The AGMM editorial team
How we write and check our guides
The AGMM editorial team writes the guides that carry its name with a research process we built to catch our own mistakes: primary sources first, every claim tied to the source it came from, a rule check that rejects anything unsupported, and a person who reads each guide before it is final.
How it moves
From topic to published guide
Nine stations, a gate that rejects, and a loop that sends a reported mistake back to the sources.
Before a word is written
Topic
From our own work, or a daily scan of news and search demand.
Sources
Three or more, from two independent organisations, each read in full and dated.
Claim ledger
Every source on file. Each fact gets a numbered citation.
Drafted, then checked twice
First model drafts
It writes the guide from those sources.
Second model checks
A model from a different family tests every cited sentence against its source.
Rule gate
Rejects an uncited figure, a broken link, too few sources or a stock phrase.
Fails: repaired or dropped, never published
Labelled, read, published
Examples labelled
Invented UK businesses are marked as examples, and their figures as estimates.
A person reads it
Someone at AGMM reads the guide before it counts as final.
A person decides here
Published
With the date it was last checked and its sources listed.
Corrections loop Returns to station 02, Sources
A reader reports a mistake. We check it against the sources, fix the guide, update the checked date and note any material fix at the end. The check goes back to the sources, never to memory.
Gilt ring: a person makes the call at this station
What we choose to write about
We write for owners and operations leads of established UK businesses. A guide earns its place when people in that position are searching for the term or the story and a plain answer would help them decide something: whether to care, what it costs, what can go wrong.
Topics come from two places. Our own work, where the same question comes up again and again. And a daily scan of news, research and search demand that spots new AI terms and business stories early. We never write a second guide on a subject we already cover, and we favour lasting explanations over news, because news is out of date within days.
How sources are chosen
Primary sources come first: the company’s own announcement, the paper, the regulator, the government publication. Reporting comes second, and only when it adds something the primary source does not.
Every guide needs at least three sources from at least two independent organisations before a word is written. Each source is fetched and read in full, and the date it was read is recorded. If we cannot find that much, the guide is not written.
How every claim is checked
Each source is kept on file as a claim ledger. A factual sentence carries a numbered citation to the source it came from, and the numbers you see in the text link to that source in the list at the end.
A first model drafts the guide from the sources. A second model, from a different family, then checks every cited sentence against the source it cites and removes anything it cannot support.
Last comes a rule check that is designed to fail. It rejects a guide with a figure that has no citation, or a figure that does not appear in the source it cites; a citation to a source we do not hold; a broken link; too few independent sources; or any of the stock phrases and em dashes we do not write. A guide that fails is repaired or dropped. It is never published with a known failure.
Worked examples are labelled
Most guides include a worked example: a UK business invented to show how the idea plays out. We say so plainly every time. Its figures are our estimates, marked as estimates, and they are never presented as a client result.
Diagrams, and why there are no title cards
Each guide has one animated diagram that explains its central idea, drawn from the guide’s own content: the labels, figures and examples in it are the ones in the text. It is real text and drawing, readable without the animation and by a screen reader.
We do not use AI-generated pictures. Where a real photograph of the actual subject helps, we use one with a licence that allows it and credit the photographer in the caption.
The voice rules
UK English, short sentences and plain words. Concrete examples from across a business, not only sales. No hype, no em dashes and none of the stock phrases that make writing sound machine-made. AGMM is an AI consultancy, and the guides say so only where it helps the reader.
Who reviews a guide
Guides the research engine drafts are marked for review when they are written, and a person at AGMM reads each one before it is treated as final. The date under the title is the date it was last checked. Guides written by Sam Wall or Adam Gillett carry their own names.
How corrections work
If something in a guide is wrong, tell us. We check it against the sources, correct the guide and update the checked date. A material correction is noted at the end of the guide, with what changed. We do not silently rewrite a claim that turned out to be wrong.