Skip to content
Published Authored byBilly Reiner

Foundation · Methodology

Methodology

Every editorial page here carries three separate dates — Published, Updated, and Last checked — and they are not interchangeable. This page says what each one means, what has to be true before either of the last two is allowed to move, and what happens when a source we cited changes underneath us.

The editorial standards page covers what counts as a citable source and how a citation is rendered. This one covers verification and dating: the sequence a claim goes through before it ships, the four classes of claim and the rule for each, and the categories of claim this site refuses to publish at all.

Three dates, three meanings

Published is when the page first shipped and it never moves. Updated moves only when the substantive body of the page changed. Last checked moves only when a source was re-opened and re-stamped. A page can be Updated without being re-checked, and re-checked without being Updated, and this site shows both rather than collapsing them into one flattering freshness number.

  • Published — the publishedISO prop, emitted as datePublished on both the Article and the WebPage node. Set once, on the day the page first shipped, and never touched again. A page that gets rewritten from scratch keeps its original publication date; if it is genuinely a different page, it gets a different URL.
  • Updated — the modifiedISO prop, emitted as dateModified. It moves when the substantive body of the page changed: a claim added, a claim removed, a section rewritten, a number corrected. It does not move for a retitled <title>, a rewritten meta description, a schema block, a new internal link, or a CSS change. That is a standing commitment and it is the rule this repository is actually built under, not a description of good intentions.
  • Last checked — the value the source manifest derives from the newest date across the page's own source records, emitted as lastReviewed on the WebPage node. It moves when a source was re-opened and re-stamped. There is no author-settable field behind it; the derivation is described on editorial standards.

How a claim gets verified before it ships

The sequence is fixed: write the claim, open the primary source, find the specific sentence or field that supports it, add a source record carrying that publisher, title, URL and the date the source was opened, then place the citation immediately after the clause it backs. If the supporting sentence cannot be found, the claim is deleted rather than softened.

The step people skip is the third one. It is easy to remember that a vendor doc “says something like this” and cite the doc as a whole; it is harder to locate the exact sentence and discover it says something narrower. That is where most would-be facts die here. The note field on the source record exists to hold what the source actually said, so the check can be repeated later without re-reading the whole document.

The last rule is the expensive one and it is the one that makes the rest credible: if the source cannot be found, the claim is removed rather than hedged. No “widely reported”, no “industry consensus suggests”, no “studies show” without the study. That constraint is why the founder page carries no install counts and no client testimonials — not because there is nothing to say, but because nothing about our own results could be cited to anything except us asserting it.

Citations go immediately after the clause they back, inside the sentence, rather than at the end of the paragraph. A paragraph-terminal footnote tells you the paragraph came from somewhere; a clause-level one tells you which claim came from where. The rendering side of this is on editorial standards.

Four kinds of claim, four rules

Not every claim is verified the same way. Vendor-documented mechanics get the vendor doc and the date it was read. Dated events carry the event's own date separately from the read-date. Claims about this site's own practice are verifiable by reading the site and get no external citation. Claims that cannot be sourced are omitted, and the categories that get omitted are named below rather than left implicit.

  1. Vendor-documented mechanics. How Shopify Catalog eligibility works, which fields a product page exposes, what a theme emits by default. Rule: cite the vendor documentation and stamp the record with the date the doc was read. If the doc is ambiguous, the prose says it is ambiguous instead of resolving the ambiguity in whichever direction suits the argument.
  2. Dated events. Shopify gave Agentic Storefronts its own admin page on 11 May 20261; Google restricted FAQ rich-result eligibility on 7 May 20262. Rule: cite the changelog or the vendor policy page, and keep the event's own date distinct from the date the page was opened. Both examples show the split — the Shopify record is stamped 2026-05-22 because that is when the changelog was read, while the event it reports happened on 2026-05-11, and the event date lives in the record's title rather than in its date field. The Google record happens to share both dates, which is a coincidence, not a convention.
  3. Self-descriptive claims about this site's own practice. Everything on this page and on editorial standards falls here. No external citation is possible for “this is what our build does”, and attaching an irrelevant one would be worse than attaching none. Rule: say plainly that the claim is self-descriptive, and describe only what is verifiable by reading the rendered site or viewing source. Where a claim of this class describes code, it is written after reading the component, not from memory of what the component was supposed to do.
  4. Claims that cannot be sourced. Omitted. Specifically and permanently: no proprietary ranking study, no traffic-lift percentages, no client-result numbers, no “X% of Shopify stores” statistics. This site publishes none of these, and this section exists to put that on the record so the absence reads as a decision rather than an oversight.

What this site does not measure

ShopifyRanked runs no crawl of the Shopify merchant population, no citation-share panel, and no A/B ranking test. Nothing published here is the output of original measurement, and no sentence on this site should be read as if it were.

Where a third party has run such a study and it is worth citing, the study is named and its methodology limitation is stated in the prose rather than buried in a footnote. The live example is the skincare category-capture guide, which names the 5W AI Visibility Index as its source and describes what the index actually is — citation share measured by sampling ChatGPT responses. That is a sampling method with real limits, and a reader deciding what to do with the number needs to know it is a sample of one model's outputs rather than a census of anything.

The practical effect: a claim on this site is either a vendor's documented behaviour, a dated event, a description of our own practice, or a named third party's finding with its method attached. There is no fifth category where a number appears without a provenance.

Corrections and re-checks

When a cited source changes, three things happen in order: the source record's date is re-stamped to the day the source was re-opened, the claim is rewritten to match what the source now says, and modifiedISO moves — because in that case the body genuinely changed. Corrections go to [email protected] with the subject line "Correction".

The order matters. Re-stamping a record without re-reading the claim is how a page ends up displaying a fresh check-date over stale prose, which is worse than displaying an old one honestly. So the re-check is the claim-level unit of work, not the record-level one: opening the source means re-reading every sentence on the page that leans on it.

When a source is retired or moves, the record keeps the original publisher and title — that is what makes the change traceable — and the prose states what changed rather than silently swapping in a replacement URL. A vendor quietly rewriting a help page is itself a fact about the field, and pretending the old version never existed throws that fact away.

To report one: include the URL of the page and, where you can, the source that contradicts it. A correction arriving with a primary source attached gets fixed the same day. Everything else about reaching a human — response time, which subject line routes where, the registered entity — is on the contact page.

The same facts, machine-readable

The three dates are not display-only. datePublished, dateModified and lastReviewed all ship in the page's JSON-LD graph, reviewedBy resolves to the same Person entity as author, and the Article node is bound to a WebPage node by @id — so an engine parsing the markup gets the identical dates a human reads in the manifest header.

The binding is what makes it checkable. Each page emits one @graph in which the Article node's mainEntityOfPage points at the WebPage node, and the WebPage node carries lastReviewed plus a reviewedBy reference to the founder Person. Those two properties live on the WebPage rather than on the Article because that is where schema.org scopes them; putting a review date on a CreativeWork is out-of-domain and validators say so.

reviewedBy resolves to the same Person entity as author, and that is not an oversight either. This is a single-author site: the person who writes the guide is the person who re-opens the sources. Pointing the reviewer at an invented second name or an editorial board that does not exist would be exactly the kind of unsourceable claim §03 refuses to publish. If a second reviewer ever exists, they get their own entity and their own page.

View-source on any editorial page to read the graph rather than taking this paragraph's word for it. The site also publishes an llms.txt manifest at the root, which lists the pillar and foundation URLs for models that read one at inference time.