The CortexGuard changelog

What it's about

There's a new card on the Dashboard called "Start here — your biggest catalog gaps". It counts seven things across your active products — products with no description, products described in under 30 words, products sharing a description with another product, products with no photo, products with images missing alt text, products with no meta description, and products with no tags — and ranks them largest first, the biggest three with a bar showing how much of your catalog each one touches. The counts are exact, not sampled, and they move as your catalog syncs and as you approve suggestions. Every row with a number has an "Enrich these" button. It opens the usual enrich step already narrowed to just those products, tells you exactly how many credits the run will use and how many you have, and if you're short, offers a top-up right there — from $10, prefilled to the amount you're missing — with a choice between topping up and enriching all of them, or enriching as many as your balance covers now. After Shopify's approval page you land back on the Dashboard with the same step open on the same gap. If some of those products already have a suggestion waiting for you, the card says so and points you to review them rather than sending them again. New stores now also get one email after their first scan — "We had a look at your catalog — here's what we found" — listing the same gaps with the same counts and one button back into CortexGuard. It only goes out if there's something to say; a clean catalog doesn't get one.

Why we made this

Most catalogs have the same handful of problems — thin or copied descriptions, missing meta descriptions, photos with no alt text — and they're exactly the things an AI shopping assistant trips over when it tries to understand a product. Until now you could see your overall enrichment progress but not which problem was biggest, or start a run aimed at just that problem. And when a run needed more credits than you had, the only way forward was a detour through Billing. Now the biggest gap is the first thing you see, and closing it is a decision you can make on the spot.

What this means for you

• You can see at a glance which gap is biggest and how much of your catalog it touches, in exact numbers — no guessing which problem to fix first.
• A run aimed at one gap spends credits only on the products that actually have that problem, and never re-sends a product that already has a suggestion waiting.
• Meta descriptions, alt text and real product descriptions are what AI assistants read to decide whether your product is the right answer — closing these gaps is the most direct way to make a store legible to them.
• If you're short on credits, you can top up from $10 without leaving the enrich step, and you come back to exactly where you were.

Good to know

If your store was connected before this went live, you may not see the card yet. The counts are built during a full catalog sync, and syncing from the Dashboard only fetches what's changed since last time — so for an existing store the card fills in after the daily automatic sync has run (the one you switch on under Settings → Automation, "Automatically enrich new and updated items daily"), and new installs get it from their very first sync. The same rule decides whether the catalog email goes out. We'd rather tell you that than show you a number that only covers part of your catalog.

The counts cover active, published products only — the same rule scans use (see the entry from 2026-09-02).

Alt text is checked on a product's first ten images; an eleventh uncaptioned photo isn't flagged.

A gap fix enriches products only; collections aren't part of it. The $10 minimum is for the top-up inside the enrich step — the Billing page's own options are unchanged.

What it's about

CortexGuard's enrichment is really several jobs in a row: look at the product's photos and work out what's actually in them, write the copy, and a handful of internal checks in between — is this image good enough to describe from, does this draft need another pass. Until now all of them ran on the same model. Now the two whose output you see — reading the photos and writing the copy — run on a newer, stronger model, and the internal checks, which nobody outside CortexGuard ever reads, run on a lighter, faster one. Picking the product out of a lifestyle shot is the hardest thing enrichment does, and it's where the stronger model makes the most difference. The analysis behind your Brand Profile, and the feedback it gives on each field, moved to the same stronger model.

Why we made this

Two reasons. The model we'd been using for everything is being retired by its provider in October, so we had to move anyway. And putting one bigger model on every job, including the internal decisions nobody sees, would have been wasteful — splitting the jobs lets us put the strong model where it shows: in what a customer, or an AI shopping assistant, actually reads.

What this means for you

• Copy and photo understanding are better where it counts — the parts of an enrichment a customer or an AI assistant actually reads.
• For AI shopping assistants, a description that correctly names what's in the photo is the difference between being the answer and being skipped — this is the model doing that job.
• No change to pricing: a product is still one credit, a collection still two.
• Nothing already written is rewritten. Only enrichments from now on use the new model.

Good to know

This isn't a re-run. Unlike the photo-aware change from 2026-08-31, nothing already enriched becomes eligible again and no credits are used on your behalf — the new model is simply used from the next enrichment onward, whether you start it or the daily automation does.

What it's about

When CortexGuard scans a store for AI visibility, it now looks only at active, published products — the ones an AI shopping assistant could actually land on. Draft and archived products, which have no live page to check, are no longer counted or checked during a scan, and the progress shown while a scan runs now reflects that same, smaller, real number.

Why we made this

A store with a large share of draft or archived products was having its scan try to check pages that could never succeed, since those products have no live storefront page at all. For a catalog with tens of thousands of products, that could turn a scan that should take minutes into one that took days for no benefit.

What this means for you

• Scans finish faster, especially for large catalogs with a lot of unpublished products.
• The AI-visibility number a scan reports is now based only on products an AI assistant could actually reach, which is a more honest measure of what's really visible.

Good to know

Very large catalogs can still take a while to scan in full — a limit on how much a single scan checks at once is something we're still working on.

What it's about

Pricing is now exactly what you see when you start a run: one credit for a product — its written copy and its alt text together, no separate charge for either — and two credits for a collection. The older "image enrichment only" run type, with its own per-image charge and its own cap, is gone, folded into the pricing above. If you export your enrichment history, you can now choose to export everything, just products, or just collections, instead of one combined file. And when you try CortexGuard for the first time in the guided Test Ground, what you see there is now a real enrichment, not a simplified preview of one.

Why we made this

Running two different pricing models side by side — one for regular enrichment, a separate one for images — was confusing and didn't match what most people expected to pay. This makes the price you see the price you get, for every kind of run.

What this means for you

• Simpler, more predictable pricing: one rate for a product, one for a collection, nothing extra for alt text.
• What you try during onboarding now behaves exactly like a real enrichment, so there's no surprise once you start using CortexGuard for real.
• No AI-assistant angle — this changes pricing and export options, not what CortexGuard writes or how a store is found by AI.

Good to know

If you were using the separate alt-text-only run, or the dedicated Images screen for a collection, both are retired — alt text now comes with every full enrichment instead (see the entry above from 2026-08-31). An old bookmarked link to that Images screen will now take you to the regular Products screen instead.

What it's about

The biggest change: when CortexGuard writes new copy for a product, it now looks at the product's own photos first and judges whether they're trustworthy enough to describe from, rather than working from text alone. Alt text for images now comes along automatically with every full enrichment, instead of being a separate pass you had to run on its own — the dedicated alt-text screen is retired because alt text isn't separate anymore. While a run is going, the Review Inbox now shows a live count of how many items have been processed so far, and the dashboard has a progress card that keeps showing that same live progress even if you refresh the page. Two smaller things landed the same day: the dashboard's enrichment-percentage numbers can no longer show something impossible like 125% — they cap sensibly at 100% now — and the dashboard itself loads faster because it reuses a result it just worked out for a short stretch, instead of recalculating it on every single visit.

Why we made this

Text alone can miss what a photo already shows — or contradict it. Checking the photo first means the copy CortexGuard writes is judged against what a customer will actually see, not just against your existing product text. And not being able to tell whether a run was still going, or how far along it was, was a real gap for anyone enriching a large catalog and wondering if it had stalled.

What this means for you

• Copy written by CortexGuard is now checked against your product's real photos, not just its existing text — for the AI shopping assistants this product exists to help with, that means descriptions are less likely to contradict what's actually in the picture.
• Alt text is no longer something you have to run separately — every full enrichment includes it, so images stay legible to AI assistants and screen readers without an extra step.
• You can now watch a run in progress instead of wondering whether it's still working, both from the Review Inbox and the dashboard.
• Two smaller fixes on the same day: the dashboard's percentage numbers can no longer read above 100%, and the dashboard itself loads noticeably faster.

Good to know

Because this change judges enrichment partly by your product photos, it treats a photo as new information the same way new text would be. The first time this ran, on 2026-08-31, every product that had a photo became newly eligible for a fresh enrichment pass and was re-enriched once, automatically, at the normal rate of one credit per product. If you noticed enrichment activity or credit use that day that you didn't start yourself, this is why. It was a one-time pass tied to this change going live, not something that recurs for the same product.

The dashboard's faster loading means its progress number can briefly hold steady for up to a short stretch before updating, rather than climbing in a perfectly smooth line — the run itself isn't paused, only how often the number refreshes.

What it's about

In the Review Inbox, when you're approving or rejecting AI suggestions for a collection's products one at a time, reaching the last couple of items used to leave the counter and navigation showing a position that didn't make sense. That's fixed — the counter now stays accurate all the way through, and if you clear every item in the list, the screen closes on its own instead of staying open with nothing left to show.

Why we made this

A review screen that shows an impossible count breaks your trust that anything you're seeing is accurate, right at the moment you're deciding whether to publish AI-written content.

What this means for you

• The review screen now behaves correctly all the way to the end of a list, with no broken counters or stuck screens.
• No AI-assistant angle — this is a fix to the review screen itself, not to what's reviewed or how a store is found by AI.

What it's about

Enrichment — whether you start it manually, run it on a chosen subset, let it run automatically on a schedule, or preview it during onboarding — now skips any product marked as a draft or archived in your store. Only products that are actually live, or that don't report a status at all, are eligible.

Why we made this

A draft or archived product isn't visible to customers yet, so writing new copy for it and spending a credit on it wasn't buying you anything. This closes that gap everywhere enrichment can run, not just in one place.

What this means for you

• You won't spend credits enriching products that aren't published, across every way enrichment can run — manual, scheduled, or scoped.
• No AI-assistant angle here — this only changes which products are eligible for enrichment, not what CortexGuard writes or how a store is found by AI shopping assistants.

Good to know

This doesn't refund credits already spent on a draft or archived product before this change — it only stops it from happening going forward.

What it's about

When CortexGuard enriches a product that has no photo at all, it now generates one in the same pass — no extra step, and no extra cost beyond the usual one credit for that product. It lands in your review inbox just like a rewritten description does, so you approve, edit, or reject it before it's used anywhere. A product that already has a photo is never touched by this. You can also tell CortexGuard, when starting a bulk enrichment, to only look at products that have no photo at all. Separately, the dashboard tile that reports how many AI shopping assistants can see your store was showing a reassuring green result even for stores CortexGuard hadn't scanned yet, or where the only thing it could actually check was a password page in front of the whole store. It no longer does that.

Why we made this

Some catalogs have products with genuinely nothing to show — no photo of any kind. A generated image gives customers, and the AI writing about the product, something to work from instead of a blank space. And a dashboard number is only useful if you can trust it — a green result for a store that was never actually checked was worse than showing no result at all.

What this means for you

• A product with no photo at all can now get one automatically, reviewed by you before it's used — you're never surprised by an image you didn't approve.
• This doesn't touch products that already have a photo, and it never replaces a real product photo with anything.
• The AI-visibility dashboard now only reports what it actually measured, so a green result means something was genuinely checked, not assumed.

Good to know

A generated photo is an AI's interpretation of your product based on its name and details — it has never seen the real item, so treat it as a placeholder illustration, not a substitute for real product photography. If you have auto-approval turned on, a generated photo can publish to your store without a manual review, the same as any other auto-approved change — worth knowing if you'd rather review images yourself.

This is on by default for new enrichment runs; you can turn it off under your enrichment settings if you'd rather leave photo-less products alone.