Marketing publishing has a numbers problem. Most cited stats are old, secondhand, or invented. We do better than the average. These are the rules.
Sources
We use four kinds of sources, in this order of preference:
- First-party data from accounts we operate or audit. Anonymized, never re-identifiable.
- Vendor-published data from primary sources (a Google blog post, not a screenshot of one in a marketing deck).
- Peer-reviewed research for psychology, statistics, and behavior claims.
- Industry surveys from named publishers with disclosed methodology.
We don’t cite anonymous LinkedIn posts, screenshots of unsourced charts, or “according to a recent study” without naming the study.
Sample sizes
When we report a percentage, we name the source it comes from and, wherever that source gives one, the sample size behind it. “55% of business-to-business (B2B) teams (n=412)” beats “55% of B2B teams.” We don’t publish a figure we can’t attribute.
Claims & citations
Every concrete claim links to the source. Three categories:
- Direct quote/number → linked to primary source.
- Synthesized claim from multiple sources → linked footnote listing all of them.
- Opinion/argument → unlinked, signaled with we think, in our experience, etc.
Charts
All charts include: axis labels, units, time period, source, and sample size where applicable. If a chart is internal Mediovsky modeling, the caption says so explicitly.
We do not truncate y-axes to exaggerate differences. We do not use 3D effects. We do not use rainbow color scales for ordinal data.
Modeled numbers
When we model something (e.g., “signal loss across a 90-day conversion window in Performance Max”) the article explains the model in plain English, lists the assumptions, and shows the result alongside the inputs.
We never present a modeled number as if it were a measurement. The word modeled appears in the caption.
How we use AI
Every article on Mediovsky is written, edited, and signed by a human, who holds editorial responsibility for it. We use AI tools in the process, and we would rather say exactly where than leave you guessing.
Where AI is used in the writing. Gathering and summarizing source material before a piece is written. Suggesting structure and angles. Checking drafts for weak claims, repetition, and broken logic. None of that reaches you unread.
Where it is not. No article is published as model output. Every number is verified by a human against the primary source that contains it, under the rules above. Every source link is opened. Every sentence is read and edited before it ships. An AI tool has never been the last thing to touch an article.
The images are AI-generated, and we say so. Every illustration on this site is made with an image model: the wide banner at the top of an article, the square thumbnail you see in listings and on shared links, and the diagrams inside a piece. They are illustrations, not photographs, not evidence, and never a document of a real event, a real person, or a real product screen. Where a piece needs a real artifact, such as a screenshot or a chart of actual data, we use the real artifact and label it in the caption.
The legal position, plainly. Article 50 of the EU AI Act, which applies from 2 August 2026, requires disclosure for AI-generated text published to inform the public. It exempts content that has undergone human editorial control and for which a person holds editorial responsibility. Both are true here, so the obligation does not apply to our articles. We publish this section anyway, because you deserve to know how the work is made.
When we get it wrong
We post corrections publicly (see corrections). Three rules:
- The correction goes on the original article, dated, with a note explaining what changed.
- The original wording stays visible, struck through. We don’t quietly rewrite history.
- If a number drove the headline of a piece and the number was wrong, we restate the conclusion at the top of the article.