By The Tool Trial Editorial Team
Last reviewed: August 23, 2026
Effective date: August 22, 2026
Why this page exists
You should be able to tell how something was made before you act on it. This page states where AI is used in producing what we publish, what it is not allowed to do, and what a person is responsible for regardless.
It is a disclosure, not a defence. If we would be uncomfortable telling you where a model was involved, that would be a reason to change the process, not to leave it out of this page.
How topics are chosen
We write about decisions a small service business actually faces and where the available guidance is thin, generic, or sold to them.
A topic has to pass three tests: a reader would recognize the problem immediately; we can add something original rather than restate what already exists; and we can support what we say from official documentation or from work we recorded ourselves. If the third test fails, we do not publish. We do not fill the gap with a confident guess.
Where our facts come from
Two sources, and nothing else. Official documentation — when we describe what a platform allows, recommends, or prohibits, it comes from that platform’s own help pages, and we record the date we checked. Our own design work — the workflows, checklists, and frameworks are ours, we can explain each one end to end, and we label it as our editorial judgment rather than as an industry standard.
Where a source is silent on something, we say so, instead of letting our recommendation borrow its authority.
Where AI helps
We use AI as a tool. In practice it helps with:
- Initial research — finding what exists on a topic and where the official documentation lives
- Organization — grouping material, spotting overlaps, structuring a comparison
- Outlines — proposing an order for a piece before it is written
- Drafts — producing text that a person then verifies, rewrites, and takes responsibility for
- Language review — tightening wording, catching repetition and awkward phrasing
- Synthetic scenarios — writing the fictional examples used to exercise a process
That assistance is real, and it saves time.
What AI is not allowed to do
AI does not publish anything. There is no automation between a model’s output and a live page. Publication is manual and explicitly authorized, one page at a time.
AI output is not accepted as fact. Anything a model produces is checked against a source before it reaches a published page. A model’s confidence is not evidence, and a fluent sentence is not a citation.
Unverifiable material is removed, not softened. If we cannot support a claim, it comes out. We do not keep it and hedge it with may or often.
The model has no vote in what gets published, and no role in approving it. It does not choose topics, it does not settle conclusions, and it takes no part in the approval step.
We do not automatically publish AI-generated replies or content to third-party platforms. When we describe an AI-assisted workflow for a reader, it produces a draft that a person reads, edits, and posts themselves.
Claiming our content is written without AI is a claim we will never make, since it would be false. Nor will we dress up what a model wrote as something we experienced.
Human review is mandatory
Every page is read in full by a person before it publishes, and that person takes responsibility for it. No model stands in for that reading, and a deadline is never a reason to skip it.
Approval is a named act on a specific text, not a general blessing. Waiting is a normal outcome: a piece is held when its sources cannot be verified, when the records are incomplete, or when the conclusion outruns what the evidence supports.
Invented scenarios, and their limit
Much of our work uses fictional material — invented reviews, invented customers, invented businesses — written to exercise a process under conditions we control.
Our synthetic work uses no real customer reviews, no personal data, no private business data, and no real transactions. No real customer is drawn into it, and no working business is put at risk to see what happens. Because the inputs are invented, we can publish them alongside a piece so you can run the same inputs yourself.
A synthetic example is not, by itself, evidence of real-world performance, and we do not present it as one.
Internal pilots are not evidence
We sometimes run informal pilots — rough passes used to shape a protocol before it is finalized.
A pilot is never official evidence. Nothing from one is written up as a result, nothing from one substitutes for a documented test, and no figure it produced appears as a finding.
Where we have not run tests
Where no test has been run, we say so. Evidence-dependent work is described as planned until its methodology, records, limitations, and conclusions have been reviewed, and until that happens the piece publishes nothing.
We do not claim to have purchased, tested, reviewed, or deployed a tool unless reviewed evidence supports the claim. Reading a product’s documentation is research, and we call it research.
Corrections
Material factual errors get corrected in the open, not edited away. When the substance of a page moves, that page states what changed.
We have not invented a correction history. Corrections will be recorded on the pages they affect, starting when there are published pages to correct.
Independence and commercial relationships
Editorial conclusions follow documented criteria and the evidence available to us. No outside party approves, reviews, or edits what we publish.
The Tool Trial currently has no advertising, affiliate, sponsorship, or paid relationship of any kind, and receives no compensation from any company it writes about.
If a material relationship is introduced later — advertising, an affiliate link, free access, sponsorship — it will be disclosed in the content it affects, visibly rather than buried, and it will not determine an editorial conclusion, ranking, or recommendation. That rule is written down now, before anything tests it.
What this page does not promise
We do not guarantee that our content is free of errors. Careful sourcing, human review, and visible corrections reduce mistakes; they do not eliminate them.
What you have read is a description of our process. It is not legal advice, and no platform has adopted it as policy.
Related pages
The standards governing what gets published are set out on our editorial policy page, and the method behind a documented test on our methodology. You can reach us through our contact page.