to-agent B2B negotiation. Real but still embryonic — under 1% of B2B GMV in 2026. Track it; do not staff against it yet.
• Wallet-level "buy-with-agent" integration. Tooling is converging fast and the early movers are going to rebuild twice.

A three-month arc that works

1. Month 1: Crawlers + JSON-LD audit + llms.txt. Start counting AI-engine referrals. Baseline established.
2. Month 2: Resource hub refresh — 8 question-shaped pages. Begin to see citation count grow in ChatGPT.
3. Month 3: Verified-buyer reviews bound to SKUs with isVerifiedBuyer. Agents start preferring your PDPs over competitors with unverified reviews.

By month four you should see measurable referral lift from AI engines — typically 4-12% of total organic referrals on a mid-market DTC store. The brands ahead of you will be at 15-25% by month six. The brands still blocked are at zero.

Closing

The agentic AI conversation in ecommerce has the same shape every new distribution channel has when it arrives. Year one: dismissed. Year two: chased. Year three: a moat. We are still in year one. The cheap window to set up properly is now.