"It's not X, it's Y": why AI loves this sentence (and what to write instead)
The setup-and-reveal sentence that shows up in every AI LinkedIn post, why models write it, and ten ways to say the same thing better.
"It's not X, it's Y" is a phrasing AI uses to sound insightful: it knocks down a straw man X before revealing Y. (Wikipedia calls this negative parallelism.) Pangram, which makes an AI detector, measured the related phrase "not just X but Y" at about three times the human rate in AI text. In our AI drafts, it showed up mostly in LinkedIn posts and academic abstracts. Eliminate the X, and be specific in stating the Y.
What is "it's not X, it's Y" called?
It's called a negative parallelism, according to Wikipedia's guide to signs of AI writing, which describes it as stereotypically an "AI sign." The guide groups three versions under that name:
- Not X, but Y: "It's not a tool, it's a teammate."
- Not just X, but also Y: "This isn't just a launch. It's a turning point."
- Y rather than X: "The goal is progress rather than perfection." Wikipedia notes this reversed form is particularly common in recent model output.
Wikipedia also lists a variation, "no X, no Y, just Z", as in "No meetings, no busywork, just results."
Why does AI write "it's not X, it's Y" so often?
Nobody has shown why this is, but here are some guesses:
It sounds insightful. Using this pattern creates a reveal, and it's a way for an AI to sound wise on a topic even if it doesn't know anything about that topic specifically.
Their training data probably contains a lot of copy that uses this phrasing, like ads, keynote speeches, self-help books, LinkedIn posts, etc. This kind of phrasing is "quote-y", which means it's probably been overused in the kind of writing AI models have been trained on.
It may have been rewarded by the tuning process. Chat models are usually tuned by having people rate the model's answers to questions. This phrasing probably reads nicely to raters, and so it likely got rewarded by the tuning process. (This part is a guess, not a measured result.)
Whatever the cause, it does seem to be a pretty common pattern in AI writing. Pangram found that "not just X but Y" type phrases appear in AI-generated text 3 times per 10,000 words, versus 1 time per 10,000 words in human writing.
Do AI models add "it's not X, it's Y" when they polish your writing?
No. In our rewriting study, 2026 models polished existing writing. These rewrites averaged 0.12 of these per thousand words, compared to 0.17 in the original writing.
That makes sense: when an AI polishes existing writing, it mostly keeps your framing of the topic instead of choosing its own.
When AI polishes writing, it often adds new words like "genuinely" and "plenty of". See our list of words and phrases that AI tends to overuse.
Does AI use "it's not X, it's Y" more when it writes from scratch?
Not overall. In our second study, we had four 2026 models write texts from scratch, and they used it on average exactly as often as people: 0.12 times per 1,000 words. But in some cases (LinkedIn posts, academic abstracts, letters, forum posts) the people we sampled didn't use it and AI did.
| Kind of writing | Human originals, per 1,000 words | AI drafts, per 1,000 words |
|---|---|---|
| LinkedIn posts | 0 | 0.51 |
| Academic abstracts | 0 | 0.32 |
| Letters | 0 | 0.18 |
| Forum posts | 0 | 0.13 |
| All 20 kinds of writing | 0.12 | 0.12 |
The model used also matters. GPT-5.4 mini uses it a lot (0.29 per 1,000 words, 2.4x more than people), but Claude Sonnet 5.5 uses it very little (0.04 per 1,000 words).
So combining the two studies, AI doesn't add them when rewriting, and it adds them when writing specific kinds of text from scratch. It also depends a lot on the model used. Of course, one such statement isn't enough to know if a text was written by AI, but if it's used several times in a LinkedIn post that isn't backed by a story, it's probably AI. We only tested about 35 originals of each kind, so don't fixate on the exact numbers here.
Is it ever fine to write "it's not X, it's Y"?
Yes, this is totally okay if you have reason to believe that some people really do believe X. Sometimes your customers really do think price is the problem, and you know it's onboarding. Then you can say "It's not the price, it's the onboarding." The trick is to make sure you're not making X up, a.k.a. using a straw man. If you find yourself about to say this, stop and ask yourself if you've ever heard anyone say X. If not, it's probably best to cut the line.
This is also a problem if you do it too much. Once is a choice. If you do it three times in a 200-word post, that's a habit, and people will notice.
How do I rewrite "it's not X, it's Y"?
One easy way to fix this is by removing the first part and adding some more detail to the second part. Here are 10 rewrites so you can see what it could look like. (The details are made up for the example, so make sure you replace them with real ones!)
Before: It's not a tool, it's a teammate.
After: It drafts the follow-up emails you'd otherwise forget to send.
Before: This isn't just a product launch. It's a movement.
After: We're launching today. Forty teams used the beta, and most asked to keep it.
Before: The problem isn't your pricing. It's your positioning.
After: Your price is fine. Buyers just can't tell what they're paying for.
Before: It's not about working harder, it's about working smarter.
After: Drop the two meetings nobody needs and you get Tuesday afternoon back.
Before: Leadership isn't about having all the answers. It's about asking the right questions.
After: My best manager asked one question every Monday: what's blocking you?
Before: We're not just a vendor, we're your partner.
After: If something breaks at 2 a.m., call us. A person will pick up.
Before: It's not the price that matters, it's the value.
After: It costs $40 a month and saves most teams about five hours a week.
Before: Content marketing isn't dead. It's evolving.
After: Blog posts still bring us leads. They just take longer to pay off than they used to.
Before: Hiring isn't a numbers game, it's a relationship game.
After: Six of our last eight hires came from referrals.
Before: No fluff. No jargon. Just results.
After: Here's what we changed last quarter and what it did to signups.
Take a second to look at what's changed here. Not only are most of these shorter, but you can see that by including more specific details (numbers, names, etc) and not trying to impress the reader with the reveal at the end, the writing is much better.
While it's easy to spot this pattern now that you know it exists, it's always good to do a final "eyeball" pass to make sure you don't have any instances of it in your writing. It's harder to manually fix other signs of AI writing like even sentence lengths, morals at the end and lines built to be quoted. If you want to learn how to spot and fix those, check out our guide to signs of AI writing in 2026, and for a full breakdown, see how to make AI writing sound human. If you write blog posts and outreach with an AI agent every day, Zavus can do this pass natively in the tool you already use. See how Zavus works for sales teams.
Frequently asked questions
What is the "it's not X, it's Y" pattern called?
Wikipedia's guide to signs of AI writing calls it a negative parallelism. The name covers "not X, but Y", "not just X, but also Y" and "Y rather than X".
Is it wrong to write "it's not X, it's Y"?
No. It's a normal way to write a sentence, and people have always written that way. It becomes a problem when it argues against a straw man or gets used too often.
Does AI add this pattern when it edits my writing?
Not in our test. Across ten 2026 models rewriting human-written blog posts, the rewrites had slightly fewer of these constructions than the originals (0.12 against 0.17 per 1,000 words).
Does AI use "it's not X, it's Y" more when it writes from scratch?
Not overall: in our 1,177 drafts it matched the human originals (0.12 per 1,000 words each). It showed up more in LinkedIn posts and academic abstracts, and from some models, such as GPT-5.4 mini.
Sources
- Wikipedia:Signs of AI writing, Wikipedia, accessed November 2026
- 9 Signs of AI Writing, Backed by Data, Pangram, 2026
- AI writing tells in 2026: data from 7,554 rewrites and 1,177 drafts, Zavus, November 2026