How we use AI in our own editorial process — and where we never do

We spend our time hands-on testing AI software, evaluating AI output quality, and writing about AI tools for a living — which puts us in an unusual position. A publication that reviews AI chatbots, generative AI writing tools, and AI automation platforms for small businesses has an obligation to be just as transparent about its own use of AI as it asks every vendor it covers to be about theirs. Vague reassurance isn’t good enough here; a reader deserves to know exactly which parts of our workflow AI touches and which parts it never comes near.

This matters for a practical reason too. AI-assisted content has become common enough across digital publishing that readers increasingly ask a direct question before trusting a review: did a person actually use this software, or did something generate a plausible-sounding review without ever touching the product? For AIBizMaster specifically, the answer needs to be unambiguous, because the entire premise of the publication — hands-on testing over marketing claims — falls apart if the testing itself isn’t real.

So this page exists to remove any ambiguity. It documents, section by section, exactly where AI tools assist our editorial workflow, where human judgment is the only thing that ever produces a verdict, and what happens when an AI-assisted claim turns out to be wrong. It’s meant to be checked against, not taken on faith.

Our underlying view is straightforward: AI is a tool for handling volume and structure, not a substitute for judgment, taste, or firsthand experience. A large language model can organize scattered research notes into a coherent outline far faster than a person can by hand. It cannot sit down with a real AI phone system, place a real test call, and notice that the tool mishandles a rescheduling request in a way its marketing copy never mentions. Those are fundamentally different kinds of work, and conflating them is where a lot of AI-assisted publishing goes wrong.

We treat this distinction as a design constraint on our own workflow, not a marketing talking point. Anywhere a task requires actually experiencing a product, forming an opinion about it, or making a judgment call that affects a reader’s decision, a human does that work directly. Anywhere a task is closer to organizing information that’s already been gathered — structuring a draft, tightening a paragraph, checking basic grammar — AI assistance is a reasonable, disclosed part of getting there efficiently.

This philosophy also shapes how we think about AI software itself, the subject of nearly everything we publish. We’re not reflexively skeptical of AI, nor are we boosters of it. We treat every AI tool the same way we treat our own use of AI: judged by what it actually does when tested, not by what a demo suggests it can do.

We don’t publish the specific prompts used to assist with drafting, for the same reason a newspaper doesn’t publish a reporter’s private notes — the meaningful thing for a reader to know is what was verified and by whom, not the exact intermediate steps that produced a first draft. What we do disclose, and will continue to disclose, is the boundary itself: which categories of work involve AI assistance and which never do, laid out plainly in the paired list above.

If a specific piece of AI-assisted phrasing is ever challenged as misleading, the resolution isn’t “the AI said it” — it’s the same accountability any staff-written sentence gets, handled through our Corrections Policy.

Beyond accuracy, we think there’s a separate ethical question worth answering directly: is it honest to use AI assistance at all in a publication whose value proposition is human, hands-on testing? Our answer is that using AI to handle organizational and structural work is honest as long as it’s disclosed and doesn’t touch the parts of the process that actually require a human — testing, judgment, and the final verdict. What would be dishonest is presenting AI-organized research as if it were independently discovered, or letting an AI-generated summary quietly stand in for testing that never happened.

We also think about the ethics of AI use in the other direction — how AI vendors themselves should be evaluated on responsible AI use, covered in our How We Test AI Software page’s security and privacy review. Holding ourselves to a lower standard than we hold the products we review would be its own kind of dishonesty.