System ready // v1.0 // point me at a store
Point me at your store.
I'll show you what AI agents can't see.
Not a checklist. We fetch your pages the way GPTBot and PerplexityBot do, read what comes back, and tell you which of your products an agent can find but never buy. You get the score and the code to fix it, free.
01 / Method
What we check, and what each part is worth
Every item is deterministic and checkable from outside your store in one pass. The weights reflect how much each one changes whether an agent can complete a purchase.
22 points
Crawler access
Whether GPTBot, OAI-SearchBot, ChatGPT-User, PerplexityBot, ClaudeBot, Google-Extended and Bingbot can reach your product pages. SALT.agency found 15% of the retailers they tested block AI crawlers outright.
28 points
The three agentic-critical attributes
priceValidUntil, shippingDetails.deliveryTime and hasMerchantReturnPolicy.merchantReturnDays. Is it still this price, when does it arrive, can it go back. 70% of the product pages SALT tested had none of the three.
13 points
Product schema
Valid Product JSON-LD with a name, an image, offers.price and offers.priceCurrency, in the response your server sends.
9 points
Server-side rendering
Whether your product content is in the raw HTML. Agents fetch and read. They don't run your JavaScript.
8 points
Stable identifiers
GTIN, MPN and SKU. Without one, an agent can show your product but can't match it to the same product anywhere else. 65% of pages SALT tested carried no GTIN.
6 points
Category pages
ItemList or CollectionPage schema, product links as real anchors, and the grid present before JavaScript runs. An agent reads your collections to work out what you sell. A grid drawn client-side is a dead end.
4 points
Returns and shipping policy
Whether your policy page is reachable, readable without JavaScript, and states a window as a number. An agent will not recommend a purchase it can't confirm is returnable.
4 points
Protocol support
A manifest at /.well-known/ucp. Almost nobody has one yet, which is exactly why it's the cleanest signal when you do.
4 points
Transactability hygiene
Returns, shipping and terms linked where an agent looks for them. Guest checkout matters too, but we can't check that from outside without placing an order, so we don't pretend to.
2 points
Homepage identity
Organization and WebSite schema, and navigation an agent can follow as ordinary links. Small, because almost nothing about a homepage decides whether a product can be bought.
Zero points, on purpose
llms.txt
We check it and give it no weight. Ahrefs found 97% of llms.txt files got zero AI traffic across 137,000 sites. Google compared it to the keywords meta tag. Being the vendor who says so seemed more useful than another point on a chart.
02 / Why it matters
The gap between ranking and being buyable
Traditional SEO gets you described. An agent needs more than that. Before it recommends a purchase it wants to know the price is still good, roughly when the thing turns up, and what happens if it's wrong.
Most stores answer none of those in machine-readable form. SALT.agency tested 141 product pages across 29 retailers in July 2026 and found 70% carried none of the three attributes. Adobe, looking at over a trillion visits to US retail sites, scored product pages at 66% machine-readability, the worst of any page type on a retail site.
That is the whole opportunity. The work is cheap, it's the same structured-data work that helps Google, and almost nobody has done it.
We should be straight about the size of it too. AI referrals are still under 1% of ecommerce traffic, and ChatGPT alone is under 0.2%. That's peer-reviewed work by Kaiser and Schulze in Marketing Science, across 973 sites and $20 billion of revenue. Anyone telling you this is already most of your traffic is selling something.
The direction is what makes it worth an afternoon. Shopify told its Q1 2026 earnings call that AI-referred orders were growing around 13× year on year, the fastest-growing inbound channel it has tracked, and that AI-sourced orders bring new buyers at about twice the rate of traditional organic search.
Every number on this page is sourced. If you want the working, it's in the report the scanner produces.
03 / FAQ
Questions we get asked
04 / Related
Where to go next
The self-assessment
Ten questions about things a crawler can't see from outside, like your feed health and your attribute quality. Nothing you enter is sent anywhere.
ROI calculator
Put your own traffic and order numbers in and see what search, service and merchandising work is worth before you commit to any of it.
Discover & Convert
How Sellarix keeps structured data, feeds and agent access current across the whole catalogue rather than one product template at a time.
Fixed the free ones? Let's talk about the rest
We do the catalogue-wide version of this: schema, feeds and agent access kept current as your products change. Join the waitlist, or book a call about your scan.