Ask Friendly is our own AI concierge, built as a Shopify app. It sits on every page of the store, answers shopper questions in a few seconds using only facts we have approved, and points people to the right product. Behind it is an admin where we approve knowledge, read every conversation, run the support inbox, moderate reviews and see how shoppers found us.
A shopper's question becomes a checked, sourced answer in a few seconds. Nothing reaches the shopper until every stage has passed.
Left: a shopper asks whether creatine chews will lower their blood sugar while on metformin. The medical guardrail fires: Friendly says it can't speak to a medication, suggests a doctor or pharmacist, and shares only the facts we have approved. Right: when a product fits, Friendly shows a card with the price and a Buy Now button.
Nine tabs: Home, Support, Reviews, Analytics, Conversations, Knowledge, Products, Survey and Settings. Screens marked "demo store data" were captured on our test store so no shopper's details appear.
Who is chatting right now, today's and this week's conversations, AI spend against the daily cap, and a callout for anything waiting on the team, such as open support cases or cards to approve.
Live admin, 10 Nov 2026
Conversations, reply speed, model spend, revenue from orders that followed a chat, how often the guardrail fired and how often Friendly had no answer, each compared with the period before.
Demo store data, 21 Oct 2026
Each reply shows its speed, cost, the knowledge it used and whether a product card was shown. Filters find guardrail hits, misses, slow or costly replies and chats that led to an order. Shoppers are shown as signed in or visitor, never by name.
Demo store data, 21 Oct 2026
Every card moves draft, submitted, approved, published. The "Waiting for you" panel lists each card in progress, so nothing sits unnoticed.
Demo store data, 21 Oct 2026
In progress, published and shipped, each with its fingerprint. Only a named person can approve, every step is written to an audit log, and if approved text is changed later, Friendly refuses it and answers from the last approved version.
Demo store data, 21 Oct 2026
One question on the Thank you page, joined to the order it came from. Each linked order is also tagged in Shopify ("Heard about us: …"), so marketing can see which channels bring real buyers.
Live admin, 10 Nov 2026 · revenue blurredMail to team@bloodsugarfriendly.com syncs in within minutes and becomes cases with owners, due times and reaction clocks. Claude drafts a reply; a person sends it from the Support tab.
New reviews sync from Judge.me. Clear ones publish, anything worth a look is held, and a monthly check makes sure low-star reviews are not held more often than high-star ones.
Answers, linked orders and revenue by channel, for 7, 30 or 90 days. It sits beside the UTM attribution in BSF Social.
These are rules in the software, not good intentions. They are why Friendly can talk about health-adjacent products without making things up.
The shopper's browser sends only the new question. Past replies are rebuilt from our database, so nobody can fake what Friendly said.
Order details unlock through Shopify's own sign-in, or the order number with the checkout email. A name typed into chat unlocks nothing.
Product content, prices and stock come from Shopify, which Relay keeps current. Friendly reads them and never changes a product.
Knowledge moves draft, submitted, approved, published. Only a named person can approve, and every step is logged.
Each approved card is fingerprinted. A changed card is refused until it is approved again.
Fast enough to help, careful enough for a metabolic-health brand.