How We Research, Write, and Keep Every Article Honest

Most reference sites tell you what they publish. Very few tell you how. This post is the second kind. Below is the actual pipeline every article on myAgents.PLUS passes through before it reaches you, including the parts that fail.

Where a topic comes from

We do not pick topics from a keyword tool and work backwards. A topic enters the queue in one of three ways. The first is reader demand: when people ask our assistant about a subject we have no page for, that question is logged as a suggestion with a running count. Subjects asked about repeatedly rise to the top of the queue. The second is hub coverage: each knowledge hub has a shelf of subjects it needs to be genuinely useful, and gaps in that shelf are filled deliberately. The third is currency, where an existing subject has changed enough that the old page is no longer accurate.

How a draft is built

Drafting is AI-assisted and we say so on every page. The draft model receives the subject, the hub it belongs to, the angle we want, and a structural brief that requires a minimum of twelve distinct sections. Those sections are not twelve headings over the same paragraph rephrased. The brief demands specific shapes: a definition, an origin, a mechanism, a comparison against adjacent concepts, a timeline of how the subject changed, common misconceptions, practical implications, limits and open questions, and a set of references. If the model returns fewer sections, or returns sections whose bodies are empty, the draft is rejected outright and never sees a URL.

The quality gate

This is the part most sites skip. Every draft is measured before publication and must clear a hard floor: at least 1,400 words of visible body copy, and at least two original visual elements such as a comparison table, a timeline diagram, a concept map, or a labelled statistics panel. Word count is measured on rendered text only, so headings, captions, and navigation do not inflate the number. A draft that misses either threshold is held back rather than published thin. We would rather have a smaller library than a padded one.

The visual requirement exists because prose alone is a weak way to explain structure. A comparison across five attributes belongs in a table. A sequence of events belongs on a track with nodes. These figures are drawn from the article's own data rather than pulled from a stock library, which means they are unique to the page and actually correspond to what the text says.

Human review

An AI-assisted draft is not a published article. Every page carries an editorial byline, and the named editor is accountable for the claims on it. Review focuses on the things automated generation is worst at: claims that sound authoritative but are unsourced, dates that drifted, comparisons that are technically true but misleading, and definitions that quietly contradict the field's normal usage. Where a claim cannot be supported, it is cut rather than softened.

Articles are living, not finished

A published article joins an evolution queue. Periodically it is reassessed: are the facts still current, has the subject moved, are there now adjacent pages worth linking to, has a section become thinner than the rest of the library? Pages that need work are rewritten in place and the version number increments, which is why you will see a version badge and a last-updated date on every article. A reference page that was written once and never touched again is a liability, not an asset.

What we archive

We regularly audit the library against the current standard rather than the standard in force when a page was written. When the floor rose to 1,400 words plus two figures, a large batch of older pages no longer cleared it. Those pages were pulled out of circulation immediately rather than left up, and they are being rewritten to the current standard one batch at a time. Anything that still fails after a rewrite stays unpublished. Removing weak pages is unglamorous and it costs us traffic in the short term, but a library is judged by its worst page as much as its best.

What we will not do

We do not publish pages that exist only to catch a search query. We do not spin one article into six near-duplicates targeting variant phrasings. We do not present AI-assisted work as though a human typed every word, and we do not present human-edited work as though no automation was involved. We do not invent statistics, quotes, credentials, or sources, and if a figure cannot be traced to something checkable it does not appear.

If you find a page that falls short of any of this, tell us through the contact page. Corrections are made at the source, on the page, with the date visible.