Free: the GTM BlueprintThe stack, by funding stageThe CRM data model, written downSeven steps, six handoffsRoles, and when to hire themA 90-day plan you can run on MondayThe three numbers that decide itNo PDF, no drip sequenceGet the blueprintFree: the GTM BlueprintThe stack, by funding stageThe CRM data model, written downSeven steps, six handoffsRoles, and when to hire themA 90-day plan you can run on MondayThe three numbers that decide itNo PDF, no drip sequenceGet the blueprint
RevOpsXL
GTM Diagnostic Book the audit

Fluency got cheap. It shows.

AI made everyone sound polished. So polished stopped meaning anything, and the vocabulary built to make you sound expert now tells a buyer the opposite: nobody here read it twice.

In short

Sounding polished used to prove you did the work. AI made fluency free, so it proves nothing now, and the tells (revolutionize, delve, a claim with no number) signal the opposite. Buyers judge you by your words. The fix is ten minutes: read it, cut the machine phrase, put back the specific.

On this page

You learn to spot it. A homepage that promises to revolutionize how teams work. A founder’s post about maximizing impact in a fast-moving landscape. A three-person consultancy writing in the exact voice of a company with forty thousand.

I read those words and I trust the company less. Not more.

Here is the uncomfortable part for anyone who sells expertise. The words meant to make you sound like an expert now do the opposite. They are the house style of a machine, and by this point most people can hear it.

The same claim, written two waysTHE MACHINE'S DEFAULTWe help companiesrevolutionize workflowswith robust, seamlesssolutions.Vague. Could be anyone.WHAT YOU'D ACTUALLY SAYWe cut a 12-steponboarding down to three.Two finance clients,same quarter.Specific. Only you.
The tools produce the panel on the left by default. The panel on the right takes ten minutes and a red pen.

The tell is measurable now

This is not a matter of taste. It shows up in the data.

In 2024 a group of researchers read through more than 15 million medical abstracts on PubMed and counted words. After ChatGPT arrived, a specific set of them climbed in a way nothing in the record ever had, not even the pandemic. Delve. Intricate. Showcasing. Commendable. Meticulous. Realm. Words almost nobody was reaching for at that rate in 2021, suddenly everywhere. The authors estimate at least one abstract in eight in 2024 had been through a model, and in some corners closer to two in five. (Kobak et al., later published in Science Advances.)

These are scientists, writing up real work they did, and the fingerprint is still that loud. Now picture a marketing team the Friday before a launch.

The open web is further along. Ahrefs ran a detector over 900,000 fresh pages in April 2025 and found about 74% of them carried AI-generated writing. Most was not pure machine output. It was the mix: a person prompting, pasting, and shipping. Which is the exact shape of a tell.

Vague is the other tell

The vocabulary is the easy one to catch. The harder one is the sentence that says nothing.

‘We help companies achieve their goals.’ ‘Solutions tailored to your needs.’ True of everyone, so about no one. A model reaches for this by default, because a claim with a real number or a named situation is a claim it might get wrong. Specificity is a risk. Machines avoid risk. So, it turns out, does anyone hiding the fact that they have nothing specific to say.

A niche expert who writes in general statements has argued themselves out of the one thing they were selling.

The same account does all of it

I have watched how senior people actually use these tools, up close. It is not the version in the demo.

One chat window holds a reply to a board email, a draft of the new landing page, a question about a knee that has been hurting, and a bit of holiday planning, all in the same half hour. The tool has no idea which of these is the company’s public face. It gives the landing page the same even, agreeable treatment it gave the knee.

So the thing that reaches the customer carries the same low, flat effort as the errand next to it. The executive feels productive. Fourteen things, handled. The output is fluent and forgettable, and it goes out with the company’s name on it.

Fluency used to be a proxy for effort

For a long time, sounding polished was a fair sign that someone had done the work. Clean writing was expensive. It needed a person who understood the subject and could shape a sentence, and you could not fake much of it at scale.

That proxy is gone. Fluency is free now. So a page that is merely smooth proves nothing, and a page built out of the tells proves something specific: nobody here cared enough to read it twice before it went live.

Buyers are already reading you this way. In Edelman and LinkedIn’s work on B2B decision-makers, 73% said a company’s thought-leadership content is a more trustworthy way to judge its ability than its marketing and product material. They read your writing to decide whether you are any good. When the writing is the machine’s and not yours, that is the answer they leave with.

Everyone ends up at the same voice

There is a second cost, and it lands hardest on the people who sell on being different.

In a controlled study, Doshi and Hauser had people write short stories with and without a model’s help. The assisted stories scored as more creative on their own. Lined up next to each other, they were measurably more alike. Better one at a time; the same in aggregate.

Read that twice if your whole pitch is that you see what others miss. The tool that makes today’s post look sharper is, across everything you publish, sanding you toward the middle. And the middle is the one place a specialist cannot afford to be seen. If you cannot say the specific thing, you do not have a position. You have a template.

The cheap thing is now the rare thing

None of this is an argument against the tools. I use them every day. They are good at the work nobody should be doing by hand: the first pass, the boring reformat, the summary of a forty-message thread. I have written about where that line sits, in augment your people, don’t replace them.

It is an argument against shipping the first draft as your voice. And the fix is not expensive. Read it before it goes out. Cut the word you would never say to a client’s face. Put back the specific number, the real example, the opinion a machine will not take the risk of holding. Say the one thing only you would have known to say.

That used to be the baseline. It is the whole difference now, because almost nobody bothers.

The effort it takes is small. Ten minutes and a red pen. When a brand will not spend even that on the words it puts its name to, it is telling you something true about how it will treat your account.

It just did not mean to say it out loud.

Common questions

How can you tell if content was written by AI?

Two tells. The vocabulary, words like delve, intricate, commendable and revolutionize that climbed across the web after 2022, and the vague claim that could describe any company. A study of 15 million abstracts found the vocabulary shift is measurable, and it is even louder in marketing copy than in science.

Does using AI for content hurt your brand?

Using it is fine. Shipping its first draft as your voice is the problem. Research on AI-assisted writing shows the output is more polished one piece at a time but more similar across everyone, so it quietly erases the difference a specialist is selling.

Why does so much AI content sound the same?

Because a model writes toward the average of everything it has read. Doshi and Hauser found writers using AI produced work rated more creative on its own but measurably more alike as a group. Left unedited, everyone converges on one voice.

Do B2B buyers actually notice low-effort content?

Yes, and they act on it. In Edelman and LinkedIn research, 73% of decision-makers said a company’s thought-leadership content is a more trustworthy way to judge its ability than its marketing materials. Weak content is read as weak capability.

How do you use AI for content without sounding generic?

Treat the model’s output as a first draft, never the final one. Read it out loud, cut any phrase you would not say to a client, and put back the specific number, the real example and the opinion a model will not risk holding. Ten minutes of editing is the whole difference.

We built our own practice on the opposite of this: models on the routine work, a person on anything that carries our name. Here is where we draw the line.

Read our AI manifesto Run the GTM diagnostic

Get the next one

New field notes twice a month. We don’t do newsletters, follow RevOps XL on LinkedIn instead.

Follow on LinkedIn

Sound familiar?

If this is happening in your stack, tell me about it. A senior expert reads these, not a bot - and you’ll get a real answer, whether or not you ever hire us.

← More field notes