What's new in sqrly

Now and then, after it has given a real answer, the digital twin asks: “Was this helpful?” One tap, thumbs up or down, in about one conversation in three and never more than once, so shoppers aren’t pestered while they decide. A thumbs down asks one optional question: what was missing? The same question appears on the store’s “Ask our expert” page, and AI agents shopping for someone can pass on the shopper’s verdict too. Their ratings are counted separately.

The merchant sees it where it matters. Home shows how helpful shoppers found the twin this week and how that compares with the week before. Conversations gets a new list, “Answers that missed”: each one with the shopper’s question, the twin’s answer and what was missing, and an Improve this answer button right there. The twin even starts the fix: it rewrites its reply into a lasting answer, works in what the shopper said was missing, and leaves out what changes, like prices and stock, which it always takes live from the store. Where only the merchant knows the answer, it asks for just that, in its own field: a few words are enough. Approve or edit it once, and the twin uses it for every similar question from then on. This is the first piece of something bigger: a twin that keeps getting better from every conversation.

Feature 10 of Shiptober. We set out to ship a new feature for the sqrly Shopify app every day until October 20, and we’re running ahead of that. Want it in your store? Email us about becoming a pilot shop.

Some AI agents don’t read and click their way through a store anymore. With a new web standard, WebMCP, a store tells a visiting agent which actions it offers, and the agent uses them directly. Every Shopify store already offers the basics this way: search the catalog, look up a product, add to cart, check out. What was missing is the part that makes a good store a good store: someone who knows the products.

Now stores running sqrly offer that too. Next to Shopify’s actions, an agent finds “ask the product expert”: it asks the store’s digital twin which product fits, which size, what goes with what, and gets the answer together with the exact products to add, in the form Shopify’s cart expects. The twin sees the page the shopper is on and what’s in their cart, and follow-up questions continue the conversation. A second action, “apply offer”, puts a current discount code on the cart, so the deal the twin mentions is the deal the shopper gets. It only accepts codes the merchant shares. The merchant sees these conversations marked “AI agent”.

Feature 9 of Shiptober. We set out to ship a new feature for the sqrly Shopify app every day until October 20, and we’re running ahead of that. Want it in your store? Email us about becoming a pilot shop.

Every store running sqrly now has an “Ask our expert” page on its own domain, in its own design. Ask a question, and the page comes back with the digital twin’s complete answer, the products it recommends and a standard add-to-cart button for each. Follow-up questions continue the same conversation. There’s nothing for the merchant to set up, and nothing in the store’s theme changes.

We built it for the AI agents that shoppers send to shop for them. A page with a question box and a full answer is what these agents handle best: nothing to wait out, nothing that appears later, and it works even without JavaScript. Agents that only read a page’s text now see the twin’s whole answer, and an agent can go straight to a question by its address. Every page of the store points agents to it. People get a full-page version of the twin too, which is especially nice on a phone. In the merchant’s Conversations, these show up marked “Expert page”.

Feature 8 of Shiptober. We set out to ship a new feature for the sqrly Shopify app every day until October 20, and we’re running ahead of that. Want it in your store? Email us about becoming a pilot shop.

More and more shoppers send an AI agent to shop for them. These agents open a store’s website, read the page and click their way through it, much like a person would. On a store running sqrly, a visiting agent can do something it can’t do anywhere else on the page: consult the store’s digital twin, which knows every product, the live stock, the current offers and everything the merchant has taught it. So we put that to the test. We sent an AI agent to our demo store with a task: find skis for an intermediate skier under $600 and put them in the cart, working only with the digital twin. It got there, but it stumbled along the way.

Now the digital twin introduces itself to visiting agents: the store’s product expert, with live stock and current offers, able to add to the cart. Every product it recommends is clearly labeled, so an agent adds exactly the one it means. While the twin is still answering, it says so, and an agent never acts on half a reply. All of its recommendations are in view at once. And the twin now sees which size is already in the cart, so it can tell the shopper, or their agent, whether to keep it or swap. Shoppers who use screen readers get all of this too.

This is the first step toward what we’re most excited about this month: a shopper’s agent and the store’s digital twin working out the right purchase together. More on that in the days ahead.

Feature 7 of Shiptober. We set out to ship a new feature for the sqrly Shopify app every day until October 20, and we’re running ahead of that. Want it in your store? Email us about becoming a pilot shop.

Until now, a merchant who wanted the digital twin to mention a deal had to set it up twice: once in Shopify, once in sqrly. With more than one or two offers, that gets old fast. Now there’s nothing to copy. One click connects the store’s Shopify discounts, and every active discount open to all customers shows up in sqrly on its own: codes, automatic discounts, buy-one-get-one deals and free shipping.

The twin uses Shopify’s own wording, so what it tells shoppers matches checkout exactly. It mentions a deal when it fits what the shopper is looking at, says when no code is needed, and knows when an offer ends. When the cart qualifies, the checkout link applies the code for the shopper. New, changed or deleted discounts reach the twin within minutes. Discounts meant for specific customers stay out of it, and the merchant can keep any code private, like a staff or partner code, with one click.

Feature 6 of Shiptober. We set out to ship a new feature for the sqrly Shopify app every day until October 20, and we’re running ahead of that. Want it in your store? Email us about becoming a pilot shop.

A digital twin learns from many places: the store’s policies and pages, the merchant’s answers, fixed transcripts and findings from the web. Over time the same fact ends up in more than one entry. The return policy says 30 days, and so does an answer the merchant wrote three weeks ago. Now the twin notices. It checks its knowledge every now and then, finds entries that say the same thing, and proposes one clear answer that keeps every fact once.

The merchant decides: merge (after editing the merged answer, if they like) or keep the entries separate. Nothing changes until they do. The merged answer counts as the merchant’s own word, keeps the products it’s linked to, and shows up on the knowledge map. Entries that disagree are never merged. A feature later in Shiptober takes care of those.

Feature 5 of Shiptober. We set out to ship a new feature for the sqrly Shopify app every day until October 20, and we’re running ahead of that. Want it in your store? Email us about becoming a pilot shop.

The sqrly home screen now opens with the last seven days of the digital twin: how many conversations it had (and how that compares with the week before), how many questions shoppers asked, and how many products went into carts and on to checkout.

Below that, three short lists: what shoppers asked most, which products the twin recommended most, and which ones shoppers actually added to their carts. The gap between the last two is often the most interesting part of the week. It closes with what the twin learned and how many questions are waiting for the merchant’s answer. No reports to set up and no export: it’s there every time the app opens.

Feature 4 of Shiptober. We set out to ship a new feature for the sqrly Shopify app every day until October 20, and we’re running ahead of that. Want it in your store? Email us about becoming a pilot shop.

Reading what shoppers asked is the fastest way to see what the digital twin gets right and where it’s missing something only the merchant knows. Now every reply in Conversations has an Improve this answer button. It opens the twin’s own words, ready to edit: add the tip, fix the detail, change the pick. Save it, and the twin uses the merchant’s version for every similar question from then on.

The fix lands in the shop’s knowledge as the merchant’s answer, which beats anything the twin found elsewhere. It’s filed by topic, linked to the products it’s about, and shows up on the knowledge map. If the same question was waiting among the open questions, it’s marked answered too. Improved replies are marked in the transcript and can be edited again anytime.

Feature 3 of Shiptober. We set out to ship a new feature for the sqrly Shopify app every day until October 20, and we’re running ahead of that. Want it in your store? Email us about becoming a pilot shop.

Shopify now publishes a basic llms.txt file (a short guide for AI systems) on every store automatically. Our scan was counting that file as something the store had done itself, which made Shopify stores look a few points readier than they are. As of today it doesn’t: Pulse recognizes Shopify’s default file, tells you it’s there, and gives it no extra credit. A file you write yourself still counts, and Pulse now shows the right way to install one on Shopify (a theme template) without losing the agent links Shopify’s version carries.

What you’ll see: if your store runs on Shopify, your score may drop by about 4–5 points at your next full scan. Nothing on your store changed, and your dashboard says so right next to the number. Our published benchmarks for ski shops and running stores keep their original scores and note the change.

Also new: the merchant-programs panel reflects how AI shopping channels work today. Shopify shares eligible catalogs with ChatGPT, Microsoft Copilot and Meta by default, and Google began switching on checkout in AI Mode and Gemini in September, so Pulse points you to your store’s Agentic settings to see which are on. And because most ChatGPT shopping picks now come from merchant product feeds, the shopping panel explains how to get your catalog into one.

A shopper who asks in German now gets a German answer. One who asks in Japanese gets Japanese. The digital twin replies in whatever language the shopper writes in, using everything the merchant taught it once, in their own language: shipping rules, the return policy, field notes. There’s nothing to set up and nothing to translate.

The facts stay exact. Product names, sizes, colors, prices and discount codes appear just as they are in Shopify, so the product cards and the checkout always match what the twin said. On stores that sell in several languages, the twin’s buttons and labels switch with the storefront, in 12 languages so far. And when a shopper asks something the twin can’t answer, the question still reaches the merchant in the merchant’s own language.

Feature 2 of Shiptober. We set out to ship a new feature for the sqrly Shopify app every day until October 20, and we’re running ahead of that. Want it in your store? Email us about becoming a pilot shop.

Teaching a guide from a blank page is slow. Reacting to a draft is fast. So right after a store installs sqrly, it now searches the public web for what the store’s guide doesn’t know yet: what customers and reviewers say about the shop and its products, the shop’s story and its team’s expertise, press coverage, and facts published elsewhere. A few minutes later the merchant gets a short list of findings to review: accept, edit or dismiss.

Every finding links to where it was found, and none of it reaches shoppers until the merchant accepts it. Opinions stay attributed (“skiers on a forum say…”), the merchant’s own word always wins, and the search never looks into anyone’s private life. In our first test on a real lure brand it came back with 11 sourced findings in about two minutes, including two articles that disagreed on the year the company was founded. That’s why the merchant has the last word.

Accepted findings join the twin’s knowledge graph like everything else it knows: filed by topic, linked to the products they’re about, and visible on the knowledge map. Pilot shops get this from day one; email us if your store should be one of them.

When a shopper asks something the guide can’t answer from the catalog or from what the shop has taught it (“Do you ship to Canada?”, “Can I return a lure I’ve fished once?”), it now says so honestly instead of guessing. The question lands on the merchant’s Knowledge page, phrased for the owner, with the shopper’s own words and how often it was asked. Answer it once and every shopper after that gets the answer. Asked on a product page, the answer is linked to that product.

A new shop doesn’t start empty either. One click imports the store’s policies and published pages (returns, shipping, FAQ, size guides) and turns them into short notes the guide can use. Shopify asks the merchant to allow “Online Store pages” and “Legal policies”, nothing about customers or orders.

Everything the twin knows forms a knowledge graph: topics, answers, imported pages and the catalog by product type, linked to each other. Merchants can explore it as a map in 2D or 3D, with open questions glowing orange wherever the twin is still thin. Pilot shops get all of this; email us if your store should be one of them.

Our second product is a Shopify app: a digital twin of your store that knows every product, color and stock level and answers shoppers in your store’s chat the way your best floor staff would. It recommends with product cards — the right color’s photo, the exact price, Add to cart — says plainly when something is the wrong choice, and never recommends what’s sold out. It has run end to end on our test store since October 2; the MVP post has the full story, including a working preview inside AI assistants like Claude.

As of today, a Shopify store can install it. Setup asks only to read your products and stock — nothing about your customers or orders — then syncs your catalog by itself, and a short checklist walks you through turning on the chat, teaching it your shipping rules and policies, and adding a deal. Answers start appearing in about two seconds. You see every conversation, and which ones led to an add to cart or a started checkout, without granting any extra access.

We’re looking for a handful of pilot shops: Shopify stores where good advice decides the sale — fit, compatibility, the right gear for the job. Each pilot gets its own private install and works directly with the founder. If that’s you, email florian@sqrly.ai with your store’s URL and what you sell.

We spent a day auditing our own reports the way we audit AI assistants: six real stores, every claim checked against the live site, everything that didn’t hold, fixed. The biggest change is a new rule at the heart of the scan: couldn’t fetch and doesn’t exist are different facts. If a store’s bot protection blocks our crawl, the report now says exactly that — with the evidence — and refuses to produce a score, instead of failing a store for pages it never saw. Anything unverified renders as “couldn’t check this scan,” never as a failure.

The scan also reads more of what stores actually publish. Product schema in all three formats — JSON-LD, microdata, RDFa — so an older storefront’s markup counts the way Google and Bing already count it. Sitemaps declared only in robots.txt. Small sitemaps crawled in full, so no section goes unread. Rental and service price lists. Policy and story pages linked from your footer. Chat widgets loaded through tag managers, plus a dozen more vendors recognized by name. And robots.txt verdicts got precise: exactly which AI crawlers are blocked, which still get in — and when Cloudflare’s managed ruleset wrote the file, the dashboard toggle that changes it.

Fact-checks got fairer too. An assistant is never flagged for a fact that appears anywhere on your own site — and when your pages disagree with each other (homepage says founded 1979, your story page says 1973), the report flags that: one fix on your site aligns every assistant at once. Our crawler now also introduces itself properly at sqrly.ai/bot — what it reads, how gently, and how to turn it away.

Reports now read how each assistant frames your store, not just whether it appears: the slot it puts you in (“the local ski specialist,” “a budget pick”), the tone, and the exact adjectives it used — quoted verbatim from its answer, never paraphrased. When one assistant calls a store “trusted” while another says “hard to verify,” that gap is the work list.

Three more things every scan now shows. Whether you’re actually set up for the merchant programs that feed AI shopping surfaces — ChatGPT’s, Perplexity’s free program, Google’s new conversational attributes — including the checks only you can see behind your admin login, with exactly where to look. How far an AI agent gets when it tries to buy from you, stage by stage, from finding your store’s agent door to reaching checkout — composed from real probes; we never place an order. And the sharpest finding of all: the questions where an assistant read your own pages and still didn’t name you — proof that being findable isn’t the same as being choosable.

For subscribers, every lost shopper question grew a button: Pulse drafts the page that answers it, grounded only in your store’s real data, with placeholders where only you know the fact. You review, edit, publish — nothing goes live without you.

And because you should be able to check us: every verdict now carries the exact question we asked, word for word, with a copy button — try it yourself in a temporary chat. Expect overlap, not identical lists; answers shuffle between sessions, which is exactly why a weekly watch beats a one-time look.

The dashboard said everything at the same volume — and for a store that isn’t winning yet, that meant a wall of red. Today’s redesign starts with what’s already working for you (real strengths, each opening its evidence), puts your next moves in the first screen, and folds three overlapping evidence cards into one store-readiness count. The full breakdown — every check and finding — is exactly where it was, one click behind the score instead of splayed across your first impression. About a quarter less text before you scroll, nothing removed.

When a shopper asks ChatGPT to buy from a store, the app often answers with a shopping carousel — product cards with prices and a buy button, built from merchant feeds. That carousel is a different pipeline from the web-search answers monitoring tools check, and the two can disagree: a brand can win the text answer and still lose every card. Reports now observe it directly — a real, logged-out ChatGPT session asks to buy from your store by name, and the report shows every card that came back: your store, or the exact reseller who captures the sale instead, at their price.

Until today the report assessed this risk from your feed signals and said so plainly — we don’t pretend to see a surface we can’t. Now we can see it, so we show it: one observed snapshot per scan (the app rebuilds the carousel every session), with what we saw and when.

Every Pulse dashboard can now draft the work it recommends. The llms.txt move writes the actual file from your store’s own scan data. Every fact-check flag (“says open 24/7 — your site says 9–6”) grows a correction kit: what’s wrong, where the assistant likely read it, paste-ready corrected copy, and how to get the page recrawled. And the sites that shape your category’s answers each get a drafted outreach pitch, grounded only in what your store verifiably offers.

Everything is a draft you review — nothing is ever published to your store without you. Weekly emails now lead with a single next move instead of a change log, and you can point your dashboard’s emails at any address, straight from the footer.

sqrly Pulse has a proper home at sqrly.ai/pulse, and you can now subscribe there directly — no free report required first. We check that your catalog is readable before any charge, and a store that already has a subscription can’t be charged twice.

The free report also got its honest name: it’s a free Pulse snapshot — the same product, as a one-time look. And every report and dashboard grew a feedback button that lands straight in the founder’s inbox.

Pulse dashboards now show which of your pages the AI assistants read when they answer shoppers — question by question, assistant by assistant. If Claude cites your boot-fitting page when someone asks who fits boots best near Tahoe, you’ll see exactly that. And if none of your pages show up in their reading, that’s worth knowing too — it means their answers about you come from other people’s sites.

Every report and weekly pulse now starts with a reachability check: TLS, bot protection, and robots rules, tested the way an AI crawler actually experiences them. A store that blocks the crawlers is invisible to the assistants no matter how good its pages are — so this comes first, and proven access loss triggers an alert email. Nothing speculative does.

We also taught our own scanner to fully honor robots.txt — if a store asks not to be crawled, we don’t, and the report says so plainly. We should live by the rules we check.

The watching service. Subscribe once at $9.99/month and your store gets a full scan immediately, a light visibility pulse every week, and a full rescan every month — with a live dashboard, competitor tracking, and an email only when something actually moves. No account and no password: the link is the login.

It exists because a report is a snapshot, and AI answers don’t hold still — assistants change their minds, rivals get featured, product data goes stale. The launch post has the full story.

Reports now ask Perplexity alongside Claude, ChatGPT, and Gemini. It’s the assistant shoppers lean on hardest for “which one should I buy” research, which makes it the most commercially important answer sheet we check.

The detailed report now asks the assistants ~30 real shopper questions about your category — never naming your store — and counts who gets recommended: you, and who instead. It maps which sites shaped those answers into a ranked get-featured list, and tests whether your product photos reach AI answers at all.

Anyone can run their own AI-readiness report at assessment.sqrly.ai/check — enter your store URL, confirm your email, get a private link. No account, no sales call, and the free DIY fixes are included, not held hostage.

Being visible isn’t enough if the assistants get you wrong. Reports now verify what each assistant says about your store against your own site — wrong hours, wrong policies, invented services get flagged with the exact quote. Plus ten buyer questions (“what does shipping cost?”) tested against your pages, and a one-paragraph verdict up top instead of a wall of scores.

The report began asking Claude, ChatGPT, and Gemini about stores live, with web search enabled — the way a real shopper’s assistant works, not from stale training memory. That week also added a live ChatGPT-shopping probe (are your products in its shopping results right now?) and turned every report into a private, shareable link.

We started building: a readiness check that reads an online store the way AI systems do — content, structured data, catalog feed — and says plainly what’s missing and how to fix it, including the fixes that cost nothing. Private and invite-only at first, while we tested it against real stores.