Skip to content

Methodology

How this site produces what it publishes, so that the claims on every other page can be checked rather than trusted.

Last updated

This page describes how AIOF produces what it publishes. It exists so that the claims on every other page can be checked rather than trusted, which is the only durable difference between research and content marketing.

How we label figures

Every number on this site falls into one of four categories, and pages say which.

Sourced
Published by a named party and linked. Platform fees, GPU rates, referral terms. If we cite it, the link is on the page and you can check it in a minute.
Third-party estimate
Modelled by someone else, with known error bands. Traffic estimates and market sizing fall here. We name the provider and the retrieval date, because these figures move and they are not measurements.
Modelled
Our arithmetic on top of sourced inputs, shown to demonstrate the shape of a calculation. Labelled as modelled every time. Not a finding.
Measured
Something we ran and recorded ourselves, published with its method. We hold this to the highest bar and consequently we publish less of it than we would like.

What we will not do

  • Invent a case study. No composite accounts, no illustrative operators presented as real, no revenue screenshots we cannot attribute.
  • Publish an income claim as a headline. The genre exists to sell courses and it is incompatible with being useful.
  • Generate pages at scale. Coverage is the target, not volume. A page we have not checked does not go up.
  • Reach a verdict before stating the criteria. Comparison pages set out what they are judging on first, so the reasoning is auditable rather than the conclusion being taken on trust.
  • Summarise a source we have not read. Reference pages carry rows saying check current terms rather than a confident guess.

How we use AI in production

Directly, and we would rather say so than have you wonder. AI assists with drafting, structure and editing on this site. It does not supply the substance.

The distinction is the whole point. Language models are good at prose and cannot know what a platform's payout screen looked like last Tuesday, what a training run cost, or which of two workflows produced more publishable images. Those are the things this site is for, and they come from sources and from work. Google's own guidance evaluates content on usefulness rather than production method, and we think that is the correct standard to be held to.

Corrections

Errors are fixed in place, with a dated entry in the revision history on the page itself and a line on the updates page. We do not silently edit a claim and leave the page looking as though it always said the new thing.

If you find something wrong, tell us. Specific corrections with a source get fixed quickly. This is the cheapest way for readers to make the site better and we would rather hear it than not.

Freshness

Platform mechanics and pricing in this niche change monthly. Pages that depend on those facts carry a fact-checked date separate from their published and updated dates, because the useful question is not when we last edited the prose but when we last read the source.