https://omgfixmd.com/blog A field manual for giving LLMs precise, multi-point feedback. Diagnoses, playbooks, and per-tool recipes for Claude, ChatGPT, Cursor, Gemini, and AI-builder surfaces like Lovable. en-us © 2026 Elad Diamant Sun, 26 Apr 2026 00:00:00 GMT https://omgfixmd.com/og.png https://omgfixmd.com/blog https://omgfixmd.com/blog/why-typing-feedback-into-the-chat-stops-working https://omgfixmd.com/blog/why-typing-feedback-into-the-chat-stops-working Sun, 26 Apr 2026 00:00:00 GMT The chat box doesn't fail at five corrections. It fails at three. By the fourth, the reviewer is knowingly shipping a message they expect the model to mis-target — some shipped feedback beats none — and the fifth correction lives its entire life inside their head. A close-read diagnostic of the typing-the-feedback failure mode, scene by scene. https://omgfixmd.com/blog/chatgpt-rewrites-whole-document-when-i-ask-for-edits https://omgfixmd.com/blog/chatgpt-rewrites-whole-document-when-i-ask-for-edits Fri, 24 Apr 2026 00:00:00 GMT You asked for one small edit. It came back as a different document. Not your prompt — the chat box has no target selection, so every ask reopens the whole file. Here's what's actually happening and the format that stops it. https://omgfixmd.com/blog/prompt-to-edit-only-specific-passages https://omgfixmd.com/blog/prompt-to-edit-only-specific-passages Fri, 24 Apr 2026 00:00:00 GMT Every "only edit X, leave the rest alone" prompt ranked by reliability, with the honest answer about why none of them work on their own — and the paired-passage format that does. Copy-pasteable templates included. https://omgfixmd.com/blog/claude-projects-vs-attaching-doc-scoped-edits https://omgfixmd.com/blog/claude-projects-vs-attaching-doc-scoped-edits Fri, 24 Apr 2026 00:00:00 GMT Projects promise persistent context. Attachments promise fresh reads. Neither scopes your edits. A frank comparison of both, plus the format that does keep Claude editing only the passages you named. https://omgfixmd.com/blog/give-cursor-feedback-long-markdown-file https://omgfixmd.com/blog/give-cursor-feedback-long-markdown-file Fri, 24 Apr 2026 00:00:00 GMT Cursor's tools are tuned for code, where ambiguity is rare. Long Markdown prose has more ambiguity per line — and Cursor edits it like code, which is why the README keeps coming back rewritten. The playbook for keeping prose edits scoped. https://omgfixmd.com/for/lovable https://omgfixmd.com/for/lovable Fri, 24 Apr 2026 00:00:00 GMT Lovable is great at the first screen and harder at the fifth edit. Here is the feedback pattern that keeps scope tight — quote the exact span of the generated Markdown or copy that needs a change, pair it with a note, send the whole batch in one turn. Works from the chat box, works better with a tool. https://omgfixmd.com/blog/give-llm-feedback-multiple-passages https://omgfixmd.com/blog/give-llm-feedback-multiple-passages Tue, 21 Apr 2026 00:00:00 GMT The chat box loses at multi-point feedback. Every LLM user hits this wall around comment five. Here is the pattern that actually works — paired passages with notes — with a manual recipe for anyone, and a tool for anyone tired of doing it by hand. https://omgfixmd.com/manifesto https://omgfixmd.com/manifesto Tue, 21 Apr 2026 00:00:00 GMT Every LLM user hits the same wall around the fifth comment: the chat box is a text area, and structured feedback on a 1,800-word answer is not prose input. A manifesto for why LLM answers need a feedback layer — and what it looks like when somebody bothers to build one. https://omgfixmd.com/blog/how-to-comment-on-markdown https://omgfixmd.com/blog/how-to-comment-on-markdown Tue, 21 Apr 2026 00:00:00 GMT Markdown has no native comment syntax. Every practical method — HTML comments, Google Docs, Notion, GitHub PR, Slack, purpose-built comment layers — ranked, with the 2026 case that matters most: giving the file back to the language model that wrote it.