Let AI channels discover your products (Agentic)
For store owners and QA testers: turn on Agentic discovery, understand the open product feed, llms.txt and MCP endpoint that power it, choose which AI channels may read your store (switching one off blocks its crawlers in robots.txt), and fix the listing problems the quality score flags.
Download this guide: PDF · Word (.docx)
On this page
- 1. Open Sales channels → Agentic
- 2. Turn on Discovery
- 3. Choose which AI channels may read your store
- 4. Check your feed, llms.txt, and the MCP endpoint
- 5. Confirm the feed URL and status on your own store host
- 6. Check product listing quality
- 7. Watch for agentic orders in your reports
- Completion checklist
- Troubleshooting
You need access to Settings in your store admin. There's nothing to connect on a third-party site for this one — Agentic publishes an open, standards-based feed that any compliant AI agent can already read; it doesn't need a partnership or developer account with any AI provider.
Open Sales channels → Agentic
In your admin sidebar, under Sales channels, click "Agentic" (`/admin/agentic`). The page is upfront about scope: Lulu publishes an open, machine-readable agent product catalog plus `llms.txt`, so assistants like ChatGPT, Google AI Mode/Gemini, Microsoft Copilot, Meta AI, Perplexity, and Claude can find, understand, and cite your products. What's guaranteed is the open feed and standards below — being placed *inside* a specific assistant's own shopping surface needs that provider's own partnership, which is outside Lulu's control.
Turn on Discovery
Toggle "Let AI channels discover my products" on. This publishes your public agent catalog feed; while it's off the feed URL returns a 404 and no product data is exposed. Leave the "Lulu Catalog — publish the open product feed" toggle on underneath it (it's the schema.org product/offer feed agents actually crawl for prices, stock and images) — turning Discovery on but Catalog off still blocks the feed.
- "Auto-enroll new AI channels as they become available" turns on newly supported channels for you automatically as Lulu adds them, so you don't have to come back and flip each one on by hand.
Choose which AI channels may read your store
Below Discovery, toggle individual channels — ChatGPT, Google AI Mode & Gemini, Microsoft Copilot, Meta AI, and "Other agents (Perplexity, Claude, etc.)" — on or off, then click Save. All five start on. Switching a channel off adds that provider's AI crawlers to your store's `robots.txt` as Disallow, and each channel's note says exactly which ones. `robots.txt` is a request that reputable crawlers follow, not a lock — the feed, `llms.txt` and the MCP endpoint stay public URLs while at least one channel is on.
- ChatGPT off: GPTBot, OAI-SearchBot and ChatGPT-User are disallowed.
- Google AI Mode & Gemini off: Google-Extended is disallowed. Googlebot keeps indexing you for Search, so AI Overviews can still cite your pages.
- Microsoft Copilot off: Microsoft has no separate AI crawler — Copilot answers from Bing's index — so your pages carry a `noarchive` header instead (Bing's signal to keep them out of Copilot answers). Bing search is unaffected.
- Meta AI off: meta-externalagent and meta-externalfetcher are disallowed. Link previews when someone shares your product still work.
- Other agents off: ClaudeBot, Claude-User, Claude-SearchBot, PerplexityBot, Perplexity-User, Applebot-Extended, Amazonbot, Bytespider and CCBot are disallowed.
- Every channel off: the catalog feed, `llms.txt` and the MCP endpoint close too, and the feed card says "Every AI channel is switched off".
- The robots.txt preview under Online Store → SEO & AI shows the effect of these switches (the bot list there shows only the blocks you set on that page).
Check your feed, llms.txt, and the MCP endpoint
The page shows your live feed URL with Copy and Open buttons — this is your store's `/feeds/agent-catalog.json`. It also points to `/llms.txt` for agent hints. Separately (not shown on this settings page, but documented for developers), the store exposes an MCP endpoint at `/api/mcp` that lets MCP-compatible agents query your catalog programmatically rather than just reading the static feed.
- The MCP endpoint is a developer-facing detail — most merchants only need to confirm the feed URL and llms.txt open correctly.
- Image links in the feed and in MCP results are full addresses on your store's domain, so agents can load your uploaded photos.
- `llms.txt` lists your collections, your newest products (while the feed is live), a link only to the policies you have written, and a Contact section built from your Contact information policy (Settings → Policies). A cleared or empty policy is never linked, and its page answers "not found".
- The feed carries the same policy links and contact details. Only text you published in the Contact information policy is used — your private store email and phone from Settings → General are never exposed.
Confirm the feed URL and status on your own store host
The sidebar's “Agent catalog feed” card shows an Enabled/Disabled pill plus your store's real feed URL — built from your store's own host (custom domain if you have one, otherwise your `.lulu.us.com` address), never the shared admin host. Use the Open button to load it directly; it should return product JSON when the pill says Enabled. The canonical URL is `/feeds/agent-catalog.json`; the same feed also answers without the `.json` suffix at `/feeds/agent-catalog`, for agents that expect either form.
- When the pill says Disabled, the feed still answers with a clear reason instead of a bare 404: a JSON body `{"error":"agentic_disabled","hint":…}`. The note under the pill says whether Discovery, the Lulu Catalog toggle, or every AI channel being off is the cause.
- The “How AI channels discover your products” card underneath explains the three pieces together: the catalog feed, `/llms.txt`, and the `/api/mcp` endpoint for MCP-aware agents.
Check product listing quality
The Listing quality card scores how "agent-ready" each active product is, from its description (100+ words is full marks), attributes (price, product type, vendor and a SKU on every variant), real images (3 or more — placeholder stock photos count for nothing), variants and stock, plus a small share for your store policies. A product with a blocking problem can't score above Fair (55), however good the rest is. A warning banner above the list counts the problems across all your active products, and each product lists its own.
- Blocking (caps the score): a sample product, placeholder images (picsum.photos and similar stock-photo services), no images, no description, or a SKU another product also uses.
- Warnings: a variant without a SKU, no brand (vendor), no category (product type), or out of stock.
- Sample products — the starter products whose description still says "Replace this sample with your own product" — are left out of the agent feed, MCP search, llms.txt and the Facebook & Instagram and Google & YouTube feeds until you rewrite or archive them. They stay on your storefront.
- The image count is exactly what the feed sends agents: still images only, not videos or 3D models.
- A policy counts as written only when it has visible text — a policy you cleared in Settings → Policies shows as Missing.
Watch for agentic orders in your reports
Any order that arrives tagged as coming from an AI agent is attributed to the `agentic` channel in your reports, the same way Point of Sale or online orders get their own channel — so once this is live you can see whether it's actually driving sales.
- An agent's checkout link opens your normal cart. Its Estimated total includes the estimated shipping shown above it (free when the cart qualifies for free shipping); checkout still prices the real rate for the shopper's address.
CHECK YOUR WORK
You’re ready when…
- "Let AI channels discover my products" and "Lulu Catalog" are both on, and at least one AI channel is on.
- Opening `https://<your-store-host>/feeds/agent-catalog.json` directly (on the store host, not the admin host) returns product JSON, not a 404.
- `https://<your-store-host>/llms.txt` opens, lists your store, and links only policies you have written.
- `https://<your-store-host>/robots.txt` disallows the crawlers of any channel you switched off.
- The Listing quality warning banner shows no sample products, placeholder images or duplicate SKUs you haven't dealt with.
IF SOMETHING DOESN’T LOOK RIGHT
Troubleshooting
The feed URL on the Agentic page shows a 404.
Check the Enabled/Disabled pill next to “Agent catalog feed” first — Disabled means Discovery or the Lulu Catalog toggle is off, and the feed answers a JSON `{"error":"agentic_disabled"}` by design, not a bug. If the pill says Enabled and it still 404s, confirm you're opening the URL shown on the page (your store's own host), not a different domain.
Discovery is on but the feed still 404s even on my store host.
Check that "Lulu Catalog — publish the open product feed" is also on underneath the master Discovery toggle — turning Discovery on alone isn't enough if Catalog is off.
I switched a channel off but its assistant still mentions my store.
The switch asks that provider's crawlers to stay away through `robots.txt`; reputable crawlers follow it on their next visit, but it doesn't remove what they already know, and it is a request rather than access control. Google AI Overviews and Copilot answer from the normal Google and Bing search indexes, which the switches deliberately leave alone so you stay in search. To close the feed, `llms.txt` and MCP entirely, switch every channel off or turn Discovery off.
A product isn't in the agent feed.
Check that it is active and published to the AI shopping agents channel, and that it isn't a sample product — a description that still says "Replace this sample with your own product" keeps it out of the feed, MCP search and llms.txt. Rewrite the description (or archive the sample) and it is included.
llms.txt doesn't link my Terms of service (or another policy).
A policy is linked only once it has content. Write it in Settings → Policies and save; an empty or cleared policy is left out on purpose, because its page answers "not found".
Will turning this on get my store featured inside ChatGPT or Gemini's own shopping results?
Not by itself. This guarantees an open, readable feed that any standards-compliant agent can use today. Being placed inside a specific assistant's own closed shopping surface needs that provider's own partnership and onboarding, which Lulu doesn't control.
What is `/api/mcp` and do I need to do anything with it?
It's a developer-facing endpoint that lets MCP-compatible AI agents query your catalog directly instead of only reading the static feed file. Most merchants don't need to configure anything for it — it's there for developers building deeper agent integrations.
My agent-ready score is low even though my products look fine.
Read the issues listed under the product. Placeholder stock photos, a SKU shared with another product, a missing description, or a leftover sample product each cap the score at 55 (Fair); a missing product type, brand or SKU, fewer than three real images, or an out-of-stock variant lower it.