Shopify’s new Sidekick Launch API lets an embedded app open Sidekick with a pre-filled prompt from inside Shopify admin. Announced October 28, 2026, it creates a useful pattern for contextual assistance: give a merchant a focused starting point without forcing them to leave the task.
The primary keyword is Shopify Sidekick Launch API. The search intent is practical and commercial: an SMB team wants to understand the platform change, the operational risk, and the implementation path before investing.
What changed
An embedded app can call shopify.sidekick.launch with a message. Sidekick opens and the prompt is sent as the merchant’s message. Shopify’s example is an Ask Sidekick button next to a report.
The launch capability is a doorway into a conversation, not proof that an action was completed. App teams still need to decide what context belongs in the prompt, what data should stay out, how the user understands the transition, and where application-specific support remains necessary.
Why this matters to a growing business
Merchants often need help at the moment they see a report, exception, or unfamiliar setting. Generic documentation makes them restate the context. A scoped prompt can reduce that setup work and make the request clearer.
Poorly designed prompts can expose sensitive details, misrepresent system state, or imply that Sidekick has authority it does not have. Contextual help should preserve merchant agency and offer an application-owned fallback.
Use Respond → Qualify → Act → Handoff
Respond
Label the button by outcome, such as Explain this report or Help investigate this issue. Tell the merchant that Sidekick will open with a prepared prompt, and let the merchant decide whether to continue.
Qualify
Determine which page, report, filter, or error is relevant. Include only the minimum safe context. Do not embed secrets, unrestricted customer data, or claims that the application has not verified.
Act
Build a concise prompt with the observed state, the merchant’s likely question, and clear boundaries. Keep irreversible or consequential changes behind the application’s normal confirmation and permission controls.
Handoff
If the issue requires application support, provide an owned route to a person with a structured summary. Contextual AI should not become a dead end when the app, billing, or integration is the actual source of truth.
A practical SMB example
Consider a small inventory app that shows a report with several products missing a supplier mapping. Beside the report, it offers Explain these exceptions in Sidekick.
The prompt names the report and asks for an explanation of the visible categories without including customer records or credentials. A separate Contact support path carries the report ID and diagnostics to the app team. This hypothetical design distinguishes general assistance from application accountability.
Implementation guidance
1. Define the event and its owner
List the moments where merchants repeatedly ask for explanation, interpretation, or next-step guidance. Avoid adding an AI button to every screen. Assign ownership for the prompt and support fallback.
2. Design the data contract
Define allowed fields, prohibited fields, prompt version, page context, merchant-visible preview, and correlation ID. Keep sensitive diagnostics in the application’s support system.
3. Test normal, boundary, and failure states
Test missing context, stale reports, different permission levels, very large values, repeated launches, unsupported requests, and Sidekick unavailability. Verify that the app remains usable without the launch.
4. Keep humans in consequential decisions
Require the merchant to confirm consequential actions through the authoritative interface. Sidekick can help explain or plan, but app-specific writes should respect existing roles, logs, and approvals.
5. Connect the experience to the wider system
Connect repeated help themes to product analytics, documentation, onboarding, and support. DIGIMAR can help develop the embedded experience and integrate follow-up into a configured workflow.
Measurement guidance
Track button visibility, launches, abandoned launches, support deflection, subsequent task completion, support escalation, resolution time, and feedback. Do not treat a launch as a solved problem.
Review prompt outcomes qualitatively. If merchants repeatedly ask the same follow-up, the prompt may lack context or the underlying UI may need improvement. Use AI-help data to improve the product, not just the prompt.
Common mistakes to avoid
- Sending secrets or unnecessary customer data in the prompt.
- Presenting a conversation launch as a completed action.
- Using vague labels such as Ask AI everywhere.
- Removing the application-owned support path.
- Measuring launches instead of resolved merchant tasks.
A 30-day rollout plan
Week 1: map
Document the current customer path, systems, owners, permissions, and failure points. Write the business rule in plain language before choosing automation.
Week 2: build
Implement the smallest useful version in a development or test environment. Add structured logs, explicit status values, duplicate protection, and a safe way to stop or reverse automated actions.
Week 3: test
Run representative cases, missing-data cases, repeated events, provider errors, and human-escalation scenarios. Confirm that every outcome has an owner and that customer-facing language stays accurate.
Week 4: launch and review
Release to a limited segment, review evidence daily, and expand only after the team understands exceptions. Record decisions so later changes do not erase the original business intent.
How DIGIMAR can help
DIGIMAR can connect this work to AI automation and integrations and web development. Where Maya is a genuine fit, the managed AI Customer Response System can be configured to support website chat, phone answering, SMS communication, qualification, appointment requests or configured bookings, structured summaries, follow-up, CRM or workflow handoff, and human escalation. The exact behavior depends on business rules and integrations.
Choose a clear next step
Choose one high-friction report or exception page. Draft the prompt, review its data boundaries, preserve a support fallback, and run a small usability test before wider release.
The durable goal is not to add technology for its own sake. It is to reduce ambiguity, preserve context, and make the next action visible. Every inquiry answered. Every opportunity moved forward.
Primary source: Start Sidekick conversations from your app, October 28, 2026.