Avoid ChatGPT detection? Edit out the tells readers notice
Short answer For your website, you don't need to beat a detector: Google judges quality, not how a page was made. What costs you is the reader spotting AI and trusting you less. Edit out the tells (stock words like "delve," lists of three, puffed-up importance, vague "experts say," leftover chat lines) and add facts only your business knows.
Want help applying this to your business? Talk it through on a free 30-minute call.
Book a callDo you need to avoid ChatGPT detection?
Not for search. Google ranks the page, not the tool that typed it:
"Appropriate use of AI or automation is not against our guidelines." (Google Search Central Blog, February 8, 2023)
The same post says Google's focus is "on the quality of content, rather than how content is produced." Detector scores play no published role in ranking, and Wikipedia's AI cleanup guide warns the detectors themselves "have non-trivial error rates"; AI detectors and SEO has the accuracy record.
The real risk is people. A customer who reads "In today's fast-paced world" on a roofer's site stops believing the roofer wrote anything. So the goal changes from "fool a tool" to "sound like a person who has done this job." That is an editing task, and the tells are well documented.
One case where hiding AI is the problem itself: reviews. The FTC's rule bans fake reviews, including those "by someone who does not exist, such as AI-generated fake reviews" (FTC, August 14, 2024). No edit makes that legal. See AI-generated reviews.
What gives AI writing away?
Eight common tells. Researchers measured the word list; Wikipedia's volunteer cleanup team documented the rest in its field guide "Signs of AI writing."
| Tell | What it looks like | The edit |
|---|---|---|
| Stock words | "delve," "underscores," "showcasing," "pivotal," "crucial," "tapestry" | Use the plain word: "look at," "shows," "key," or nothing |
| Puffed-up importance | "plays a pivotal role," "stands as a testament to" | Say what it does: "cuts heating bills about a third" |
| Lists of three | "fast, reliable, and affordable"; three short phrases in a row | Keep the one that's true and provable; cut the others |
| "Not just X, but Y" | "It's not just a cleaning, it's peace of mind" | State Y plainly, with a fact |
| Vague sources | "Experts agree," "studies show" | Name the source, or delete the claim |
| Wrap-up endings | "Despite these challenges, the future looks bright" | End on the last useful fact, or the next step for the customer |
| Chat leftovers | "Certainly!", "I hope this helps," "Let me know if you'd like..." | Delete. These were meant for you, not the reader |
| Formatting habits | Every heading in Title Case, bold in every sentence, emoji before headings | Sentence-case headings; bold one thing per section at most |
The word list isn't folklore. A study of over 15 million medical paper summaries found words like "delves," "underscores" and "showcasing" jumped in 2024, far above their earlier trend. From that excess, the authors estimate at least 13.5% of 2024 summaries were processed with AI tools (Kobak et al., Science Advances, 2025). Readers see the same words everywhere now, which is why they stand out.
Which tells are out of date?
Some famous tells no longer hold, so don't waste edits on them.
- Em dashes (the long dash). Wikipedia's guide now lists heavy em-dash use as a "historical" sign. It says some AI companies tried to make newer chatbots, OpenAI's GPT-5.1 among them, use fewer, and a July 2026 study it cites found ChatGPT used them less than professional writers.
- Perfect grammar and formal tone. The guide lists these as "ineffective indicators." Plenty of people write cleanly.
- Starting sentences with "Additionally." Old AI did this a lot, but the guide notes many human style guides allow it, so on its own it proves little.
Tells shift as models change. The ones that last are about content, not punctuation: vague claims, filler importance, no first-hand detail. Fix those and the style tells mostly go with them.
How do I edit AI text so it sounds like me?
Work in this order, because each pass makes the next one shorter.
- Delete the chat leftovers. Search for "Certainly," "I hope," "Let me know," "Here is." They never belong on a page.
- Cut the opening and the closing paragraph. AI intros restate the question; AI endings summarize what was just said. Start with the answer. End on the last useful line.
- Search the stock words. Use your editor's find tool on the words in the table. Replace each with the plain word, or cut the sentence if nothing is left.
- Turn every vague claim into a fact you know. "We offer fast service" becomes "Most repairs are done the same day; parts we don't stock take 2 to 3 days." This is the step no prompt can do for you.
- Break up the threes. Keep the item a customer would care about. A list of two or four reads as a person choosing.
- Read it out loud. Any sentence you wouldn't say to a customer at the counter, rewrite the way you would say it.
- Check every number and rule. Google's guidance calls it "critical to manually factcheck and review all AI-generated content for accuracy and trustworthiness before publishing" (Google Search Central).
Step 4 is also what Google asks for. Its self-assessment questions include:
"Does the content provide original information, reporting, research, or analysis?" (Google Search Central)
A page full of your own prices, timelines and job details answers yes, whoever wrote the first draft. More on proving experience in E-E-A-T for a small business.
Can a better prompt prevent the tells?
Partly. Telling ChatGPT "no intro, no conclusion, plain words, don't use the words delve, pivotal or crucial" removes some style tells. It can't add facts it doesn't have, so the vague-claim tell survives any prompt. The fix is to feed it your knowledge first; ChatGPT prompts for blog writing shows an interview prompt that does that.
Should I use an AI humanizer tool?
For a business site, no. These tools reword text to change a detector score. They don't add facts or experience, which is what Google says it rewards, and rewording can introduce errors you then have to catch. Google also warns against producing many pages "without adding value for users" (Google Search Central), however they're worded. Your editing time does more than any rewording tool.
Example
A plumber's service page
The AI draft says: "Our team of skilled professionals delivers reliable, efficient, and affordable solutions." The owner rewrites it: "Two licensed plumbers, both with 15 years on the job. Water heater swaps are usually done in 3 hours. Flat rate quoted before we start." Same length, and now only this plumber could have written it.
A bookkeeper's blog post
The draft ends: "Despite these challenges, staying organized is crucial for long-term success. I hope this helps!" She deletes both lines. The post now ends on her last tip, about saving receipts in one folder per month.
A dental office's FAQ
A staff member worries the FAQ "sounds like AI" because it's clean and formal. Checked against the table, it has no stock words, no threes, and real prices and hours. Nothing to change. Clean writing isn't a tell.
FAQ
Will Google penalize my site if a detector says my page is AI?
Google has published no use of detector scores. It acts on content made mainly to manipulate rankings, however it was produced (Google Search Central Blog).
Does removing em dashes make text undetectable?
No. Wikipedia's guide now treats heavy em-dash use as an outdated sign, since newer models use fewer. Vague, fact-free writing gives AI away far more than punctuation.
Can people reliably spot AI writing?
Not well. Wikipedia's guide warns "Humans are notoriously bad at distinguishing human and LLM-generated text." That's another reason to edit for the reader, not for a score.
Should I tell customers I used AI?
Only where they'd wonder how the content was made. Google's guidance says sharing how content was created "can help give your readers more context" (Google Search Central).
Different for you?
- A client or platform runs your content through a detector: AI detectors and SEO
- You want ChatGPT to draft in your voice from the start: ChatGPT prompts for blog writing
- You're tempted to have AI write reviews: AI-generated reviews
- You'd rather have pages written and kept current for you: SEO agent