Shopify Knowledge Base Moves Into the Admin: Agentic Commerce Readiness (2026)
Shopify now keeps the facts AI agents read about a store inside the admin, and the standalone Knowledge Base app goes away on January 15, 2027. Here is what changed, what merchants have to do, and how Shopify apps and agencies can turn it into useful conversations.

On October 26, 2026 Shopify moved Knowledge base into the Shopify admin at Sales channels > Agentic > Knowledge base, replacing the standalone Knowledge Base app, which Shopify says will no longer be available after January 15, 2027. Knowledge base is where a merchant reviews, edits and publishes the non-product facts AI agents get about the store, such as policies, contact details and services. Most merchants have not opened it yet, which makes it a concrete, dated reason for a Shopify app or agency to start a conversation.
Snipe Outbound runs outbound for software companies and agencies that sell to ecommerce brands, so we read changes like this for the selling angle. Every product fact below comes from Shopify's changelog post or its Help Center page on the knowledge base, both checked November 3, 2026.
What did Shopify change?
- New location. Knowledge base is built into the admin at Sales channels > Agentic > Knowledge base.
- The app is being retired. The standalone Knowledge Base app "will no longer be available after January 15, 2027," per the changelog.
- Drafted from the storefront. Shopify drafts facts from the store's own published storefront pages, and each fact links back to the page it came from. Per the changelog, nothing Shopify drafted reaches AI agents until the merchant publishes it.
How does the knowledge base work?
According to Shopify's Help Center, the published knowledge base is the source of truth AI agents use for non-product information about a store. When a shopper asks about a store in an AI chat, the agent can query the knowledge base through the Storefront MCP. Points that matter in practice:
- The published knowledge base is publicly available through the Storefront MCP and can be used by AI channels including Shop and third-party platforms.
- Facts generated from a storefront scan stay unpublished until the merchant publishes them. Facts linked to admin settings, such as store policies and contact details, are published when Shopify creates them.
- The Help Center says the knowledge base can hold a total of 1,000 facts, including generated and additional facts.
- Shopify rescans the store and offers new findings as suggested updates the merchant accepts or rejects. Shopify says it never changes knowledge base content on the merchant's behalf.
- Password-protected stores only show admin data, because the scan cannot reach the storefront. Custom storefronts and stores on the Agentic plan can add only additional facts.
- Facts are grouped into categories such as About the Store, Brand and Identity, and Company policies, with Shopify flagging missing policies.
Which merchants have to act before January 15, 2027?
| Merchant situation | What Shopify says happens | Action needed |
|---|---|---|
| Used the Knowledge Base app, fewer than 1,000 facts | Facts the merchant wrote or edited moved over and were published | None required; review what agents now see |
| Used the app, more than 1,000 facts | Facts did not move over; Shopify emailed these stores | Add the facts to keep before January 15, 2027 |
| Never used the app | Facts are drafted from the storefront and stay unpublished | Nothing required, but drafted facts reach agents only once published |
The third row is the large one. A store that never touches the page still has its admin-linked policies and contact details published, while everything Shopify drafted from its storefront stays unpublished.
The two details that create the opening
Two lines in Shopify's own documentation do all the work here, and most vendors will read past both of them.
- Policies publish on their own. Facts linked to admin settings, such as store policies and contact details, are published when Shopify creates them. So a merchant who opens the page sees content already live. The screen looks handled.
- Scanned facts do not. Facts generated from a storefront scan stay unpublished until the merchant publishes them. That is the substantive material, and it sits in a queue nobody has opened.
Put together: "configured" and "actually answering correctly" are two different states, and from inside the admin only the first one is obvious. That gap is the thing worth selling against, and it has a date attached to it.
What does an agentic commerce readiness check look like?
A short, honest check a merchant (or the agency or app helping them) can run in the admin:
- Open the page. Sales channels > Agentic > Knowledge base. Note how many drafted facts are unpublished.
- Fix the sources first. Every fact links to a storefront page. If the shipping or returns page is vague, the drafted fact will be too. Edit the page, then request a rescan.
- Check missing policies. Shopify flags missing store policies; fill them in the admin.
- Publish deliberately. Publish the facts you would be happy to see quoted by an AI assistant, category by category or field by field.
- Add what the scan cannot see. Use additional facts for details that live nowhere on the storefront, within the 1,000-fact limit.
- Review suggested updates on a schedule. Accept or reject what each rescan proposes.
We are not claiming this changes how often a store appears in AI answers; Shopify does not publish that, and neither do we. It does change whether the store's own facts are available to agents at all.
Who this is a wedge for, and who should skip it
A dated platform change is one of the better reasons to email a merchant: it is specific, it is verifiable, and it is not about you. But it is not a wedge for everyone, and the vendors it does not fit will burn a quarter finding that out.
It is a reasonable wedge if your product touches what a shopping agent gets asked about.
- CRO and site agencies. Drafted facts come from storefront pages, so weak policy, shipping and FAQ pages produce weak facts. Offer one observed gap on their store with the fix.
- SEO and AI-search agencies. The knowledge base is a first-party channel to AI agents through the Storefront MCP. Offer a review of which facts are published and which are missing.
- Returns, subscriptions, sizing, warranty and support-automation apps. The facts agents read about those topics now sit in a merchant-controlled page. Help merchants keep them accurate.
- Product-content and PDP tooling, and 3PLs with policy-heavy shipping rules. The rules you already manage are exactly the rules an agent will be asked to repeat.
- Former Knowledge Base app users with more than 1,000 facts have a hard January 15, 2027 deadline and are the most time-sensitive segment.
Skip it if the connection to an agent's answer is a stretch. Ad-buying agencies with no on-site surface, pure creative shops, and apps whose contract value cannot carry a project sale will all struggle to get past the first reply. Saying so is not modesty. A merchant can tell within one sentence whether the person emailing them has a reason to care about their returns policy, and the vendors above cannot fake one. If you are weighing whether this is your angle at all, our guide to the best marketing agency options for Shopify apps covers the wider channel question.
What agents get asked, and where stores answer wrong
The useful way to frame readiness for a merchant is not "your knowledge base is incomplete." It is the list of questions a shopping agent will be asked about their store, and which of them the store's policy pages alone cannot answer:
- Does this store ship to a given country, and what does that cost?
- What is the return window, when it differs by SKU, by sale status or by bundle?
- Can a subscription be paused or skipped, and on what notice?
- Which size should a specific body or room measurement buy?
- Is this product eligible for the bundle promotion the shopper just read about?
- What does the warranty cover, and who honours it?
Each of those has a wrong answer that a generic shipping page will happily supply. The loss is not a ranking position; it is an order that an agent routed somewhere else because it could not confirm a detail. That sentence is the one the outbound message is built on. We are not going to put a number on how often it happens, because nobody has measured it and an invented figure is worse than no figure.
How to find the stores where this is true
The honest constraint first: you cannot see from outside whether a merchant has published their facts. The admin is the admin. Any message that claims to know their knowledge base state is a bluff, and merchants who have opened the page will catch it.
What you can observe is which stores have the kind of catalogue and policy complexity that makes a wrong agent answer expensive. That is the segment, and it is buildable from signals anyone can check:
- Platform and plan. Shopify versus Plus versus a custom storefront, which also determines what the merchant can do in the knowledge base at all.
- Revenue band and order volume. The cost of a wrong answer scales with both.
- Apps installed and, more usefully, apps removed. Subscriptions, returns, sizing and helpdesk apps tell you which rules are complicated enough to need software.
- Whether policy pages exist and how specific they are. A single 90-word returns page on a 4,000-SKU catalogue is the whole pitch in one observation.
- Catalogue complexity. Variant counts, bundles, made-to-order items, regional pricing.
- Ad activity and hiring. Both tell you whether anyone is paid to care this quarter.
This is a process, not a tool. You run it the same way every week: pull the store universe, layer the fit signals, verify the observable fact per store by looking at the storefront, and keep only the ones where you can point at something. Our guides on finding Shopify stores by app or tech stack and building an ecommerce lead database cover the mechanics.
The five-part campaign frame
Trigger, segment, observable, message, ask. No templates here on purpose: the frame travels, the swipe copy does not.
- Trigger. The dated platform change. January 15, 2027 is the reference point, and it belongs in the campaign plan rather than in the first line of every email.
- Segment. Stores where a wrong agent answer costs real money, selected on the signals above. Keep the segment small enough that you could describe any store in it from memory.
- Observable. One verifiable fact per store, from their storefront. Not their admin, not their traffic, not a guess at their revenue.
- Message. Written as the store's outcome, in their words. Buyers who use AI assistants to research vendors describe their own situation rather than searching a product category: Empact's buyer research found 61% of buyers describe their use case in the prompt and 45% add a constraint, and adding a single constraint changed the first brand ChatGPT named in 21 of 30 categories. The same logic holds in a cold email: the specific situation gets the reply, the product category does not. Our guide to cold email for ecommerce agencies goes further on the offer, and how to sell to Shopify merchants covers what merchants actually respond to.
- Ask. A January slot. In November and December, stores are in and then recovering from BFCM, and they will not start a project. That is our read from selling to merchants rather than a published statistic, but it is consistent enough that asking for a December kickoff mostly buys you a polite no. Ask for the thing they can say yes to.
Sequencing: November, December, January
- November. Build the segment and write the message. Sell the project internally first: decide what you are actually delivering, what it costs you to deliver, and who does it.
- December. Warm the list and publish the proof asset, so that the January conversation has something to point at. Book January slots now, while calendars are empty.
- January. Run it, with the January 15 cutoff as the dated reference for the segment it genuinely applies to. If you sell to Shopify apps rather than stores, how to get demos for your Shopify app covers the same sequencing from the app side.
What not to do
- Do not imply Shopify requires this. It does not. Only stores that used the old app with more than 1,000 facts have a required action.
- Do not quote a number for how many stores are unconfigured. Nobody has that data, including us.
- Do not send an "audit" of a knowledge base you cannot see. Audit the storefront instead, which is public, and say that is what you did.
- Do not lead with the feature name. Most merchants have not heard of it yet. Lead with the question their customers are asking an AI assistant.
Where this sits next to your own AI-search visibility
This is a store-side surface. It is not the same problem as whether an AI assistant names your app when a merchant asks for a recommendation, and the two should not share a budget line.
They do share a clock, though, and the clock is not generous. AirOps tracked AI answers through 2026 and found that on 62.8% of questions, the brand named most in January and February was still the one named most in September, while brands that first appeared in Q3 picked up roughly a tenth of the mentions incumbents got. The number of brands named in an average answer fell from 7.7 to 5.3 over the same period. Fewer slots, and the ones that exist are sticky.
Which is why the store-facing campaign and the visibility work are different jobs on different timelines. If your own inbound is the part that is shrinking, that is the subject of a separate piece: Shopify app outbound when AI search cuts your inbound.
Frequently asked questions
Where is the Shopify knowledge base now?
In the Shopify admin at Sales channels > Agentic > Knowledge base, per Shopify's October 26, 2026 changelog.
When does the Shopify Knowledge Base app stop working?
Shopify's changelog says the standalone Knowledge Base app will no longer be available after January 15, 2027.
Do merchants have to do anything?
Only stores that used the app with more than 1,000 facts must add the facts they want to keep before January 15, 2027. Other stores are not required to act, but storefront-drafted facts reach AI agents only after the merchant publishes them.
How do AI agents read the knowledge base?
Shopify's Help Center says agents can query it through the Storefront MCP, and that the published knowledge base can be used by AI channels including Shop and third-party platforms.
How many facts can a knowledge base hold?
Shopify's Help Center says a total of 1,000 facts, including generated and additional facts.
Do store policies publish automatically?
Per Shopify's Help Center, facts linked to admin settings, such as store policies and contact details, are published when Shopify creates them. Facts generated from a storefront scan stay unpublished until the merchant publishes them, which is why a store can look configured while its substantive facts are still in a queue.
Is agentic readiness worth selling to stores as a service?
It depends on what you already sell. If your product touches what a shopping agent gets asked about, such as returns, shipping, subscriptions, sizing or warranty, it is a credible reason to open a conversation. As a standalone audit it is thin, because the part you would audit is inside an admin you cannot see.
Where should you start?
If you sell to Shopify merchants, this is a timely reason to reach the right ones. On a call, Leon walks through a Market Map of fitting stores, the signal we would use, and the first offer we would test. No promised meeting counts, because they depend on your contract value, segment size and offer.
Book 30 minutes with Leon to plan outreach around agentic commerce.