AI in Marketing

Before you publish that AI post, ask it where it buried your point

A new writing benchmark says AI text hides the main point and invents strange phrases. What to ask AI before you publish, and how to fix only what's broken.

01 - The feeling

Why does AI-written text feel so tiring to read?

You asked AI for a post. It looks fine. Every word is spelled right. And still, by the third line, you're tired.

You read the paragraph twice. You still can't say what it wants from you.

That's not you being picky. A new writing benchmark says it's real, and it names what causes it.

Most people react the same way: they paste the text back and type "make it better". The AI rewrites everything. The new version is smoother, and just as tiring.

I do the opposite. Before anything gets rewritten, I ask the AI to find the problems. Then I fix only those lines.

What most people do

"Make it better." The AI rewrites the whole post, adds new phrases, and the main point is still hidden somewhere in the middle.

What you'll do instead

Ask the AI to point at the lines where the point is buried and the words sound strange. Fix only those. Everything else stays yours.

02 - The study

What did the writing benchmark find?

The study comes from Zonder Labs. One of the authors, Zhizhuo Zhou, shared it in a post on X on 9 November, with a longer write-up on their web page. Here is what they found, by their measure.

They checked writing against principles of clear writing from a classic textbook, Style: The Basics of Clarity and Grace by Joseph M. Williams and Joseph Bizup. They compared AI articles with human articles, and Claude Code replies with human answers on Stack Overflow.

In the authors' words: "AI writing breaks these principles far more often than human writing, with Claude Code replies by far the worst offenders."

Their short version of why AI text is hard to read: it "often obscures main ideas and invents awkward phrases, making sentences more complex than necessary."

The part that surprised me

Everyone hunts for long dashes and the word "delve". The authors say those AI signatures "are not what makes writing sloppy."

So you can remove every long dash and still publish a text nobody can follow.

One honest note. Their write-up focuses on AI writing at work, like work chats and coding replies, plus articles. They didn't test social media posts. But a post has the same job as a work message: make one point clear to a busy reader. That's why the check below works for any text you're about to publish.

03 - The faults

What are the three faults that make AI text hard to read?

The authors sort unclear AI writing into three faults. Here they are in plain words, with their own examples.

1. The "who" and the "what they did" are hidden. They call it hidden characters and actions. A clear sentence puts the person first and the action right after. AI often turns the action into a noun, so you have to dig for who did what.

Hidden

"A request for grapes was made upon the duck's arrival at the lemonade stand."

Clear

"A duck walked up to a lemonade stand and asked, 'Got any grapes?'"

2. Strange word combos. They call it vague wording. AI glues nouns together into phrases no person would say out loud. By their measure, this was the most common fault in Claude Code. Their example: calling a lemonade stand a "beverage retailer". Some they found in real AI writing: "a label's awareness entry" and "length floor drops a block".

3. Empty lines and a buried point. They call it empty statements and misplaced emphasis. You understand every word and still can't find the point. Or one sentence carries two points that fight each other.

Buried

"Nothing else depends on anything but the result, so I'll report as soon as the poll returns."

Found

"I'll report the result when the poll finishes."

Now look at your last AI post. My guess: you'll find at least one of the three in the first paragraph.

04 - The check

What should you ask AI before you publish?

Here is the whole idea: the first ask is not "rewrite". The first ask is "show me".

Paste your text into a new chat and add this:

Copy this: the find-first check
Don't rewrite anything yet. Read my text and only point out problems.

1. My main point: tell me in one sentence what you think it is,
and in which line you first found it.
2. Hidden who and action: quote every sentence where the person
and what they do are hidden in nouns ("a request was made"
instead of "she asked").
3. Strange word combos: quote every phrase no person would
say out loud.
4. Empty lines: quote every sentence I could delete without
losing anything.

Number each problem. Don't fix them. Just list them.

My text:
[paste your post here]

Read the answer to question 1 first. If the AI names a different main point than the one you meant, that's your biggest problem. Your point is buried.

Before you paste anything, write one sentence: what should a reader remember? Keep it next to you. This is what you'll compare against.

A fresh chat reads your text like a stranger. That's what you want: your reader is a stranger too.

Use the prompt above, word for word. Don't add "and make it better" at the end.

Does the AI's one sentence match yours? If not, move your point to the first line, in plain words.

Highlight each quoted line in your draft. Now you have a short list, not a whole text to redo.

Newsletter

Get my before-you-publish check in the newsletter. It goes out in my free newsletter Anjela about AI: sign up here →

You get one email with my free guide Teach Claude your style, then one workflow every Thursday.

05 - The fix

How do you fix it without rewriting everything?

Now you have a numbered list. This is where most people go back to "make it better". Don't.

Fix the lines yourself, if you can. It takes a few minutes, and the text stays yours.

If you want the AI's help, give it one line at a time, and tell it what must not change:

Copy this: fix only one line
Fix only problem number [2].
Say it the way a person would say it out loud.
Put who did it first and what they did right after.
Keep every other sentence of my text word for word.
Show me only the new line.

"Show me only the new line" matters. If the AI returns the whole text, it will quietly change things you liked.

Rewrite everything

You lose your own phrases. New strange word combos can sneak in. You have to read the whole text again from zero.

Fix only the list

You change three or four lines. You know exactly what changed. Your voice stays where it was.

Then run the check one more time on the fixed text. If the AI finds your main point in the first line, you're ready.

06 - Next step

What do you do today?

AI text tires readers because the point hides, the words get strange, and the lines say nothing. By the authors' measure, that happens much more often in AI writing than in human writing.

You don't need a better rewrite. You need to see the problem first.

  • I wrote my main point in one sentence myself
  • I asked the AI to find problems, not to rewrite
  • The AI found the same main point I meant
  • My main point is in the first line
  • I fixed every sentence where the who and the action were hidden
  • I replaced every phrase nobody would say out loud
  • I deleted the lines that said nothing
  • I fixed only those lines and kept the rest in my words

Your next step: take the next post you were going to publish today. Before you touch anything, paste it with the find-first check and read what it says your main point is.

Free workshops

Want to see how I use AI to build a business model, viral content and a selling event? Join my free workshops on 18, 19 and 20 Nov, 6:00 PM Dubai time. One free sign-up covers all evenings: save your free seat here →

FAQ

Why is AI-written text hard to read?

By the benchmark authors' measure, AI writing often hides who did what, glues words into strange phrases and buries the main point, which makes sentences more complex than they need to be.

Which AI did worst in the benchmark?

The authors say Claude Code replies were by far the worst offenders when compared with human answers on Stack Overflow.

Is removing long dashes enough to fix AI text?

No. The authors say signs like long dashes and the word "delve" are not what makes writing sloppy. The real problems are a hidden point, strange phrases and empty lines.

Why not just ask AI to make my text better?

A full rewrite changes lines you liked and can add new strange phrases. Ask AI to list the problem lines first, then fix only those and keep everything else word for word.

Channel

Breakdowns and notes - no fluff

New material from Anjela on AI, expertise and marketing. Subscribe to the channel.

Subscribe on WhatsApp