Why “It’s Not Just X — It’s Y” Became the Dead Giveaway of AI Writing (and What to Do About It)

If you’ve been reading a lot of press releases, LinkedIn posts, or blog articles lately and keep stumbling on the phrase “It’s not just X — it’s Y,” your instincts aren’t wrong. That neat little turn of phrase has exploded across the web — and not because humans suddenly fell in love with parallel structure. According to a new TechCrunch report, it’s one of the clearest signals that a large language model (LLM) helped put words on the page.

Why does this one line keep showing up? How are detection tools using it? And, most importantly, what should editors, marketers, and founders do now that a once-useful flourish is practically a red flag?

Let’s unpack the trend, the why behind it, and a practical playbook to keep your content authentic and effective in an era where synthetic prose is everywhere.

Read the original TechCrunch report for context

The strange rise of a robotic turn of phrase

TechCrunch reports that the “It’s not just X — it’s Y” construction has become so pervasive in AI-generated text that it’s now “near-definitive evidence” of synthetic writing in formula-driven formats like press releases and corporate updates. Max Spero, CEO of AI detection startup Pangram, says 2025-era frontier models show a high base rate of this “tic,” and it gets amplified further when tools generate templated, emotion-light copy at scale.

That lines up with what many editors and readers have noticed since 2024. Once this structure sneaks into your mental autocomplete, you start seeing it everywhere:

It’s not just growth — it’s sustainable growth.
It’s not just a feature — it’s a paradigm shift.
It’s not just innovation — it’s customer-centric innovation.

Add a handful of hedges (“in many ways,” “to be sure,” “arguably”), a numbered list, and a couple of vague claims, and you’ve got what AI detectors now recognize as an ultra-common AI signature — especially in marketing, PR, and investor communications.

Where the tic comes from: data bias and templates

Why this phrase? The simplest answer: training data and incentives.

Overrepresented in training data. The modern web is full of punchy, parallel constructions that promise contrast and payoff. LLMs trained on vast swaths of internet text learn that “It’s not just X — it’s Y” often scores engagement and clarity, so they reuse it frequently.
Rewarded by prompts and templates. When you prompt a model to “make it more compelling” or “add emphasis,” it reaches for familiar, high-confidence patterns. Many content tools, style guides, and corporate templates leaned into these patterns during the 2023–2025 adoption boom.
Safe and generalizable. The phrase fits nearly any topic and creates the illusion of depth without committing to details. In high-volume content operations, “safe and generalizable” often beats “risky and specific.”

It’s the linguistic equivalent of stock photography: eye-catching, easy to insert, and everywhere.

How detectors leverage it (and why they don’t stop there)

No credible detector flags a single phrase in isolation. But when “It’s not just X — it’s Y” appears alongside other machine-favored patterns, the signal strengthens. According to TechCrunch, tools like Pangram analyze n-gram frequencies — essentially counting and comparing recurring word sequences — and then combine those stats with additional markers:

Over-hedging and equivocation (“may,” “could,” “on the other hand”)
Repetitive cadence across paragraphs (parallel openings, mirrored clauses)
Listicle-heavy structures lacking concrete specifics
Generic modifiers (“robust,” “seamless,” “transformative”) packed around abstract nouns

Pangram reportedly flags outputs from GPT‑5 derivatives and Claude 3.5 with high accuracy (around 95%) when this phrase clusters with companion signals. That doesn’t mean the phrase is definitive proof, but it is a strong component in a broader probabilistic model that scores “AI-likeness.”

For editors and legal teams, the takeaway is nuance: detection is a risk lens, not a gavel. You want layers of evidence and a path to human verification, not a single-phrase ban list.

Why editors care: trust, brand risk, and SEO

The fallout is tangible. TechCrunch notes editors at outlets like Barron’s and Forbes now reject submissions laced with the line. Why such a strong reaction?

Trust signals matter. Readers are on high alert for bland, automated prose. Overused AI tics can erode credibility even if the piece is accurate.
Compliance is tightening. The EU AI Act introduces transparency rules for certain generative uses, and U.S. policymakers are moving on synthetic media and deepfakes. Publishing content that “reads AI” without disclosure creates reputational and regulatory risk.
SEO prefers originality. Search systems increasingly reward helpfulness, specificity, and uniqueness. Recycled phrasing correlates with lower information gain and higher similarity scores — not a place you want flagship content to live.

In short: if you run a newsroom, brand, or investor-facing site, this pattern isn’t just a style quirk. It’s a risk indicator for authenticity, compliance, and performance.

The limits and pitfalls of AI detection

Let’s keep our heads: humans have been using contrastive structures for centuries. A skilled writer might deploy “It’s not just X — it’s Y” for rhetorical effect, with rich specifics before and after. So where does detection stumble?

False positives. Human writers in PR-heavy verticals often mirror the same patterns, especially under tight deadlines. A detector that overweights one phrase can mistake hurried humans for machines.
False negatives. As more users prompt models to avoid obvious tics (adversarial prompting), surface markers get scrubbed even while the underlying “AI cadence” persists.
Domain skew. Legal, medical, and financial writing favor hedges and parallelism for safety and clarity. Raw frequency alone isn’t proof of machine authorship.

The smart stance: treat detection as triage. Use it to prioritize reviews and ask for sourcing, drafts, or interview notes — not to summarily reject work without context.

The bigger context: regulation and platform responses

The AI writing boom didn’t happen in a vacuum. A few macro moves are reshaping incentives:

EU AI Act. The European Union’s AI Act introduces transparency obligations for certain high-risk and generative AI use cases, pushing organizations toward disclosure and provenance practices. See the evolving text at the EUR-Lex portal.
U.S. policy pressure. While the U.S. lacks a single sweeping AI statute, there’s growing movement on synthetic media and deepfakes, plus executive-branch actions framing responsible AI. For a primer, review the White House Executive Order on AI (Oct 2023).
Platform tooling. Detection firms are partnering with publishers to pre-screen submissions. Meanwhile, labs at companies like Google DeepMind and Microsoft explore “style randomization” and controllable generation to diversify outputs. See the DeepMind blog and Microsoft’s open-source Guidance project for steerable generation patterns.

Don’t expect a silver bullet. Watermarking and provenance tech are promising but imperfect, especially against paraphrasers. For background, see “A Watermark for Large Language Models” on arXiv.

What to do now: a practical playbook for creators and editors

You don’t have to ban AI or phrases to keep content credible. You need process, style rigor, and provenance. Here’s a field-tested playbook.

1) Build an authenticity-first workflow

Capture sourcing artifacts. Save interview recordings, emails, datasets, and field notes. Link them in internal briefs. Provenance beats vibes.
Require outlines and drafts. Ask contributors for outline timestamps and draft evolution. Real work leaves a trail.
Attribute AI assistance. If writers used AI for ideation or grammar, say so briefly in a footnote or disclosure line where policy requires.

2) Tune your brand voice away from templated rhetoric

Maintain a live “anti-tic” list. Include overused frames like “It’s not just X — it’s Y,” “In today’s fast-paced world,” and “At the end of the day.” Update quarterly.
Teach alternatives (examples below). Don’t just say “avoid X”; offer specific sentence patterns that achieve the same rhetorical effect without the cliché.
Emphasize concrete nouns and verifiable claims. Reward specificity in edits: numbers, names, places, timelines, quotes.

3) Institute a style lint before publish

Run a stylistic pass. Search for common AI telltales: repeated contrast frames, hedges stacking two or more per sentence, adjective clusters before abstract nouns.
Check for information gain. Ask: what’s new here? A quick desk check against your own archives and top-ranking SERPs will reveal redundancy.
Pull a quick detector score — then verify. Tools are inputs, not verdicts. If a score is high, escalate to human review and ask for sources and drafts.

For a baseline on detection thinking, see Turnitin’s overview of AI writing detection.

4) Raise the cost of fakery with “receipts”

Embed primary material. Screenshots, redacted docs, code snippets, charts with source links, or 20-second audio clips from interviews.
Add byline credibility. Short bios with relevant expertise and LinkedIn/Twitter links signal a human behind the words.
Include datestamped updates. A small changelog at the bottom grows trust over time and makes templated drive-bys less likely.

5) Calibrate prompts and tools

If you do use AI:

Steer for specificity. Prompt for concrete, contextual details and citations. Ask the model to propose interview questions and data sources rather than deliver polished prose.
Randomize structure. Request varied sentence lengths, unconventional openers, and rhetorical devices (e.g., anaphora, metonymy) — then edit fiercely.
Keep humans in the loop. Treat AI as a sparring partner for research and brainstorming, not a ghostwriter.

6) Train editors to spot cadence, not just keywords

Read aloud. AI cadence reveals itself in rhythmic sameness and “finish your sentence” predictability.
Check paragraph openings. If three or more start with the same structure (“In many ways…,” “Not only…,” “It’s not just…”), slow down and scrutinize.

Strong style alternatives to “It’s not just X — it’s Y”

You don’t need a ban list if you have better options. Here are crisp alternatives that create contrast without the cliché. Tailor them to your voice:

The real shift isn’t X. It’s Y — and that changes Z.
X got us to the starting line. Y gets us over the finish.
Most teams chase X. The leaders optimize for Y.
On paper, it’s X. In the field, it’s Y.
We used to believe X. The data points to Y.
If X explains the what, Y explains the why.
X is the headline. Y is the footnote that moves markets.
X solves today’s pain. Y builds tomorrow’s moat.

Pro tip: Follow any contrast with specifics. Add a figure, timeframe, named customer, or source link. Contrast without context is just polish.

How this impacts different teams

For marketers and PR pros

Refresh your templates. Rewrite intros and transitions that rely on AI-favored scaffolding. Audit your press release boilerplates.
Specify claims. Replace “transformative platform” with the measurable outcome you drove for a named customer.
Vet vendor outputs. Tools like Jasper and Copy.ai are powerful, but their defaults skew generic. Fine-tune prompts to your brand corpus and edit in-house.

For newsroom editors

Tighten contributor guidelines. Require interview notes and sources, plus light disclosure if AI assisted drafting.
Kill listicles that don’t add reporting. Aim for fewer, denser pieces with original quotes, documents, or on-the-ground details.
Implement a two-pass review. First for facts and sourcing; second for cadence, clichés, and redundant structure.

For academics and students

Document your process. Keep a research log and annotate drafts. If institutional policy allows AI assistance for brainstorming, cite the tool and version used.
Avoid generic scaffolds. Professors can spot templated rhetoric even when content is correct.

For product and legal teams

Draft a disclosure policy. Decide when and how you’ll acknowledge AI assistance to meet legal and ethical expectations.
Pilot provenance tools. Explore document fingerprinting, cryptographic signing, or content credentials for high-stakes outputs.

What’s next: will the tic disappear?

TechCrunch notes detectors got much better after 2024, just as AI writing flooded content pipelines. Meanwhile, labs at DeepMind and Microsoft are experimenting with style randomization and controllable generation to diversify outputs. If next-gen models train on debiased data and incorporate stronger guardrails, the obvious tics may fade.

But two realities will stick around:

Adversarial prompting cuts both ways. Users will always find ways to mask or mimic style. Detectors and writers will keep leapfrogging.
Authenticity will keep winning. Search algorithms, regulators, and readers all push in the same direction: more receipts, less varnish.

In other words, the phrase may retire, but the underlying arms race is just getting started.

A quick editor’s checklist to keep content human and high-performing

Does this piece include at least two concrete specifics (numbers, names, dates, places) per 300 words?
Are there verbatim quotes, firsthand observations, or unique data points?
Is every claim either linked to a source or clearly labeled as analysis or opinion?
Do the first two paragraphs avoid stock framing like “In today’s fast-paced world,” “Not only…,” “It’s not just X — it’s Y,” or “At the end of the day”?
If AI assisted, is the contribution disclosed where appropriate and has a human editor substantively revised the output?
Would a rival convincingly claim they could have written the same piece without changing a word? If yes, you need more originality.

Frequently asked questions

Q: Is “It’s not just X — it’s Y” definitive proof of AI? A: No. It’s a strong signal in context, especially when it clusters with other markers like over-hedging and listicle-heavy structure. Treat it as a prompt for deeper review, not a verdict.

Q: How do AI detectors actually work? A: Most blend statistical analysis (n-gram frequencies, perplexity, stylometric features) with heuristics and sometimes model-based classifiers. The best tools combine multiple weak signals into a stronger overall score and are calibrated by domain.

Q: Can I just prompt a model to avoid this phrase and beat detectors? A: You can reduce obvious flags, but detectors look at broader cadence and repetition patterns. More importantly, merely hiding tics doesn’t create originality or trust. You still need specifics, sources, and human editing.

Q: Are watermarks the solution? A: Watermarking and provenance tech help, but they’re not foolproof, especially against paraphrasers or screenshots. Think of them as part of a layered approach that includes disclosure and editorial process. For background, see the LLM watermarking paper.

Q: Does this apply to non-English content? A: Yes. Each language has its own overused AI constructions. Detectors are expanding multilingual support, but accuracy varies by language, domain, and model training mix.

Q: How should brands disclose AI use? A: Follow applicable law and platform rules. A lightweight note such as “Edited with the assistance of AI tools; reviewed by [editor name]” can be enough for low-stakes content. For high-stakes material, include more detail about sources and human oversight.

Q: Will using one or two AI-favored phrases tank my SEO? A: Not by itself. Search systems don’t penalize phrases; they reward helpfulness and uniqueness. The issue is correlation: content that leans on generic scaffolds often lacks originality. Focus on adding real information gain.

Q: What’s the single best way to avoid sounding like a bot? A: Put something on the page a bot can’t fake: firsthand reporting, proprietary data, named customers, or lived experience — and edit for specificity and rhythm.

The clear takeaway

The “It’s not just X — it’s Y” boom isn’t just a style fad; it’s a symptom of how we’ve been using AI: fast, templated, and generic. Detectors noticed. Editors reacted. Regulators are circling.

You don’t need to fear AI or ban phrases to keep trust. You need a system. Anchor your content in specifics and sources. Audit your style for lazy scaffolds. Disclose assistance where it matters. And put a practiced human editor between draft and publish.

Do that, and you’ll sidestep the tics, outlast the detection arms race, and ship content that reads unmistakably like you — not just today, but as the landscape keeps shifting.

Further reading and sources: – TechCrunch on AI’s “It’s not just X — it’s Y” tic: https://techcrunch.com/2026/04/20/ai-writing-its-not-just-this-its-that-barrons/ – EU AI Act overview and text: https://eur-lex.europa.eu/ – White House Executive Order on AI: https://www.whitehouse.gov/briefing-room/presidential-actions/2023/10/30/executive-order-on-the-safe-secure-and-trustworthy-development-and-use-of-artificial-intelligence/ – DeepMind blog on research updates: https://deepmind.google/discover/blog/ – Microsoft Guidance for steerable generation: https://github.com/microsoft/guidance – Turnitin on AI detection: https://www.turnitin.com/blog/ai-writing-detection

Discover more at InnoVirtuoso.com

I would love some feedback on my writing so if you have any, please don’t hesitate to leave a comment around here or in any platforms that is convenient for you.

For more on tech and other topics, explore InnoVirtuoso.com anytime. Subscribe to my newsletter and join our growing community—we’ll create something magical together. I promise, it’ll never be boring!

Stay updated with the latest news—subscribe to our newsletter today!

Thank you all—wishing you an amazing day ahead!