Skip to main content

Humanify

☰
Best ChatGPT Prompts to Avoid AI Detection

If you have pasted a draft from ChatGPT into a detector and flagged it, then perhaps the next step in this case is to try a more robust prompt. If you use the right prompt for ChatGPT, the text will be different, it’ll feel more like you than like some template. While there is no magic or magic words to make a clean result 100% of the time, a few do well. Here they are and what each one is doing. The following blog will discuss how to avoid AI detection.

Is there any way to avoid being detected as AI by a prompt?

Yes, to a point. AI detectors scan for common patterns in generic model output, such as length of sentences, lack of “small hedges and asides,” flat or too-polished phrasing. Instead of editing them out after creation, you can tell ChatGPT to not use those patterns in the first place.

The secret to making it difficult for ChatGPT to detect AI is to focus on the specific patterns you want to discourage rather than simply saying “make this sound human.”Getting ChatGPT to avoid being detected as AI isn’t as much about making it “sounds human” as it is about targeting the specific patterns to discourage.

A prompt will not ensure a clean score. Detectors don’t always agree and they are updated regularly, so sometimes a quick “clear” in one detector tool does not clear the other, or the same detector a month later. Use each prompt listed here as a means to decrease AI patterns, rather than aim to make text impossible to detect forever.

Best ChatGPT prompts to avoid AI detection

AI detectors are primarily detecting three key factors: sentence length variance, the use of concrete or generic language, whether the language is used in a way that mimics thinking aloud. All the following prompts are directed at one of those signals, which you can use to select the signals that correspond to your text rather than guessing. Afterwards, pass through a detector. Such prompts eliminate boring AI writing, but they don’t guarantee it won’t be detected.

1. Vary sentence length and rhythm

Target: Even sentence length, uniform (good tell of AI output)

Prompt: Write each sentence below in a natural way, varying sentence length. Use short, concise sentences (5-8 words) and longer sentences (20+ words). Try to keep the average to be between 15 and 16 words per sentence – but don’t do it mechanically by alternating between different sentence lengths. Paste: [paste your text]

This is often the most helpful hint on this list, as one of the most obvious differences between human and AI writing is the rhythm of the sentences.

2. Add specific action steps and numbers

Target: instead of vague, generic statements

Prompt: Rewrite this text to include specific actions and numbers when they are applicable. Don’t make vague statements like “somebody did something,” or “the situation changed in some way,” or “the event happened” but rather use actual numbers if I provided them. Type your text:

3. Speak in your own jargon and be casual

Target: generic business language, not a specific domain language.

Prompt: If I am talking to a colleague who is knowledgeable about this area, how would I explain it? Do not employ jargon from your industry or topic – use natural jargon, not generic business jargon – and do not use anything that sounds too corporate. Paste your text here:

4. Start sentences with natural discourse markers

Target: lack verbal skills people are not aware of

Prompt: Rewrite this text and begin a few sentences with phrases that would typically open a conversation like “So,” “Honestly,” “For now,” “Of course,” or “Well.” Avoid shoehorning it into every sentence, but enough to make it sound as though someone were thinking aloud. Spell: [spell your text]

5. Turn it into concrete sequential steps

Target: abstract summaries instead of a real process

Prompt: Restate this as a series of specific steps/actions in the way that someone would explain a process he or she is performing at this very second: what happens first, what happens next, what does the next step in the process look like, etc. Do not summarize, tell what actually happened. Text: [insert your text]

6. Include hedged, reasoned explanations

Target: points that are made without providing any justification.

Prompt: Write a text that uses a couple of hedged, reasoned explanations rather than flat statements, such as, “it looks like,” “so far,” or “because right now.” State and explain a claim rather than just make a statement. Paste your text here:

7. Add one more every day

Target: too generic or profound language

Prompt: You will want to reword this text and include one simple, easy-to-understand metaphor or comparison, something from the real world, not some abstract or technical comparison. Only use one, don’t use too much. Copy and paste your text into the text box.

8. Use simple verbs and informal conjunctions

Target: Uses transitions that are too formal or nominalizations.

Prompt: Convert abstract nouns to concrete, active verbs: abstract – actualization of the plan was done, concrete – we did the actualization of the plan. Replace formal words (then, by the end and and) with everyday words. Text: [paste your text]

9. Restate it in your own words

Target: Speaks in a voice other than his/her own

Prompt: Here is a sample of my own writing: [paste 2-3 paragraphs that I wrote myself]. Examine the sentence length, tone and word choice. Rewrite the following text in the same voice as the original, but as close as possible. Duplicate the style, but not the content of the sample. 

This one is more effective on having a sample of writing to eat. If a prompt could sound like you, not “less” like AI, it would be the closest it can get.

How to use these prompts?

Don’t place all 9 at once. Choose 2 or 3 related to the signals you feel the draft is missing (a stiff uniform paragraph would probably require #1 and #6 more than #7, for example), run the rewrite, and then read it back. If it sounds like a template, repeat it using a new prompt from the list.

When you’re satisfied with your version, double-check against an AI detector before using it anywhere. You can only determine if a specific AI detection prompt was effective for your text if you compare the results of the various detectors, and the degree of AI use in the original text.

Conclusion

The key to avoiding AI detection is to target signals that are associated with a specific task, rather than asking ChatGPT to be “vaguely human.” Try to use 2 or 3 that correspond to what your draft is lacking, then test with a detector, and if it is still flagged up, an AI “humanizer” can then completely humanize it before you submit or publish.

FAQs

Q1. Is it possible to trick ChatGPT?

Yes, a well-targeted prompt can decrease the number of patterns that are detected, such as sentence length or general phrases. Don’t rely on it, as detectors change and update, but it is a very initial step before picking up any other tool.

Q2. Which is the most effective prompt to get AI text to sound human?

The sentence variation prompt (a mix of short and long sentences with uneven rhythm) seems to have the greatest impact when used in isolation because sentence length is one of the more obvious AI signs. It is often supplemented by a hedged-explanation prompt.

Q3. Can prompts fool AI into making text unrecognizable?

No one single prompt is going to make text undetectable forever, and assuming anything is a prompt to bypass AI detection is not going to work. Prompts eliminate AI style patterns in the generation process. If the detector continues to flag the result a humanizer passes and a recheck is the safest move.

Q4. How to get ChatGPT to write naturally?

Encourage it to alter sentence length, substitute specific detail for general statements and incorporate small natural features such as the use of hedged phrases or a casual discourse marker. Telling it to read a bit of your own writing and copy that style is better than telling it to use a generic voice.

Q5. Are these the same as ChatGPT prompts to avoid AI detection?

They both process the same signals, but in a different way. The prompts have been designed to ensure that the text appears more natural, rather than fool a particular detector, so results should not be considered an automatic pass.