Publishing Principles
Last updated: August 8, 2026
Everything WISLR publishes comes from data we gathered ourselves and questions we thought of ourselves. We do not summarize other people’s studies and present the result as research.
Where the numbers come from
Our findings are first-party. They come from three sources.
Servers we operate or are given access to. Request logs, edge logs and origin response records. When we report how often an AI crawler fetches a resource, that count comes from a log file, not from a vendor dashboard or a published estimate.
Live sites we measure directly. When we report what software a company runs, we loaded the site and read what it served. Findings are dated because a site can change the week after we look.
Our own client work. Always anonymized. Store names, domains, IP addresses and infrastructure details are removed before anything is published, including from screenshots and sample records.
When we cite someone else’s number, it is attributed in the sentence that uses it and it is labeled as theirs. Industry figures from firms like Bain and BCG appear as context around our measurements. They are never the finding.
Every claim is confirmed a second way
A single signal is a hypothesis. Before a measured claim is published, we confirm it from a source that cannot be shaped for our benefit, ideally from a different network or a different client.
This test removes findings regularly, including ones we wanted to be true.
In our August 2026 study of luxury ecommerce platforms, three findings were cut for failing it.
A result showing that half of luxury brands block AI crawlers described the conditions we measured under rather than anything the brands had done. Repeating it under ordinary conditions produced the opposite answer.
A finding that 32 luxury domains could not serve HTTPS was traced to wrong domains in our own compiled list, one of them a misspelling we had introduced. The brands were fine.
A headline example naming two brands as running a particular platform was wrong about both. The error compounded: a brand identified incorrectly is by construction excluded from the comparison set, so every misidentification inflated the statistic it was being used to illustrate.
None of the three reached publication.
We fact check
Measurement and fact checking are separate jobs. Confirming that a server returned a given response says nothing about whether we named the right company, the right owner or the right date. Every assertion in a piece is checked against a primary source before it runs.
Entities. Company names, brand ownership, group structure and acquisitions are verified at the time of writing rather than from memory. Ownership moves. In the luxury study, our list had Stuart Weitzman under Tapestry, which had been true until Caleres acquired the brand. The brand’s own site settled it.
Software identification. A site can mention any product name anywhere in its code, so finding a platform named in a page proves nothing on its own. We credit a platform only once we have established that it is actually serving the site. During the luxury study, one of the largest houses in the world carried a stray reference to a commerce platform in its own configuration, enough for a careless read to file it under that platform. Nothing on the site runs on it.
Domains. We confirm that a domain resolves, that it belongs to the company we think it does, and that it is not parked. One entry in that same study pointed at an industrial supplier that happened to share a fashion label’s name, and another was a misspelling we had introduced. Both would have become data points.
What we actually landed on. A measurement is only as good as the page behind it. Sites redirect, to a regional variant, to a corporate site, to a parked domain, sometimes to a vendor’s own site. Any of those will be recorded as though it were the company we set out to measure unless someone checks. Anything that finishes somewhere unexpected is set aside for a person to look at rather than counted.
Dates and figures from elsewhere. Announcements, filings and third-party statistics are traced to the original document, not to an article describing it.
Quotes and characterizations. We do not paraphrase a company’s position from secondary coverage.
Where we cannot verify a detail and the piece works without it, the detail goes. Where the piece depends on it, the piece waits.
What is published alongside a finding
The denominator. Every percentage is published with the count behind it.
The date of measurement. What a company runs in August may not be what it runs in November, so every finding carries the date we observed it.
What we could not resolve. Cases we failed to settle are reported as unresolved rather than assigned to whichever answer looked most likely. Where a range depends on how the unresolved cases fall, we say which end excludes them.
What the data cannot tell you. A pattern in a scan shows what companies did, not why they decided it. Where we are inferring, we say we are inferring.
Authorship
Every research piece carries a named human author who did the analysis and is accountable for it. We use AI tools in our work, for drafting, for code and for data processing. No article is published without a human verifying each factual claim against the underlying data.
Corrections
When we get something wrong, we correct it on the page, dated, describing what changed. We do not quietly edit a number and leave the article looking as though it always said that.
If you believe something we published is wrong, write to [email protected] with the specifics. Evidence that contradicts our finding is the most useful thing you can send us.
Independence
WISLR sells AI visibility software and consulting. Our research often concerns platforms and services that compete with, partner with, or could be sold alongside what we offer.
We do not accept payment to produce a finding, and no vendor reviews our research before it is published. Where a study touches a company we have a commercial relationship with, that relationship is disclosed in the study. Where the honest answer is bad for our commercial interest, we publish the honest answer.