Editorial transparency
How and why we use AI
AI helps research, write, edit and maintain Digital Publishing News. Here is what that means for the work you read.
Why we use AI
A publisher’s work crosses advertising, search, email, subscriptions and software. Useful information is scattered across product documentation, research reports and company announcements. AI helps us examine that material, connect it to a publisher’s question, and develop explanations and examples.
Our aim is to make a useful publication with a maintainable reference library. Producing more pages is not a success measure by itself. A reader should leave knowing something they can use, with enough evidence to question our conclusion.
What AI contributes to this edition
OpenAI’s Codex has been used to research public sources, propose article topics, draft and revise articles, explain terms, construct illustrative calculations and diagrams, and write the site’s code. AI use is substantial; it is not limited to spelling or grammar checks.
Codex also reviews the work through different publisher roles, such as an advertising operator, newsletter manager or publication owner. These simulated reader reviews help identify missing explanations and weak recommendations. They are not interviews, independent experts or evidence that real readers found the site easy to use.
The website itself is built from saved content files. Its ordinary page generation, search filters, calculators and version checks use software rules rather than a live language model. Visiting a page does not ask an AI to generate a new answer.
How an article is developed
We start with a specific question a publisher needs answered. Research identifies the relevant primary material and its date. A product’s own documentation can establish how its feature is described; it cannot, by itself, establish that the feature will improve a publisher’s results.
The draft must add something useful: a worked calculation, a comparison, a decision procedure or an explanation of a common mistake. Numbers used to illustrate a situation are identified as hypothetical. We do not invent interviews, account screenshots, firsthand product tests or publisher performance data.
Review checks whether the headline’s promise is actually answered, whether the cited evidence supports the claims, and whether a reader can follow the reasoning without already knowing the terminology. The rendered page is then inspected for reading, navigation and presentation problems. The writing standard describes the full procedure.
What Shaun reviews, and what a review credit means
Shaun Connell is the Managing Editor. He confirmed that he reviewed all 35 articles in edition 0.4.0. The site preserves those reviewed copies and links to them from the article records.
Articles revised after that confirmation say that an earlier version was reviewed. An AI revision does not inherit a claim that Shaun approved the new wording. Staff Writers is the collective AI-assisted byline; it is not a set of invented human identities.
Editorial review applies to the version Shaun actually reviewed.
How we check facts and examples
Sources are attached to articles and definitions so readers can inspect the original material. When a claim is disputed, consequential or dependent on several facts, the research must examine appropriate corroborating evidence. Two pages repeating one announcement do not count as independent confirmation.
Calculations must show their inputs, units and comparison population. For example, comparing advertising prices only on sold impressions can hide the revenue lost when more opportunities go unfilled. We explain the denominator because it can change the decision.
Software checks catch missing records, broken internal links, inconsistent metadata and stale review versions. An accessible source URL does not prove a claim is true, and a passing test does not establish that an article is useful. AI can still misread a source or produce confident but mistaken reasoning.
How updates and corrections work
Article publication dates, material update dates and source-check dates serve different purposes. Adding a dictionary link does not make an old article new. Substantive revisions retain a history so readers can distinguish the current explanation from an earlier reviewed copy.
The local workbench can identify possible dictionary links and store a decision to accept or reject each one. It does not silently rewrite the surrounding text. Accepted links are tied to the article version being inspected.
Feedback records preserve the affected page and version. During a repair pass, Codex or an editor reproduces the problem, changes the source, and repeats the reader’s task.
What this means for readers
You should be able to judge the work by the answer, reasoning and evidence. The AI disclosure explains the process; it is not offered as proof that the conclusion is correct.
Our use of AI for production is separate from whether external search or training crawlers may access the site. The For robots page describes the edition and its readable content. It does not grant crawler permissions or guarantee that an AI service will cite an article.