AI Readiness
In plain English
More and more people shop by asking an AI assistant — "find me a linen throw under £80" — instead of typing into Google.
When that happens, the AI doesn't look at your website the way a person does. It reads a list of facts about each product: name, description, price, photo, brand, category, product code. If any of those facts are missing, the AI can't recommend your product. Not "ranks it lower" — it can't include it at all.
Here's the catch: your website can look perfect and those facts can still be empty. The website shows the price from one place; the AI reads it from another. If that second place is blank, you're invisible.
This screen checks every product and gives it a score out of 100. It also flags something more important: whether an AI could actually sell the product. For that it needs three things — a price, a photo, and stock available. Miss any one and the answer is no.
How to use it
- Click "Daily Ops" in the left-hand menu, then the "AI Readiness" tab.
- Look at the percentage at the top. That's how much of your catalogue an AI could actually sell. Aim for 90% or more.
- Filter the list to "Not ready" and sort by score, lowest first.
- Click a product. You'll see a tick-list of what's missing, with the most important thing at the top.
- Fix the missing fields — usually a longer description, a price, or a photo.
- Fix "can't be sold" products first, even if their score looks higher than others. Those are the ones losing you sales right now.
- Check back after your next product sync. Scores update automatically.
⏱ ~15 min · 💳 Starter+ · 🎯 Be sellable on the fastest-growing buying surface
Why this matters for your business
A growing share of purchase journeys now starts in a conversation rather than a search box. Someone asks ChatGPT for a linen throw under £80, asks Perplexity which running shoe suits flat feet, asks Google's AI to compare three espresso machines. The agent answers from structured product data — title, description, price, image, brand, category, identifier — pulled from feeds and structured markup.
Here's the part that catches merchants out: your storefront can look immaculate and your products can still be invisible. A theme renders a price from a variant object; a feed reads a price field. If that field is empty, your beautifully-designed page is a product the agent cannot quote, cannot compare, and cannot recommend. It isn't ranked low. It isn't there.
This tab grades your catalog against what agents need, product by product. Not as a vague health score — as a weighted checklist where each missing field has a known point cost and a known consequence. And it separates two very different failures: poorly represented (an agent might mention you but can't describe you well) and not transactable (an agent physically cannot sell this), because those need different urgency.
What this typically unlocks
| Outcome | Typical result |
|---|---|
| Catalog visible to AI shopping agents | From ~50–65% to 90%+ in one clean-up pass |
| Products an agent can actually sell | Measured explicitly, not assumed |
| Time to find the worst products | Seconds — sort by score ascending |
| Fix prioritization | Automatic — gaps listed in weight order |
| Feed rejection rates downstream | Sharply lower — same fields Google and Meta require |
| Re-scoring after a fix | Next catalog sync, no manual re-run |
What you actually get
Nine weighted checks, totalling 100 points.
| # | Check | Weight | Passes when | Importance |
|---|---|---|---|---|
| 1 | Rich description | 20 | At least 120 characters of real text (HTML stripped) | High |
| 2 | Price set | 18 | A parseable price greater than zero | High |
| 3 | Product image | 14 | An image URL is present | High |
| 4 | Brand / vendor | 10 | Vendor field populated | Medium |
| 5 | Product type | 10 | Product type populated | Medium |
| 6 | Descriptive title | 8 | Title at least 10 characters | Medium |
| 7 | SKU / identifier | 8 | SKU populated | Medium |
| 8 | Meta description | 7 | Meta description populated | Low |
| 9 | Attributes | 5 | Material or weight populated | Low |
Bands
| Band | Score | What it means |
|---|---|---|
| Ready | 80–100 | Agents can find, represent, and compare this product properly |
| Partial | 50–79 | Surfaced, but described thinly — loses comparisons on missing detail |
| Not ready | 0–49 | Effectively invisible or unusable |
Transactable — the check that matters most
Score and transactability are different questions, and conflating them is the most common mistake merchants make here.
transactable = has price AND has image AND not out of stock
That's 32 of the 100 points, plus live availability. A product can score 68 — a respectable Partial — and still be not transactable if it's missing an image. An agent might mention it in passing but cannot put it in a basket.
Conversely a product can score 79 (just short of Ready) and be fully transactable — sellable today, just under-described in comparisons.
Stock is treated as three states, not two:
| Stock state | Effect |
|---|---|
| In stock | Transactable if price and image are present |
| Out of stock | Not transactable — flagged as a stock issue, not a data issue |
| Unknown | Never penalised — you're not marked down for missing signal |
That last row matters: an unknown-availability product isn't assumed broken. Guessing against the merchant would inflate the problem count and destroy trust in the percentage.
The shop-level rollup
| Metric | What it tells you |
|---|---|
| % transactable | Share of catalog an agent could actually sell — the number on the Executive Pulse |
| Average score | Overall representation quality |
| Ready / Partial / Not ready counts | Where the catalog sits by band |
| Total scored | Catalog size in scope |
How it works (without the technical bits)
How findings are raised
Any product that is not both Ready and transactable raises one finding — one stable finding per product, refreshed on each re-score rather than duplicated.
Severity escalates by consequence, not by score alone:
| Situation | Severity | Reasoning |
|---|---|---|
| Not transactable | Critical | An agent cannot sell this at all |
| Not ready (< 50) but transactable | Warning | Sellable, but poorly represented |
| Partial (50–79) and transactable | Info | Working, with room to improve |
Each finding names the top three missing fields by weight, so the body of a row reads like an instruction:
Make "Linen throw — sand" ready for AI shopping agents
AI agents can't sell this yet — Readiness 44/100.
Add: Rich description, Price set, Product image.
When the product is out of stock rather than incomplete, the wording changes to say so — because "restock this" and "fill in this field" are different jobs for different people.
Findings carry a confidence of 0.7. That's deliberately not 1.0: the score grades catalog completeness, which is an honest proxy for agent-readiness rather than a live measurement of what a specific agent did. Claiming certainty would over-rank these against measured findings like leaks.
Auto-resolution
Readiness findings auto-resolve after 7 days. The window is short because the catalog is re-scored on every sync — a finding older than a week is describing a product state that has almost certainly changed. Fix the fields and the finding closes on the next scoring pass without you touching it.
Real merchant scenarios
Scenario A — The 47% that looked fine
Setup. Home-goods merchant, 2,800 SKUs, a well-designed storefront and no obvious problems.
First scan:
Agent-ready products 47% (avg 61/100 readiness)
Ready 680
Partial 1,290
Not ready 830
Cause. Descriptions had been written as short marketing lines — "Soft. Simple. Yours." — averaging 34 characters. Beautiful on the page, and a fail on the 120-character check worth 20 points. Meta descriptions were absent across the board (another 7). Together that capped most of the catalog at 73 — Partial, never Ready.
Fix. Bulk-generated longer descriptions and meta descriptions through Content Studio, keeping the short line as an opening sentence.
| Before | After | |
|---|---|---|
| Agent-ready | 47% | 91% |
| Average score | 61 | 88 |
| Not ready | 830 | 74 |
The insight. Nothing was broken. The catalog was written for humans reading a page, not for machines reading a field. Both audiences now get what they need.
Scenario B — Transactable vs. score
Setup. Furniture merchant confused that products scoring 68 were flagged critical while products scoring 54 were only warnings.
The explanation:
| Product | Score | Price | Image | Transactable | Severity |
|---|---|---|---|---|---|
| Oak side table | 68 | ✓ | ✗ | No | Critical |
| Walnut shelf | 54 | ✓ | ✓ | Yes | Warning |
The side table had a rich description, brand, type, SKU, and attributes — but no image. An agent cannot present a product it cannot show. The shelf was thinner across the board but had the two fields that matter for a transaction.
Action. Photographed the 140 image-less products first, even though they had higher scores than the products below them. Transactable percentage rose 12 points in a week.
The lesson. Score is representation quality. Transactable is a gate. Fix gates before you polish.
Scenario C — A feed import that broke 1,200 products
Setup. Supplements brand, agent-ready percentage fell from 81% to 44% overnight.
Cause. A supplier feed re-import had mapped columns one position off. Vendor and product type wrote empty across 1,200 products; 210 also lost their price.
What the tab showed. 1,200 findings — one per product, not one per field — with the 210 price-less products escalated to critical because they'd become non-transactable.
Fix. Re-ran the import with corrected mapping. Readiness returned to 83% over two syncs, and every finding auto-resolved inside the 7-day window without manual closing.
Why it was caught in a day. Nothing on the storefront looked wrong — those products still rendered fine, because the theme reads different fields. The percentage on the Executive Pulse was the only signal, and it was unmissable.
Scenario D — Out of stock, not incomplete
Setup. Seasonal apparel merchant saw 340 critical findings appear at the end of a season.
What they said. Not "add a description" but "Out of stock — AI agents can't sell this right now."
Why the distinction matters. All 340 products had perfect data. Sending a catalog team to "fix" them would have wasted a week finding nothing wrong. The finding correctly identified this as an inventory state, not a data defect.
Action. Products genuinely discontinued were archived; the rest were left to resolve on restock. The team spent zero hours on false catalog work.
Scenario E — Downstream feed benefits
Setup. Merchant running Google Merchant Center and Meta Catalog with persistent disapprovals — roughly 18% of the catalog rejected on missing-attribute errors.
What happened. They cleaned up for AI readiness, not for feeds. The nine checks overlap heavily with what Merchant Center and Meta require: brand, GTIN/identifier, product type, image, price, description.
| Before | After | |
|---|---|---|
| Google Merchant Center disapprovals | 18% | 3% |
| Meta Catalog rejections | 14% | 2% |
| Agent-ready | 58% | 89% |
The point. Structured-data quality is one problem with many symptoms. Fix it once and the AI surface, the shopping feeds, and your on-site search all improve together.
Best practices
✅ Fix transactability before score. Price, image, stock. A product that can't be sold is worth more attention than one that's merely thin.
✅ Attack description length first among the rest. At 20 points it's the single largest check, and it's the one most catalogs fail.
✅ Bulk-generate rather than hand-write. Descriptions and meta descriptions are exactly the job for Content Studio and bulk operations.
✅ Watch the percentage after every import. A drop of more than a few points overnight almost always means a mapping or feed error, not gradual decay.
✅ Treat 90%+ as the target, not 100%. Archived, seasonal, and intentionally-hidden products will always sit below the line.
✅ Re-check after theme or app changes that touch product metafields — those can silently blank a source field.
❌ Don't optimise the score and ignore transactable. 100 products moved from 70 to 82 is worth less than 40 products made sellable.
❌ Don't pad descriptions to clear 120 characters. The check strips HTML and counts real text, and agents summarise what's there — filler makes you lose comparisons you'd otherwise win.
❌ Don't dismiss out-of-stock findings in bulk. They resolve themselves on restock; dismissing hides a genuine inventory signal.
❌ Don't assume a good-looking storefront means good data. This is the failure mode the tab exists to catch.
Plan tiers
| Capability | Free | Starter | Pro | Agency | Enterprise |
|---|---|---|---|---|---|
| Per-product readiness scoring | — | ✓ | ✓ | ✓ | ✓ |
| Band + transactable flags | — | ✓ | ✓ | ✓ | ✓ |
| Missing-field breakdown | — | ✓ | ✓ | ✓ | ✓ |
| Shop-level rollup on the Pulse | — | ✓ | ✓ | ✓ | ✓ |
| Findings in the Actions worklist | — | ✓ | ✓ | ✓ | ✓ |
| Filter, search, and sort the product table | — | — | ✓ | ✓ | ✓ |
| Bulk fix from Content Studio | — | — | ✓ | ✓ | ✓ |
| Readiness trend over time | — | — | ✓ | ✓ | ✓ |
| Multi-store readiness roll-up | — | — | — | ✓ | ✓ |
Frequently asked
Does this measure what ChatGPT actually sees? It grades the structured catalog data that feeds and markup are generated from — an honest proxy, and the thing you control. It is not a live query against any specific agent, which is why findings carry 0.7 confidence rather than certainty.
Why is description weighted highest? It's what an agent uses to decide whether your product answers the question, and to compare you against alternatives. A product with no description can be listed but rarely gets recommended.
Why 120 characters? Below roughly that, there isn't enough substance to summarise or compare. It's a floor for usefulness, not a target — most well-described products run several hundred.
My product has an image on the storefront but fails the image check. The check reads the product image field. Theme-rendered images, metafield images, and app-injected galleries don't populate it. Set a primary product image.
Does variant-level data count? Scoring is at product level, using the primary variant's price and SKU. A product with a priced primary variant passes even if some variants are incomplete.
How often does it re-score? On every catalog sync. Fix a field and the score updates on the next sync — no manual re-run.
Why did my percentage drop without me changing anything? Usually stock: products going out of stock become non-transactable. Check whether the drop is in transactable-% alone or in average score too — the first is inventory, the second is data.
Do out-of-stock products hurt my average score? No. Availability affects transactability only. Data completeness is scored independently.
See also
- Daily Ops overview — the hub and Executive Pulse
- Actions worklist — where readiness findings are ranked
- Revenue Leaks — out-of-stock as a money figure
- Bulk operations — fixing fields at scale
- Image SEO — image coverage and quality
- Barcode & GTIN — the identifier check
- Schema markup — structured data on the storefront side
- Google Merchant Center — the same fields, downstream