Updated June 2026
AI-written text has gotten very good — but certain patterns still show up often enough that people have started to notice them. None of these prove a text is AI-written (humans do these things too!), but they’re fun, useful patterns to recognize.
⚡ Quick overview
- These are patterns, not proof — humans write this way too, especially in formal contexts.
- The most reliable “tell” is usually a mix of several patterns together, not just one.
- Good editing (human or AI-assisted) removes most of these naturally.
The 7 tellsThe big caveatWhat is happeningTry it safelyHow to verifyReview and maintainSourcesFAQ
7 patterns people associate with AI writing
- Overly balanced “on one hand / on the other hand” framing for topics that don’t really need it.
- Tricolon overload — lists of exactly three things, repeatedly: “fast, efficient, and reliable,” over and over.
- Grandiose openers — “In today’s rapidly evolving world…” or “In the ever-changing landscape of…”
- Heavy use of certain transition words — “Moreover,” “Furthermore,” “Additionally” appearing very frequently.
- Perfectly even structure — every section exactly the same length, every list exactly the same number of items.
- Vague confident summaries — “Ultimately, it comes down to balance and careful consideration” without saying anything specific.
- Em dashes used constantly — for asides, lists, and emphasis, far more than typical human writing.
| Pattern | Also common in human writing when… |
|---|---|
| “On one hand / other hand” | Academic essays, balanced op-eds |
| Lists of three | Marketing copy, speeches (it’s a classic rhetorical device!) |
| “In today’s world…” | Student essays, generic intros written under deadline |
| Heavy “moreover/furthermore” | Formal/academic writing styles |
Why this matters beyond a fun game
AI-detection tools exist but are known to be unreliable, and falsely flagging human writing (especially from non-native English speakers, who may use more formal/textbook phrasing) is a real problem. Treat “this might be AI-written” as a curiosity, not an accusation — especially in contexts like grading or hiring where false positives have real consequences.
Why AI-writing detection by ‘vibe’ is unreliable
Short sentences, polished transitions, repeated structures, neutral tone, and certain phrases can occur in both human and AI-assisted writing. Editing tools also blur the boundary: a human draft may be heavily rewritten, while an AI draft may be thoroughly reported and edited by a person.
Detection is especially difficult for short passages, non-native writers, formulaic professional formats, and text produced with accessibility or grammar tools. A list of stylistic tells can support discussion, but it cannot prove authorship.
Turn the idea into a safe experiment
Collect human, AI, and collaboratively edited samples on the same topic. Remove author labels and ask reviewers to explain their judgments, then compare confidence with accuracy.
- Save the exact prompt, model, settings, date, and whether any search, calculator, code, image, or browsing tool was enabled.
- Run the same request twice. Different wording may be normal; conflicting facts need verification.
- Change one variable, such as prompt specificity, source material, temperature, or tool access.
- Compare the result with an independent source or deterministic tool.
- Share the funny or surprising result with its context, not as proof that every model always behaves that way.
How to verify a surprising AI claim
| Claim type | Best verification | Do not rely on |
|---|---|---|
| Arithmetic or counting | Calculator, code, or manual count | The model’s confident explanation |
| Recent product fact | Current official documentation | Old screenshots or model memory |
| Scientific or historical claim | Primary paper or reputable reference | A citation that has not been opened |
| Image anomaly | Inspect the full image and generation settings | A cropped viral repost |
| Model behavior | Repeatable test with recorded conditions | One entertaining example |
For academic, employment, or disciplinary decisions, use documented process evidence, drafts, source notes, version history, and a fair conversation. Do not rely on one detector score or one reader’s intuition.
When you publish an example, label reconstructed prompts and edited screenshots clearly. Preserve the funny part without inventing a result the model never produced. For factual articles, separate durable concepts from dated product numbers and include the retrieval date for fast-changing claims. That small amount of context turns a viral anecdote into something readers can learn from and reproduce responsibly. It also prevents a model-specific quirk from being presented as a universal limitation across every AI system, version, and configuration currently available to users.
Use the example without turning it into a myth
Save the original prompt and output when possible, then label edits, reconstructions, and screenshots. When a model update changes the behavior, add that context rather than pretending the old example is still universal.
For a fun comparison, invite readers to reproduce the test and report the product, model, date, and settings. For a serious claim, require stronger evidence. This preserves the entertainment while teaching the more valuable habit: AI behavior is conditional, versioned, and best understood through repeatable observation. Include counterexamples when they materially change the conclusion, and avoid ranking products from a single playful prompt, anecdote, isolated result, edited example, or screenshot.
Official references and further reading
FAQ
Are AI detectors accurate? They’re imperfect and can produce false positives/negatives — useful as one signal among many, not a verdict on their own.
Is it bad if my writing has some of these patterns? Not inherently — good writing sometimes uses these devices intentionally. The issue is when they’re overused to the point of feeling generic.
Bottom line: these patterns are a fun pattern-matching game, not a detector — the real value is using them as a checklist to make your own writing (AI-assisted or not) feel less generic.