By The Tool Trial Editorial Team
What this page is for
You should know how something was made before you act on it. This page explains how we produce what we publish: where our guidance comes from, how we design a test when we run one, where AI helps, and what has to happen before anything appears on the site.
It describes our working method. The rules that govern our writing — sourcing standards, corrections, independence, disclosure — live separately, in our editorial policy.
Where our guidance comes from
Most of what we publish is not a test. It is guidance: a workflow, a checklist, a decision framework built for a small business that has to act this week.
That material comes from two places. The first is official documentation. When we describe what a platform allows, recommends, or prohibits, we take it from that platform’s own help pages and record the date we checked it. The second is our own design work — sequences, risk frameworks, and trackers we built ourselves and can explain end to end.
If we cannot support a piece from one of those two, we do not publish it. We do not fill a gap with a confident guess.
How we design a documented test
Some questions cannot be answered from documentation. For those we run a test, and the design comes first.
Before anything is run, we write down what we are trying to find out, what would count as a meaningful result, and what would not. We fix the scenarios and the criteria in advance and freeze them. We record the configuration in writing — what was used, which version, which settings — because a result nobody can reproduce is not a result.
During the run we measure rather than estimate, and we write values down as they happen. A number reconstructed afterward from memory is an estimate, and we label it as one.
Afterward we keep the records: the frozen inputs, the measurements, the failures, and the cases where the process stopped. Those records are what a conclusion has to trace back to.
Synthetic scenarios
Much of our testing runs on invented material — fictional reviews, fictional customers, fictional businesses — written to exercise a process under conditions we control.
We do this deliberately. Our synthetic testing uses no real customer reviews, no personal data, no business account data, and no real transactions. Nobody’s actual customers are involved, and no real business is exposed to an experiment.
Working this way is also more useful for you. Because the inputs are invented, we can publish them alongside the piece, and you can run the same inputs yourself.
It has a limit we state plainly: a synthetic scenario cannot establish real-world performance by itself. Invented material has no account history, no prior relationship, and none of the context that makes a real situation complicated. It tells you how a process behaves under conditions we chose, which is useful and is not the same thing.
What counts as evidence
We hold reported results to a higher bar than guidance.
A piece is not a completed test until its methodology, records, limitations, and conclusions have been officially reviewed. Until that review happens, the work is described as planned or in progress, never as done, and the piece does not publish in a reduced form with softer wording.
We also run informal pilots — rough passes used to shape a protocol before it is finalized. Internal pilots are not official evidence. They are not reported as results, they are not used to fill a gap in a documented test, and no number from a pilot is presented as a finding.
Where a completed test has material limitations or unresolved gaps, those stay visible in the published piece rather than being softened or pushed to the end.
Where AI helps, and where a person decides
We use AI, and we would rather tell you where than let you guess.
It helps us organize research, build outlines, prepare drafts, review language, and write the synthetic scenarios described above. That assistance is real and it saves time.
What it does not do matters more. AI output is not accepted as fact: anything a model produces is checked against a source before it reaches a published page, and material we cannot verify is removed rather than softened. AI does not choose what we publish, and it does not approve anything.
We also do not automatically publish replies or content to third-party platforms. When we describe an AI-assisted workflow, it produces a draft for a person to read, edit, and post themselves. Nothing posts on its own.
Human review before publication
Every page is read in full by a person before it publishes, and that person takes responsibility for it. This step is not delegated to a model, and it is not skipped because something is running late.
Publication is manual and explicitly authorized, one page at a time. The Tool Trial is independently operated, and we do not describe ourselves as larger than we are.
What we publish, and what we hold back
We hold back more than we publish. A piece waits when its sources cannot be verified, when its records are incomplete, when the conclusion outruns what the evidence supports, or when what we found is too narrow to be useful.
Waiting is the normal outcome, not a failure. A page that overstates what we know costs a reader more than a page that never appeared.
The limits of this method
Our tests are small. They usually cover one configuration, on one day, using scenarios we invented. That is enough to show how a process behaves under specific conditions, and not enough to generalize to every business, every tool, or every situation.
We do not draw legal, regulatory, medical, or compliance conclusions from our work. Where a topic touches those areas, we say so in the piece and recommend a qualified professional.
When we update
We revisit pages when the facts underneath them change — when official guidance is revised, when a product behaves differently, or when new work affects a conclusion we drew. Updated pages carry the date of the update.
If something here does not match what you see on the site, tell us through our contact page.
Sources
- Google Search Central — Creating helpful, reliable, people-first content
- Google Business Profile Help — Reply to reviews
Both pages were checked on 2026-08-14.
Citing these does not make The Tool Trial affiliated with, endorsed by, certified by, or partnered with Google or any other organization.
This page describes our working method. It is not legal advice, and it is not the official policy of any platform.