Last updated: 21 August 2026
This site is about AI tools, and it is made with AI tools. It would be strange — and a little dishonest — not to tell you exactly how.
The short version: AI agents run the tests, gather the logs and write the first drafts. A human being checks every number against those logs, edits the draft, and signs their name to it. Nothing publishes without that signature. The byline is the person who is accountable for the article, not the software that typed it.
1. Why we tell you this at all
Two reasons.
The first is that you are entitled to know. If you are about to act on a number in one of our articles, how that number was produced is part of the number.
The second is that Google asks publishers to make this self-evident. Its guidance on creating helpful content asks whether “the use of automation, including AI-generation, is self-evident to visitors through disclosures or in other ways”, and notes that such disclosures are useful for content where someone might reasonably wonder how it was created. A site full of measurement logs is exactly that kind of content. So here is the answer, in detail, rather than a line of small print.
2. What the AI actually does here
We run a set of AI agents as a working department. Their jobs:
Running the tests. The agents execute the actual sessions — invoking the tools, issuing the API calls, driving the browsers, building the pipelines. When an article says a task ran 412 times, agents ran it 412 times.
Collecting the evidence. Agents capture the raw terminal and API output without editing it, record the structured measurements (model and version, timestamp, input size, tokens in and out, elapsed time, attempts, cost), log every failure, take the screenshots, and note the environment.
Writing the first draft. An agent turns the evidence set into prose. Its inputs are the logs and the measurement record — not other websites. If a draft contains a fact that is not in the evidence set or in a cited source, that is a defect and it gets removed at review.
Routine editorial work. Structure, headings, tightening sentences, consistency of terminology, formatting tables.
Publishing mechanics. Scheduling, uploading, internal linking, and the technical parts of putting a page on the internet.
3. What the AI does not do
It does not decide what is true. The evidence decides, and a human checks the article against the evidence.
It does not invent numbers. Every figure in an article traces to a stored log line. If a number cannot be traced, it is deleted rather than softened.
It does not have a byline. No article on this site is attributed to an AI, and no AI persona is presented as a person. There are no fictional authors here.
It does not replace first-hand experience. An agent can run a benchmark; it cannot tell you that a tool felt wrong to use, or that a documentation page sent us down a dead end for forty minutes. Those observations come from the person, and they are the reason the person is in the loop.
It does not publish. The pipeline is capable of publishing autonomously. We have deliberately not wired it that way.
4. What the human does — and why the byline is theirs
Before any article goes out, the person named in the byline:
- Checks every number against the raw log. Not a summary of the log. The log.
- Verifies the first-hand claims. If the draft says a tool was used, the reviewer confirms it was actually used. If it says something was tested on a particular date and version, the reviewer confirms the record says so. Any claim that cannot be confirmed is cut, not hedged.
- Removes overstatement. Drafts tend to sound more certain than the evidence warrants. Bringing the language back down to what was actually measured is most of the editing work.
- Adds what the logs cannot hold — judgement, context, the recommendation, the caveats that come from having sat through the session.
- Signs off. The review date is recorded alongside the article.
This is why the byline is a person’s name. Google’s own guidance treats content that is “written or reviewed” by someone who knows the subject as equivalent — review is a real form of authorship. A person who has verified every claim against the evidence and who will answer for it is the correct author of record.
The reverse is what we refuse to do. Putting a human name on something no human read would be deceiving you about who is responsible for the page. That is not a grey area for us; it is the one line the workflow is built to make impossible. An unreviewed draft cannot pass the publishing gate.
5. Our publishing limit, and why it exists
We publish at most one article a day and five in a week.
We could publish far more. The pipeline could produce dozens of plausible articles a day, and search engines do not penalise volume as such: Google has repeatedly said that page count is not a ranking factor, and has rejected the idea that there is a limit on how many pages it will index from one site. Those are reported remarks from Google staff rather than a line in a policy document, so treat them as guidance rather than as a rule we can quote at anyone.
We cap it anyway, because the limit is not about search engines. We cannot honestly measure faster than that. One article requires one real working session plus a complete evidence set. Two articles a day would mean one of them came from somewhere other than a measurement — and an AI-written page that is not backed by a measurement is precisely the thing Google’s spam policy calls scaled content abuse: pages generated at scale without adding value.
So the rule below the cap matters more than the cap: no evidence set, no article. In a week where the sessions produce three well-evidenced pieces instead of five, we publish three. The quota is a ceiling, never a target.
6. What you will see on each article
Every article carries, at the foot:
Drafted with the help of AI tools, then fact-checked and edited by the named author before publication. Tools mentioned were tested first-hand.
That last sentence is only ever used when it is true. Where we have written about something we did not run ourselves — reading documentation, summarising someone else’s published benchmark — the article says so in its own words, in the body, where you will actually see it.
Each article also shows its publication date and its last-updated date, and its test setup section states the date the measurements were taken and the exact versions used. A benchmark without a date is not a benchmark.
7. Where the evidence lives
Behind each article there is a stored evidence set: the unedited logs, the structured measurements, the failure list, the screenshots, the environment. The article’s Test Setup section is written so that a reader with the same tools can reproduce the run and get their own numbers.
If you reproduce one of our tests and get a materially different result, we want to hear about it. Tell us what you ran and what you got. If you are right, we correct the article and say so — see the corrections policy on our About page.
8. This page changes when the process changes
If we change how articles are produced, this page is updated in the same change, not afterwards. A disclosure describing a process we have abandoned would be a false statement about who is responsible for our content, which is the exact failure this page exists to prevent.
If something here is unclear, or you think our practice does not match this description, write to us through the Contact page and say so directly.