Shopify Knowledge Base 2026: How to Review Store Information for AI Agents?

A return-policy answer looks outdated, but no one knows whether the old app or the new admin entry is authoritative.

Fastest fix: For Shopify Knowledge Base 2026, verify every fact against its current source page and policy, then edit, leave unpublished, or publish it. If you used the old app, check migration separately. Publishing a fact does not guarantee an AI agent will display or use it.

Shopify store owners: decide which store information AI agents may use.
Merchandising and content teams: confirm return, shipping, and contact details against current pages.
Customer support and AI project leads: test answer boundaries and assign ownership for updates.

Last updated October 28, 2026; checked against Shopify’s admin migration announcement and Knowledge Base help documentation.

For store owners: distinguish store facts from product data

Shopify Knowledge Base is for store information that can help answer customer questions, such as contact details or gift services. It is not a substitute for your product catalog. Shopify documents product information for Agentic storefronts separately in its guidance on products in those storefronts.

That distinction affects what you should review. A product title, variant, or price belongs in product data. A statement about how customers contact your team, whether gift wrapping is offered, or how a store policy works belongs in store information—provided the statement is accurate and approved.

Before editing facts, make a short ownership list. For each topic, name the person accountable for the underlying policy, not just the person who can access the admin.

• Returns and exchanges: assign the owner who approves eligibility, exceptions, and the current process.
• Shipping and delivery: assign someone who can confirm service areas, handling rules, and any stated timeframes.
• Contact and service: assign the person responsible for support channels and business contact details.
• Gift services or other store-specific information: ask the relevant team whether the offer is active and what limits apply.

Use this list to define the boundary of what an agent can safely answer. If the team cannot confirm a policy or its exceptions, do not turn an uncertain internal assumption into a customer-facing fact.

For old-app administrators: compare migration states

Shopify announced that Knowledge Base moved into the admin on October 26, 2026, under Sales channels > Agentic > Knowledge base. The old standalone app is scheduled to stop being available after January 15, 2027, according to Shopify’s Knowledge Base migration and app-availability guidance.

Those dates make migration a record-checking task, not a reason to assume your information is already complete. There may be two different kinds of content to inspect:

• Manually edited information from the old app: compare each item you deliberately maintained with what now appears in the admin.
• New facts drafted from published store pages: check the wording and source page, then decide whether the draft reflects current policy.

Do not treat a newly generated draft as proof that an older manually maintained answer transferred correctly. Conversely, do not keep duplicate facts just because the old and new views look different. Check what is present, what is approved, and what still needs a responsible owner.

Use Shopify’s migration guidance to verify the current rules before deciding that an item has moved. Then use the admin to identify content you need to preserve. Do not assume that every old-app fact migrates in the same way or that a particular item limit applies to every store. If the official guidance states a limit relevant to your account, compare your saved content against it and identify any items that need attention before the old app is no longer available.

Migration check: Keep a copy of old content that still expresses an approved policy until you have confirmed its replacement in the admin. If no matching fact or source page exists, recreate or revise it only after the policy owner confirms the wording.

For content operators: trace each fact to its source

A fact without a reliable source is difficult to maintain. For each entry, open the linked store page and compare its current wording with what the fact says. Check the actual policy, not just whether the page loads.

Look closely at these common failure points:

• Returns: Does the linked page explain the current process and any eligibility conditions? A simple “returns accepted” statement may be misleading if exceptions matter.
• Shipping: Does the source explain where the store ships and any conditions that affect delivery? Do not add a promise that the page or policy does not support.
• Contact details: Do the listed details still reach the team responsible for the issue? An old address or inactive contact route can send customers into a dead end.
• Service descriptions: Are offers such as gift services still available, and are their limits clear?
• Conflicting wording: Does another published page state something different? Resolve the conflict at its source before approving a summary.

If the source page is outdated, fix the page or confirm the policy with its owner first. If the fact itself is wrong but the page is current, edit the fact to match. If neither the policy nor its owner can be confirmed, leave it unpublished or remove it from the set under review. Do not publish an AI-drafted statement just because it sounds plausible.

Keep store-level facts separate from catalog details. This is especially important when multiple people update product listings and policy pages. A clear source path helps the next reviewer see whether a statement is an approved policy, a product attribute, or an assumption that still needs confirmation.

For support leads: test what the fact can safely answer

A clean source page is necessary, but it does not prove that the fact supports a clear customer answer. Use questions your support team actually receives. For example, test a basic return request, a delivery question, and a request to contact the store. These are test prompts, not a forecast of how any particular AI channel will respond.

For each prompt, record:

• What the approved fact supports directly.
• What it does not establish, such as an exception or a promise.
• Whether the source page gives enough context to answer without guessing.
• Who should decide if the answer is incomplete or disputed.

A safe answer should stay within the policy. If a question depends on an exception the fact does not cover, the answer should not invent one. Mark the gap and send it to the policy owner. Then update the source page or fact only after that owner confirms the change.

Shopify provides separate technical material for its store question-and-answer interface. That documentation can help teams understand an interface, but it is not evidence that every customer-facing AI channel will use a particular fact in the same way. Keep your acceptance notes channel-specific.

Quick answers about access, migration, and publication

Where do you open Knowledge Base?

Use Sales channels > Agentic > Knowledge base in the Shopify admin. Shopify’s announcement confirms the move. If the menu is not visible, check your store selection and access level before treating the missing entry as a migration failure. Shopify’s admin permissions overview explains how users and permissions relate to admin access.

What is the pre-publication check?

Open the source page for each fact and confirm that the wording, policy, and any relevant conditions still match. If the page is missing or conflicts with current practice, correct the source or fact before publishing. Treat a generated draft as a proposal for review, not as approved policy. A reviewer should be able to point to the page and the owner who confirmed the underlying rule.

Does old-app content move automatically?

Check the old content against the facts shown in the admin rather than relying on a blanket assumption. Shopify’s migration guidance is where to verify the current rules. Record any missing content, confirm it with its policy owner, and preserve approved information before the old standalone app stops being available after January 15, 2027.

Does publication mean an AI agent will use the fact?

No. Publication is a content-approval state; it is not proof that a channel is available, has retrieved the fact, or will display it in response to a particular question. Record these as separate checks. Do not describe publishing as a guarantee of visibility, ranking, answer accuracy, or sales.

For the Agent project lead: choose the right state

After review, make an explicit decision for each fact. A draft should not move forward simply because the team has finished reading it.

• If the fact matches a current source and its owner approves it, publish it.
• If the wording is incomplete but the policy is confirmed, edit it and repeat the source check before publishing.
• If the source is stale or conflicts with current practice, correct the source or leave the fact unpublished until the owner resolves the conflict.
• If the team cannot confirm the policy, do not publish a guess. Escalate the question or remove the unsupported fact.
• If old-app content appears missing, compare records and rebuild only the information the owner confirms should remain.

Keep these states distinct in team conversations: fact published, channel available, and AI agent used the fact. They are separate questions. Shopify’s Agentic storefront documentation describes the broader channel context, but it does not make a published fact a promise of exposure or a particular answer.

For the handoff owner: keep an audit trail

A review that cannot be handed over will become a repeat review. Store a compact record for each fact so another person can understand why it was approved and what needs attention next.

Include:

• The fact text or a stable reference to it.
• The source page checked.
• The policy owner and reviewer.
• Its current state: draft, under review, published, or removed.
• The review date and any unresolved question.
• The reason for the last edit, if the change affects customer expectations.

Use access controls that match each person’s role. Shopify’s admin user and permission guidance is relevant when several people need to review the backend. Avoid sharing credentials just to speed up a content check; assign access through the store’s normal permission process.

If reviewers work across locations or cannot use the same workstation, decide whether a controlled macOS environment is genuinely needed for the admin review. It may help standardize access to a Mac-based review workflow, but it does not replace Shopify permissions, policy approval, or channel testing. For a team that needs that setup temporarily, you can compare a remote Mac option for a U.S. location with your existing devices. If the review is a recurring operational need, check VMSPIN plans against the team’s access, security, and support requirements. A long-term, heavy workload or a need for physical ports may make an owned Mac a better fit.

The review sequence to use this week

Start with the policy owner and source page, not the draft’s fluency. Work through the following sequence before making a fact available:

1. Confirm the current entry point. Open the Knowledge Base in the admin and make sure you are reviewing the intended store.
2. List facts by owner. Assign responsibility for returns, shipping, contact information, and other store-level topics.
3. Separate old content from new drafts. Compare manually maintained old-app information with facts drafted from current published pages.
4. Open every cited source. Check the live wording, policy, exceptions, and contact routes against the fact.
5. Test real support questions. Record what the approved information can answer and where it must stop.
6. Choose a state. Edit, leave unpublished, remove, or publish only after the relevant owner confirms the content.
7. Record the handoff. Save the reviewer, source, status, date, and any outstanding decision.

Shopify Knowledge Base 2026 is best treated as a store-fact review surface, not as a shortcut to guaranteed AI visibility. If your current process relies on scattered notes, stale policy pages, or one person’s memory, fix that source-of-truth problem before adding more facts. A remote Mac may suit a team that needs a temporary, controlled backend review environment; it is not necessary for every store, and it cannot decide policy or guarantee an AI response. Review your source pages first, then choose the working environment that fits the team.