Fact-Checking Policy
This page describes how a piece gets checked before it goes up. It is written for the reader who wants to know whether a number here was verified or copied, and it is specific about where our beat goes wrong, because a general promise to be careful is worth nothing.
Last updated: August 2026.
Who checks a piece, and when two editors are required
A draft arrives with a source noted against every factual claim: a link, a document, a dated screenshot, a log file, or the name of the person who said it. Editors read the draft against those sources rather than against the prose, which is slower and catches more. A sentence with nothing behind it gets cut.
A second editor is required rather than optional in six cases: when we report that a named company was breached, when we say a company misled its customers, when a piece carries a pricing table, when it dates an outage, when it reaches a verdict that could cost a company sales, and when the subject is a former employer of anyone involved.
Primary sources, links, and what does not count as evidence
We go to the document rather than the description of it. The clause in the terms of service, not a summary of it. The invoice, not the price on the landing page. The regulator’s filing, not the trade-press write-up. The engineer on the record, not somebody who spoke to the engineer.
Some things look like evidence and are not. A marketing page is not evidence of performance. Another outlet’s article is not the source of a fact, so we go to their source or credit the claim to them. A security vendor’s press release is not an independent severity rating for a flaw in a product it sells protection against.
Claims are linked to their sources inside the article, and the link points at the primary thing wherever one exists. When the source is a page that changes underneath us, such as a checkout page or a status page, we quote the exact wording and record when we captured it. Where a document is not public, the article says so and describes it as precisely as we can.
Prices, uptime and other people’s studies
Every price is captured with the term, the prepay requirement, the currency, whether tax is included, and the renewal rate, and checked from more than one location, because checkout pages quote different numbers by country. Uptime promises are read rather than repeated. A figure quoted to three decimal places, measured per calendar month, with maintenance excluded and remedy limited to a credit you must request in writing, is a different product from the percentage in the advertisement.
For studies we look for who paid, the sample size and how it was recruited, and whether the question asked supports the headline. A benchmark commissioned by one of the hosts in it is described that way in the sentence, not in a note at the bottom. Our own testing is published with the configuration and the number of runs, because one measurement on a shared server measures the neighbours as much as the host.
Images, screenshots and video
Reverse image search happens before use. Photographs of flooded or burning datacentres circulate for years and get re-attached to each new incident, and a stock photograph of a server aisle is not a picture of the facility in the story. We check metadata where it survived and ask the sender what they were standing next to.
Screenshots of status pages, checkout pages and support transcripts are either captured by us with the time visible, or requested in original form rather than as a cropped picture of a picture. Personal data in reader screenshots is blacked out. Where an image is illustrative rather than evidence, the caption says so, and we publish no generated images of real places, people or incidents.
Reader reports, social posts and an outage happening now
A viral post saying a host is down is a lead rather than a fact. Site owners misdiagnose their own resolver, an expired certificate, or a DNS change they made the previous evening, and one loud account is sometimes one broken laptop. Before an incident becomes a story we want checks from more than one network, timestamps from readers in different places, and monitoring output rather than a description of a spinning browser tab. We ask before quoting a reader by name.
The pressure to publish first on a developing incident is real, and our answer is to publish thin and honest rather than fast and confident. An early outage post says what we confirmed, when we confirmed it, what we asked the provider, and what we do not yet know, and it is marked as developing.
The same standard whoever is making the claim
A claim gets the same treatment regardless of who benefits from it being believed. A provider’s incident statement, a security vendor’s advisory about that provider, a competitor’s briefing against them, and an angry reader’s email all meet the same question: what can you show us. The firm selling the product that would supposedly have prevented the outage is not a neutral witness.
That applies to us. The people who run this site worked inside hosting companies and have opinions about some of them, so those companies get the same test and a disclosure note, as set out in our editorial standards. We sell no position and take no affiliate commission, so nobody buys a shortcut through verification, which is documented on our ownership and funding page. Policy reaches this beat through domain governance and breach law, checked the same way whichever party is arguing.
When the checking fails anyway
It sometimes does. At that point it stops being a verification question and becomes a correction, handled under our corrections policy: fixed in the original article, with a dated note, decided by our managing editor rather than the writer, and logged with where the error came from.
We work to the principles in the IFCN code: non-partisanship, transparency about sources and funding, an open methodology, and a corrections practice that is published and used. We are not a signatory and do not claim to be. If you think we failed to do something on this page, quote the sentence and write to support@hosting123.org.
