I run a desk of human writers for a living. Here is the test I use on any draft, machine or human, before it goes out under a client’s name.
The draft is fine, and that is what worries me.
I read a lot of marketing copy. Some of it comes from clients who wrote it themselves. Some comes from agencies. A growing share comes out of a tool that produced it in nine seconds. The tool drafts are the ones that worry me, and not for the reason most people give.
The usual complaint is that you can spot them. The rhythm is too even. Every paragraph has three points. Every post opens with a question and closes with a soft nudge. You can train a model to drop those habits and the next version will. I have watched that happen more than once. The tells move. The thing underneath them does not.
Nobody is on the hook for the sentence
When a person writes under their own name, a sentence costs them something. A client reads it. A competitor screenshots it. A customer replies with a fair objection and the writer has to answer or look like they were bluffing. That cost shapes every word choice, and readers hear it even when they cannot name it.
A model pays nothing. It never wakes up the next morning and has to stand behind a claim that landed badly. It samples what was common, then what raters preferred, and settles on the answer least likely to lose a vote. The result reads fluent and weightless at the same time. I went into the mechanics of that in an essay on why AI voice is unaccountable speech with better manners: https://kolbinah.com/en/item/people-don-t-distrust-ai-written-posts-they-distrust-posts-nobody-stands-behind–841c5b4484e2
So the question I ask about any draft is not whether a machine wrote it. It is who signs it. If the answer is a tool, the reader works that out in about two seconds, usually by scrolling to the comment section and counting the replies that never got answered.
What the cheap draft actually saves you
Here is the arithmetic that convinces owners to try it. A junior writer costs real money. A tool costs thirty dollars a month. The tool produces a month of posts in an afternoon. On paper the saving is obvious.
The saving is real and it is smaller than it looks. Writing a post was never the expensive part of running an account. The expensive part is the reply that has to read one specific question and answer it with a lead time or a price, in the brand’s voice, without inventing a fact. That work does not compress. A tool can generate forty posts while you sleep. It cannot sit in your comment section at four on a Monday afternoon and answer a buyer who is still deciding.
I have watched the same pattern repeat. A founder fills a calendar with generated posts. The feed looks alive. The comment section goes quiet because no one is watching it. Then a customer asks how long delivery takes and gets nothing back for nine days. The post did its job by the only measure anyone tracked, which was that it went out on time.
The number that predicts the outcome
Ask any provider, mine included, for one figure: the median time between a comment landing and a human answering it. Not the average. Averages hide the problem, because nine replies in ten minutes will flatter a tenth that sat for three days.
Then read ten replies at random and open them. Does each one answer what the person actually asked? Would you put your name under it? A reply that says “Great point, thanks for sharing” tells you everything about the desk behind it. So does a news-site comment that could be pasted under any article on any site. That test works on day one, before there are any results to argue about.
What good AI-assisted content looks like
I am not against the tools. My writers use them. The useful version is a draft that gets a person ninety percent of the way there, and then a human editor who has read the client’s archive finishes it. That editor knows which opinions the brand will not touch and which words it never uses. They cut the sentence that sounds impressive and says nothing. They add the specific fact that only this client would know.
The distinction is who carries the risk. A tool produces text. An editor produces text that someone is willing to defend in public. Those are different products and they cost different amounts.
I argued in an earlier piece that most agencies sell content creation when what a business actually needs is voice systematization, and that argument holds whether the writing is done by a person or a machine: https://kolbinah.com/en/item/starbucks-proves-strategy-beats-scale-every-time–4096f1c42b0c
The three questions I would ask before paying anyone
First, who answers the comment that comes in at four on a Monday, and can you get that person’s name? If the answer is the system, you have bought the thing you were already worried about.
Second, can they show you a post that contains something only you would say? Not a theme. A sentence with your fingerprint on it. If every sample could belong to any company in your category, it belongs to none of them.
Third, who owns the account if you leave? If the answer is anyone but you, you are renting your own audience.
A provider who passes all three is rare. One who fails the first is a scheduling tool with a logo.
What to do with the draft sitting in front of you
If a tool wrote it and a person read it and signed it, ship it. If a tool wrote it and nobody read it, you have not saved money. You have moved the cost to the reader, who will pay you back by scrolling past.
GuruVox is a human desk. Writers and editors, daily posts and replies, comments on news sites, Shorts and Reels cut from the client’s own material, and the Studio at my.guruvox.com so the work has somewhere to live. One flat fee. We do not touch ad spend. Thirty days, money back if it is wrong for you. If you want that desk running your voice, write to apply@guruvox.com or start at https://guruvox.com The people who notice are not reading for style. They are reading for risk. A buyer deciding whether to trust you with money is scanning your last ten posts for evidence that a person is on the other end, and the evidence they want is a reply with a fact in it, not a caption with a hook in it. That reply is what they are looking for.
I built GuruVox around that test because the market kept failing it. Owners were paying for output and getting output, then wondering why the phone stayed quiet. The desk writes the posts, answers the comments, places the comments on news sites, and cuts the Shorts from material the client already recorded. The client keeps the logins, the followers, and the credit. We never hold the account, because an agency that holds your account has already decided how the relationship ends.
The cost of getting this wrong is not the monthly fee. It is the eleven days a real question sat unanswered under a product photo while the buyer went to whoever replied first. That number does not show up in any report a tool generates, because the tool was never measuring the thing that lost you the sale.
If you want to see what the desk actually ships before you commit to anything, the log is the place to look, and I will walk you through a week of it on a call. Start at https://guruvox.com or write to apply@guruvox.com.