Storemend Methodology: How the Shopify Audit Corpus Is Built

By the Storemend Team. Dataset version 2026-06-19 (Q2 2026). Refreshed quarterly.

Storemend is a research and audit operation. Every public number Storemend publishes traces back to one corpus of live Shopify storefronts, audited with the same ~140 checks a paying customer receives. This page documents how that corpus is built, how stores are classified, where the statistical floor sits, how often the data refreshes, and what each term means. It is the site-level companion to the report-specific State of Shopify 2026 methodology, and it covers the whole research operation rather than a single report cut.

There is no named author byline here by design. The authority is the corpus, the classification schema, and the reproducibility of the method, not a personality. Read this page as the receipts for everything else on the site.

What Storemend audits: ~140 checks per store

Every Shopify store gets the same ~140 deterministic checks, whether it is one of the corpus 1,091 or a single paid audit. The audit runs one method, not two, so the public research and the $39 customer report are the same instrument pointed at different inputs.

The checks fall into a handful of load-bearing groups:

Check group	What it inspects
Structured data	Product and Organization JSON-LD presence, required fields, offers block, aggregateRating, FAQPage schema
AI-shoppability	llms.txt at the apex, robots.txt AI-crawler policy, server-rendered vs client-injected schema
SEO foundation	Title tags, meta descriptions, canonical signals, sitemap, indexability
Performance	Render-blocking head content, page-builder JavaScript overhead, mobile speed signals
Trust and conversion	Social proof density, Open Graph metadata, checkout integrity, policy pages

Each check is deterministic: given the same live storefront, the audit returns the same result. There is no scoring model that drifts between runs and no human judgement call inside a single check. That is what makes quarter-over-quarter comparison honest.

The corpus: 1,091 Shopify storefronts

Storemend audited 1,091 live Shopify storefronts for the Q2 2026 corpus, and 1,077 classified cleanly to a primary cluster. The 14-store gap is stores with no dominant cluster after deduplication, expected at the current schema-version cut and reported rather than hidden.

The corpus is a curated sample of publicly accessible Shopify storefronts audited for the Q2 2026 cut, dataset version 2026-06-19. It spans eight verticals with a long-tail catch-all for stores that do not fit one named category. Seven named verticals carry enough volume to report; the remainder sit in the catch-all.

Metric	Value
Stores audited	1,091
Stores classified to a cluster	1,077
Total findings surfaced	8,409
Average findings per store	7.8
Checks per store	~140
Dataset version	2026-06-19 (Q2 2026)

Every percentage in a Storemend report is calculated against the 1,077 classified stores unless stated otherwise. When a number reads against the full 1,091, the text says so.

How classification works

Each storefront is assigned one primary cluster based on its highest-severity finding, then placed in an AI-shoppability tier. Classification is rule-driven off the deterministic check output, not a judgement call, so the same store lands in the same cell every time it is audited.

The pipeline is three steps:

Fetch and parse. The homepage and a representative product page are fetched from the live storefront surface and inspected for JSON-LD, Open Graph metadata, robots.txt directives, llms.txt presence, and rendered head content.
Score the checks. All ~140 checks run and produce a per-store finding stream. The Q2 2026 corpus produced 8,409 findings total, 7.8 per store on average.
Classify. Each store gets a primary cluster (its dominant failure pattern) and an AI-shoppability tier (invisible, transitional, or ready). Findings carry a severity of high, medium, or low.

The result is a grid of cohort cells: vertical by cluster by AI-shoppability tier. The single strongest signal in the Q2 2026 corpus is that the dominant cluster is identical across every named vertical: structured-data absence. The rate varies by category; the pattern does not.

The n=50 statistical floor

A vertical needs 50 or more audited stores before its rate is treated as definitive; below that, the number is reported as directional. The floor exists so a small sample never gets quoted as a hard finding.

In the Q2 2026 corpus, six named verticals clear the floor and one sits below it. Supplements, at 46 audited stores, is marked directional everywhere it appears. Its rate is published for completeness, not as a definitive category claim.

Vertical	Stores audited	Above n=50 floor?
Apparel	205	Yes
Beauty	171	Yes
Food and CPG	146	Yes
Home	122	Yes
Electronics	108	Yes
Pet	87	Yes
Supplements	46	No (directional)

Reporting a below-floor number transparently, and labeling it, is the honest posture. Hiding it or rounding it up to look definitive is not.

Per-vertical findings

Across every named vertical, at least six in ten audited stores ship no Product or Organization JSON-LD. Schema absence is category-wide infrastructure debt, not a single-store problem.

The table below pairs each vertical with its audited count, its schema-absent count and rate, and the number of stores classified to the invisible AI-shoppability tier. Supplements is directional (below the n=50 floor).

Vertical	Audited	Schema-absent	Schema-absent rate	Invisible-tier stores
Apparel	205	138	67%	143
Beauty	171	126	74%	128
Food and CPG	146	102	70%	108
Home	122	83	68%	88
Electronics	108	70	65%	76
Pet	87	53	61%	57
Supplements (directional)	46	31	67%	33

Category-specific breakdowns of what these gaps cost live in the matching guides: supplement store schema, electronics store mobile speed, and pet store shipping zones.

The corpus-wide headline

72.2% of audited Shopify storefronts are functionally invisible to AI shopping assistants, and zero hit the full AI-ready bar. The top tier was empty across the entire 1,091-store corpus.

Corpus finding	Value
Functionally invisible to AI shopping	72.2%
No Product or Organization JSON-LD	68.1% (733 of 1,077)
Stores at the full AI-ready bar	0
Total findings	8,409

"Functionally invisible" means the store cannot be cited by ChatGPT, Perplexity, or Google AI Overviews when a buyer asks for a recommendation in the store category, because the structured-data surface those assistants read is absent or incomplete. The full corpus write-up sits in the State of Shopify 2026 report.

Definitions

Precise terms, because the findings only mean something if the words behind them are fixed.

AI-shoppability. Whether an AI shopping assistant (ChatGPT, Perplexity, Google AI Overviews, and similar surfaces) can retrieve and cite a storefront when a buyer asks for a product recommendation. It is a function of structured data presence, crawl accessibility, and server-rendered content, not brand size or ad spend.

Schema-absent. A store that ships no valid Product or Organization JSON-LD on either its homepage or a representative product page. This is the single most load-bearing gap in the corpus and the one most operators assume their theme already handled.

Invisible tier. The lowest of three AI-shoppability tiers. A store in the invisible tier has no Product or Organization JSON-LD detected, so an AI shopping assistant has no structured signal to cite. The transitional tier has partial signal (for example, Product schema with three or fewer fields, or Organization schema without Product). The ready tier requires complete Product JSON-LD plus FAQ schema, llms.txt, and an AI-crawler policy. Zero stores in the Q2 2026 corpus reached ready.

Refresh cadence

The corpus refreshes quarterly, and each cut keeps the same structure so quarter-over-quarter comparison is honest from the start. The current dataset version is 2026-06-19; the next cut lands with the Q3 2026 cohort.

Because the audit method is deterministic and the classification schema is fixed, a store audited in Q2 and again in Q3 can be compared directly. A rate that moves between quarters reflects the market moving, not the instrument changing. The machine-readable corpus is available on request for research purposes.

Frequently asked questions

How many checks does a Storemend audit run? Roughly 140 deterministic checks per store, covering structured data, AI-shoppability, SEO foundation, performance, and trust and conversion. The public research corpus and the $39 paid audit run the same ~140 checks.

How large is the corpus? 1,091 Shopify storefronts were audited for the Q2 2026 cut, and 1,077 classified to a primary cluster. The corpus produced 8,409 findings, 7.8 per store on average.

What is the n=50 statistical floor? A vertical needs 50 or more audited stores before its rate is treated as definitive. Below that, the number is reported and labeled as directional. In Q2 2026, supplements (46 stores) sits below the floor.

What does "invisible to AI shopping" mean? The store has no Product or Organization JSON-LD for an AI shopping assistant to cite, so it cannot appear when a buyer asks ChatGPT, Perplexity, or Google AI Overviews for a recommendation. In the Q2 2026 corpus, 72.2% of audited stores were in this position.

How often does the data refresh? Quarterly. The current dataset version is 2026-06-19. Each cut keeps the same classification schema so quarter-over-quarter comparison holds.

The Storemend audit runs the same ~140 deterministic checks on a live Shopify store and returns a prioritized fix list. $39 one-time, no subscription, 30-day no-questions refund.

Cite this data

Storemend, State of Shopify 2026, n=1,091, storemend.com/methodology

Methodology: n=1,091 Shopify storefronts audited (1,077 classified), ~140 deterministic checks per store, dataset version 2026-06-19 (Q2 2026), refreshed quarterly.

By the Storemend Team. Dataset version 2026-06-19. Last updated 2026-07-10.

Cite this block

Storemend, State of Shopify 2026, n=1,091, storemend.com/methodology. Methodology: n=1,091 Shopify storefronts audited (1,077 classified), ~140 deterministic checks per store, dataset version 2026-06-19 (Q2 2026), refreshed quarterly.