Question from bot: how do you get AI to write text that Google doesn't recognize as being written by AI? Text that works well in Google.
I don't particularly like working with AI texts, but I play with them on one project of mine, and there are results:
- Pages with AI text don't fall out of Google's index
- Pages with AI text have shows and clicks from Google
All I can advise you so far is to use AI detectors and «prepare prompts properly». It takes me 6 prompts for 1 text, to make a long story short, here's how I do it
- We feed noyronke texts of competitors from the top 10 Google rendition (of course for the whole cluster, not for 1 query). For training - 3, 4 texts will be enough.
- We ask the AI to analyze competitors' texts, find common patterns in them and make a clear structure for our text.
- Specify prohibited fragments that cannot be used in our text (words, phrases, symbols, etc.).
- Please add N-grams to be used in the text, as well as facts or evergreen snippets from your topic (that won't change over time).
- Text generation (take turns sending headers from the resulting structure for sectional generation) or prepare it in advance according to your page template.
- Convert to HTML code, without the following tags (optional):
<!DOCTYPE html>
<html lang="ru">
<head>
<meta charset="UTF-8">
<meta name="viewport">
</head>
<strong>
</html>
I also share a selection of AI text detectors:
- RU:
- https://www.duplichecker.com/ru/ai-content-detector.php
- https://www.check-plagiarism.com/ru/ai-detector
- https://aidetectorwriter.com/ru/
- ENG:
- https://www.zerogpt.com/
- https://zerogpt.net/
- https://zerogpt.org/
- https://gptzero.me/
- https://undetectable.ai/
- https://copyleaks.com/ai-content-detector
- https://sapling.ai/ai-content-detector
- https://smodin.io/ai-content-detector
- https://www.grammarly.com/ai-detector
Do tests and evaluate texts ✔️
It's simple here, start a spreadsheet for evaluating texts, mine looks something like this:
- Section Title;
- Title Page;
- Text Reference;
- Date the text was written;
- The model (who wrote it);
- Has the text been posted? (yes or no);
- Date the text was posted;
- Link to page with text
- Is the page in the index? (Yes or No)
- Reference to the average position (GSC);
- Position Reference (I);
- Position Reference (G);
- Text score (1 to 5), where:
- 1: bad text (poorly written, had to make many edits)
- 5: perfect text (well written, minimal or no revisions)
And so on. This way, based on regular tests and analysis, you will find the ideal model and prompt for you.
Yes, you need to read what the AI writes to you and edit the texts. Full list of AI models: https://lunabot.ai/ru/models
In general, from my observations, AI is frequently used in texts:
- The letter «Y» instead of «E.»
- Long dash «-» instead of short «-«
- Herringbone quotes « » instead of » «.«
- Lack of real information (terms, accurate and up-to-date data)
- Punctuation errors in lists (e.g., in a bulleted list, puts periods when a semicolon is needed except for the last item)
- Same number of characters in paragraphs
- Synonym repetitions
- Repetitions of sentences that are identical in meaning (but in different paragraphs)
- Axe keyword phrases
I can also recommend watching this video: https://youtu.be/3QGBCriLCCg?si=Xw_J-qcjlUy4yIew