Products SEO & GEO Agent Paid Ads Agent Website Agent Conversion Agent
How it works Pricing FAQ Blog Book a call

Make ChatGPT undetectable? Reddit tricks vs. what works

Updated · 5 min read

Short answer For a business website, chasing "undetectable" is the wrong goal. Google judges pages on value, not on how they were written, and no detector score feeds rankings. Most popular tricks (typos, look-alike letters, translate-and-back, humanizer tools) make text worse or match Google's spam examples. The only trick that helps: add facts and experience only you have.

Want help applying this to your business? Talk it through on a free 30-minute call.

Book a call

Can you make ChatGPT undetectable?

Against one detector, on one day, sometimes. Reliably, no, and for a business website it doesn't matter. Google doesn't rank by detector scores. It acts on pages that offer "little to no value to users, no matter how it's created" (Google Search Central: Spam policies). So the useful question is not "will a tool flag this?" but "will a customer get something from it?" For why detector scores don't touch rankings, see AI detectors and SEO.

Here is how the tricks people share most often hold up.

TrickHelps the reader?Verdict
Add your own facts, prices, photos and storiesYes, a lotDo it
Cut stock phrases and filler ("In today's world", "delve")YesDo it, for readers
Prompt for "high perplexity and burstiness"Rarely; often choppierSkip
Add typos on purposeNoBackfires
Swap letters for look-alike charactersNoBackfires
Translate to another language and backNoBackfires
Run it through a humanizer or paraphraserNoBackfires
Strip "hidden characters" to remove a watermarkNoMisses: the watermark is in word choice
Re-run detectors until all say "human"NoWasted time

Does prompting for "perplexity and burstiness" work?

It tunes a signal detectors moved past years ago. Perplexity means how predictable each next word is; burstiness means how much that predictability swings across the text. GPTZero, which made the terms popular, now says:

"As of autumn 2023, GPTZero no longer uses perplexity and burstiness for its AI detection because we migrated to a deep-learning based architecture." (GPTZero)

The prompt usually yields jumpy sentence lengths with the same generic content. Researchers did find "simple prompting strategies" could bypass detectors (Liang et al., 2023), but detectors get retrained, and a reader gains nothing either way.

Should you add typos or "human mistakes"?

No. Google's own quality checklist asks:

"Does the content have any spelling or stylistic issues?" (Google Search Central: Creating helpful content)

The same list asks whether a page looks "sloppy or hastily produced." A typo in a quote or a price makes a customer trust you less. Deliberate errors trade a real cost for a guess about one tool.

Do look-alike letters or invisible characters fool detectors?

They fool some, and they break your page. A 2024 study swapped letters for identical-looking ones (a Latin "A" for a Cyrillic "А") and found the trick "can effectively circumvent state-of-the-art detectors" (Creo and Pudasaini). But to a computer, a word spelled with a Cyrillic letter is a different word. The customer searching "roof repair" is typing Latin letters, not the ones on your page. Copy-paste, spell check and translation tools hit the same wall.

Does translating it and back help?

It garbles meaning, and Google names it. Google's scaled content abuse examples include generating pages "through automated transformations like synonymizing, translating, or other obfuscation techniques" where little value is provided (Google Search Central: Spam policies). A few translated pages won't trip that by themselves, but a service page that comes back saying "ceiling restoration" for "roof repair" loses the customer anyway.

Do humanizer and paraphrasing tools work?

They reword; they don't add anything. Google's scraping examples include content modified "only slightly (for example, by substituting synonyms or using automated techniques)" (Google Search Central: Spam policies). That rule is about copying other sites, but it shows how Google sees synonym-swapping: no added value. Detector makers also adapt. Turnitin says it "will flag if a student has likely used an AI bypasser" (Turnitin, August 28, 2025).

Does ChatGPT put a hidden watermark in its text?

In the EU, it is starting to. On October 5, 2026, OpenAI said that over the coming weeks it will add "an invisible watermark to eligible ChatGPT and Codex text output in the European Union," and that API customers anywhere can opt in (OpenAI). It is not a global default. The watermark is "an invisible statistical signal to the model's word choices," not hidden characters, so "cleaner" tools that strip characters don't touch it. OpenAI also says editing can weaken it and its detector is open only to approved researchers.

What actually makes AI text read like a person wrote it?

What a person adds. Google's checklist asks whether content shows "first-hand expertise and a depth of knowledge" (Google Search Central: Creating helpful content). AI can't supply that; you can.

  1. Give ChatGPT your notes first: real prices, job details, the questions customers ask you.
  2. Cut the opening and closing paragraphs. They're usually filler.
  3. Swap every vague claim for a number or a named example.
  4. Check every fact, rule and price.
  5. Read it aloud. Change anything you'd never say to a customer.

More on which human parts Google rewards: AI content plus human input. To keep your own voice: AI content creation that keeps your voice.

When the text pretends to be a customer. The FTC's 2024 rule bans businesses from writing or buying reviews that misrepresent "that the reviewer or testimonialist exists" or had experience with the business (16 CFR 465.2), and the FTC's announcement names AI-generated fake reviews (FTC, August 14, 2024). Making a ChatGPT review "undetectable" doesn't make it legal. See AI-generated reviews.

Example

A plumber who added typos

He sprinkles a few typos into AI-written service pages after reading a tip online. A detector score moves; two customers mention the spelling on calls. He fixes the typos and adds his $150 drain-clearing price and a photo of a real job instead.

A bakery using a $20-a-month humanizer

The owner runs every blog post through a humanizer. The posts come back longer and vaguer. She cancels it and spends the time adding one seasonal recipe note per post that only her kitchen knows.

A landscaper with translated pages

He machine-translates his English pages to Spanish and back to "make them unique." Half the plant names come back wrong. He keeps the original English, fact-checks it, and adds before-and-after photos.

FAQ

Is it illegal to make AI text undetectable?

Editing AI text isn't illegal. Passing it off as a real customer's review is, under the FTC's fake review rule. Schools, clients and publishers can also set their own rules.

Will Google penalize ChatGPT content that a detector catches?

No. Google judges the page's value, not a detector score. See Does Google penalize AI content?

Does removing em dashes make ChatGPT text undetectable?

No source shows it does. Swapping punctuation changes how text looks, not what it says.

Which AI detector does Google use?

Google hasn't said it uses any. Its spam policies target low-value pages however they're made.

Different for you?

Keep learning

NextAI score meaning: how to read AI detection scoresWhat a 30% or 90% AI detection score means in Turnitin, GPTZero and Originality.ai, how often human writing gets flagged, and what to do next.

Ask us one question

One line is enough. We reply by email.

Sources

  1. Google Search Central: Spam policies for Google web search (updated 2026-08-28)
  2. Google Search Central: Creating helpful, reliable, people-first content (updated 2026-10-05)
  3. OpenAI: Our approach to EU text provenance rules (October 5, 2026)
  4. GPTZero: Perplexity, burstiness, and statistical AI detection (with later update note)
  5. Creo and Pudasaini, SilverSpeak: Evading AI-Generated Text Detectors using Homoglyphs (COLING 2025 workshop)
  6. Liang et al., GPT detectors are biased against non-native English writers (Patterns, 2023)
  7. Turnitin: What are AI bypassers, and why should educators care? (August 28, 2025)
  8. FTC: Federal Trade Commission Announces Final Rule Banning Fake Reviews and Testimonials (August 14, 2024)
  9. eCFR: 16 CFR 465.2, Fake or false consumer reviews and testimonials