How to choose an AI shopping assistant for Shopify: 20-point checklist

What should you look for in an AI shopping assistant for Shopify?

Look for five things, in this order: answers that are accurate to your live store, honest behavior with a human always reachable, a light footprint on your storefront, minimal access to customer data, and a pricing unit and measurement method you understand. Features like bundles or proactive greetings matter only once those five hold.

Most demos look good because they use friendly questions. The checklist below is built around the questions that break assistants in production: a sale price that changed this morning, a discount code that does not exist, a shopper writing in Arabic, a phone on a slow network. An AI shopping assistant is only useful if it survives those.

How do you check that the assistant's answers are accurate?

Ask where every number comes from. Prices, compare-at prices and stock should be read live from Shopify for each answer, not copied into a knowledge base the night before. Then ask what happens when the model writes a wrong figure anyway: is the draft checked before the shopper sees it, or streamed straight to the screen?

Streaming looks fast, but it means a sentence is visible before anyone can check it. A guarded assistant generates the whole reply, compares each figure with live data and approved facts, and only then sends it. Run your own test: change a price in your admin, then ask the assistant about that product within a minute. More on why this fails in AI chatbot hallucinations: prices and stock.

• 1. Live prices and stock. Read per answer from the store, never stored.
• 2. Figure checks before sending. No streaming of unchecked text.
• 3. Approved store facts. Shipping thresholds, return windows and codes come only from facts you approved.
• 4. No invented codes or dates. Ask it for a discount code you never created; it should say none exists.
• 5. Right variant. Product cards show the variant the shopper asked for, with the live price.

Does the assistant disclose it is an AI and hand off to a human?

It should do both by default, and you should not be able to switch disclosure off. Shoppers deserve to know they are talking to software, and some laws now require it. A clear one-tap route to a person on your team protects you on complaints, exceptions and anything the assistant should not decide alone.

In the EU, Article 50(1) of the AI Act requires AI systems that interact with people to be designed so people are informed they are dealing with an AI, unless that is obvious; the provision applies from August 2, 2026 (EU AI Act, Article 50). If you sell to EU shoppers, check with counsel how it applies to you. Our guide to AI chatbot disclosure and human hand-off covers the wording.

• 6. AI disclosure. The window is labelled as AI; it never claims to be human.
• 7. Human hand-off. Always offered, with your business hours and a fallback.
• 8. No pressure tactics. No false urgency, fake scarcity or fake social proof; at most one offer, and none to an upset shopper.
• 9. Safe topics. It refers medical, legal and safety questions to the label or a professional.

Does it speak your shoppers' languages, including right-to-left?

Check two things: whether it answers each shopper in the language they write in, and whether the chat layout actually flips for right-to-left languages such as Arabic, Hebrew and Urdu. Translating text is easy; mirroring the window, price placement and punctuation is where widgets can break. Test it on a phone, not only on desktop.

Ask also which language it uses by default and whether you can change that. If your catalog is in English but a shopper writes in Spanish, the assistant should answer in Spanish while keeping product names exact. See multilingual customer support for Shopify.

• 10. Reply language. Matches the shopper, with a default you control.
• 11. RTL layout. Mirrored window and correct number direction for Arabic, Hebrew and Urdu.

Will it slow down your store or break your theme?

It should not edit your theme code, and the script that loads on every page should be small. Shopify's Built for Shopify standard requires storefront apps to use theme app extensions, which are removed automatically on uninstall, and Shopify suggests keeping each extension's compressed JavaScript to about 10 KB.

Those rules come from Shopify's Built for Shopify requirements, which also say an app must not reduce the storefront Lighthouse performance score by more than ten points, and from the theme app extension configuration docs, which list a suggested 10 KB limit for compressed JavaScript referenced by the extension. A sensible design loads a tiny launcher and fetches the chat window only when a shopper opens it. Test it yourself with the before-and-after method in will a chat widget slow down your Shopify store?.

• 12. Theme app extension. No pasted snippets in theme.liquid; an app embed you switch on in the theme editor.
• 13. Small loader, lazy chat. The full UI loads on open, not on page load.
• 14. Matches your theme. Colors and fonts follow the theme instead of a generic bubble.

What customer data does an AI shopping assistant need?

Usually much less than it asks for. Shopify splits protected customer data into levels: Level 1 excludes names, addresses, phone numbers and emails, while Level 2 includes them and adds stricter requirements. An assistant that recommends products and answers policy questions rarely needs Level 2. Ask which level and which scopes it requests, and why.

Shopify's protected customer data documentation lists the Level 1 duties (process only the minimum data, state why, set retention periods, encrypt at rest and in transit) and the extra Level 2 duties such as access logs and an incident response policy. Also ask whether your data trains anyone's AI models: Shopify's API License and Terms of Use restrict using API and merchant data to train or improve AI systems.

• 15. Protected customer data level. Level 1 is enough for most selling; ask for the reason if it wants Level 2.
• 16. Retention and deletion. How long transcripts are kept, and whether shoppers can wipe a chat.
• 17. No training on your data. In writing, for the vendor and its AI model providers.

How do AI shopping assistants charge: per conversation or per resolution?

Most charge by usage, and the unit matters more than the headline price. Some count a conversation once the AI replies at least once; others count a resolution, meaning a request closed without a person. Before you compare prices, find out exactly what is counted, what happens at the limit, and whether overages are billed automatically. For sourced side-by-sides of specific apps, see the Shopify chatbot comparison.

Two published examples, checked October 20, 2026. Gorgias bills its AI Agent per automated interaction, defined as a customer request fully resolved without a human agent, at $0.90 on annual or $1.00 on monthly contracts, with $1.50 for overage (Gorgias pricing). Tidio sells Lyro as a monthly conversation quota and counts a conversation once it has at least one AI reply, however many replies follow (Tidio pricing). Neither model is wrong; they reward different things, so estimate your own volume against each.

• 18. Clear billing unit and limits. What counts, what does not, and what happens at the cap.

How will you know the assistant is actually earning its fee?

Ask whether it can measure incremental revenue with a holdout: a small random share of visitors who do not see the assistant, compared with everyone else. Revenue that a dashboard simply attributes to chat includes shoppers who would have bought anyway, so it overstates impact. A real trial period lets you run that comparison on your own traffic.

Set up the test before you go live, decide the run length in advance, and do not stop it early because the first week looks good. Our holdout test guide explains the setup.

• 19. Holdout measurement. Lift against a control group, not only attributed revenue.
• 20. A trial on your own store. Long enough to run a measurement and read real transcripts.

The 20-point AI shopping assistant checklist

Use this table in vendor calls and during the trial. Mark each line pass or fail, and treat any fail in the first nine lines as a reason to stop: those are the points where a wrong answer reaches a real shopper. The rest decide speed, data risk and whether the cost is predictable.

How Tycoon answers this checklist

Tycoon is an AI sales associate for Shopify, in early access, built around the first nine checks. Prices and stock are read live for every answer, every draft is checked whole before sending, and it may only state shop facts the merchant approved. It labels itself as an AI assistant and always offers a person from your team.

It works in 16 languages with right-to-left layouts for Arabic, Urdu and Hebrew, appears through a theme app extension with a loader designed to stay around 10 KB, and requests protected customer data Level 1 without names, emails, phones or addresses. A conversation is one shopper in a 24-hour window with at least one AI-written reply, up to 10 replies; at the limit it keeps working in a lighter mode with no overage fees. Measurement uses a holdout: one 30-day lift test on Growth, continuous on the Tycoon plan. Details are on pricing and security.

Frequently asked questions

Sources

1. Shopify.dev: Protected customer data · checked October 20, 2026
2. Shopify.dev: Built for Shopify requirements · checked October 20, 2026
3. Shopify.dev: Theme app extension configuration · checked October 20, 2026
4. Shopify API License and Terms of Use · checked October 20, 2026
5. EU AI Act, Article 50: Transparency obligations · checked October 20, 2026
6. Gorgias pricing · checked October 20, 2026
7. Tidio pricing · checked October 20, 2026