# Text editing with AI has been getting worse lately — retained record

> Retained reddit record from reddit:ChatGPT.

Canonical: https://highsignal.app/data/records/647a39e917ae6217

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:ChatGPT)  published 2026-11-13 16:29 ingested 2026-11-14 02:35

[original source ↗](https://www.reddit.com/r/ChatGPT/comments/1xvceug/text_editing_with_ai_has_been_getting_worse_lately/)

## retained text



```
Please tell me I'm not alone. This was never an issue in my previous use of AI bots, but over the past month or two I've noticed a recurring change in their behavior (chatgpt, claude, qwen, deepseek. All with think mode enabled): when asked to modify existing text, they increasingly describe the requested changes instead of applying them to the actual text they output.

**Deliberately simplified example:**

**Text I want to modify:** "A cat sitting on a table.**"**

**Instruction to AI:** Replace the cat with a bird and the table with a tree branch.

**Expected output:** A bird sitting on a tree branch.

**Actual output:** "The cat is replaced with a bird, sitting on a tree branch instead of the table."

ARE YOU BEING FOR REAL??? 😠

With complex multi paragraph prompts, the AI often mixes previous edits, obsolete wording and change descriptions into the output instead of simply applying the requested changes.

You then have to review and clean up the entire prompt yourself. Despite explicit system instructions to return only the complete revised text, this keeps happening... I've started noticing this much more frequently lately. This was never a pattern I observed.

(I'm a free user with no subscription, so it's possible the free models have been significantly downgraded.)

Has anyone else experienced this?
```



document fetched 2026-11-14 00:20 · canonical https://reddit.com/r/ChatGPT/comments/1xvceug/text_editing_with_ai_has_been_getting_worse_lately
