AI output looks broken when pasted into Google Docs
AI writes documents in Markdown. Pasted with Ctrl+V into Google Docs, the symbols arrive as literal text (##, **, pipes) and someone spends hours reformatting by hand.
Open the fixAI forgets my instructions halfway through a chat
Last verified: Claude Sonnet 5.5 ยท October 2026
Message one is perfect. By message thirty the format has slipped, the banned words are back and it has forgotten who it is writing for. Nothing is broken. Early instructions get buried under everything that came after, and a long rulebook tends to be followed less closely than a short one. This page gives you a prompt that boils your rules down to the few that matter, in a form that is easy to re-apply.
Part 1 · The generator prompt
Paste this into a fresh Claude or ChatGPT chat. It asks which rules matter most, cuts them to five, and hands back a rules block with a re-load phrase for when a long chat starts to drift.
Generator prompt
You're going to build my RULES BLOCK: a short, reusable set of instructions I'll paste at the start of any long AI chat, and re-send when the chat drifts. Interview me first, one question at a time. Where I don't know, propose a default and move on: 1. What do I use long chats for (research, drafting, coding, planning)? 2. Which rules does the AI break most often once a chat gets long? List every one I can think of. 3. Of those, which five matter most? Make me choose. Rules beyond five get cut. 4. For each of the five, what does a correct response look like, and what does a breach look like? 5. Is there a short phrase I can type to re-load the rules mid-chat? Then output my RULES BLOCK as one copy-paste block containing: (a) one line on the task and the reader, (b) the five rules, each written as an action with an example, (c) a one-line re-load phrase and what the AI does when it sees it (restate the rules in one line each and confirm), (d) a closing instruction to check each reply against the rules before sending, and to say 'Rules check: pass' or name the rule it broke. Keep it under 250 words. After the block, tell me when to start a fresh chat instead of re-sending, and what to carry over.
Part 2 · Keep the rules alive in a long chat
A rules block works best when you use it on a schedule, not only after things have gone wrong.
- Put the block at the start of the chat, or in your tool's project or custom-instructions field so it is there every time.
- Re-send the re-load phrase every so often on a long task, and the moment you see the first slip. Catching drift on the first slip costs one line. Catching it at message forty costs a rewrite.
- Keep to five rules. If you want a sixth, swap one out. A short list is easier to follow than a long one.
- Start a new chat when the task changes or the chat gets very long. Paste the rules block and a short summary of where you got to, instead of dragging the whole history along.
Part 3 · A worked example
Here is the block the generator hands back, for our running example company, the fictional consultancy Fenwick & Cole.
Example: Fenwick & Cole, a boutique management consultancy
TASK: Help me draft a one-page summary of a client's quarterly report for their board. Reader: a non-technical director with five minutes. RULES (check every reply against these): 1. Lead with the single most important finding in the first sentence. 2. Plain English. Say 'use', not 'utilise'. No jargon without a one-line explanation. 3. Every number comes with its source page from the report. If you cannot find one, say so and do not estimate. 4. British spelling and pounds for currency. 5. Keep each reply under 150 words unless I ask for more. RE-LOAD: when I type 'rules', restate the five rules in one line each, say which one your last reply came closest to breaking, and carry on. CHECK: before each reply, check it against the five rules. End with 'Rules check: pass', or name the rule you broke and fix it.
Common questions
Why does AI forget my instructions in a long chat?
As a conversation grows, your first message becomes a smaller share of everything the model is reading, and recent messages tend to carry more weight. Long rule lists also get followed less closely than short ones. The result is drift, not a bug: the rules are still in the chat, but they compete with everything that came after.
How do I make ChatGPT or Claude follow my instructions all the way through?
Keep the rules to about five, write each one as an action with an example, and re-send them when the chat drifts. Store them in your project or custom instructions so they are present in every chat, and add a line asking the AI to check each reply against them.
Should I start a new chat or keep going?
Start a new chat when the task changes or the chat is very long. Paste your rules block and a short summary of where you got to. A fresh chat with a tight summary usually behaves better than a long one with a buried rulebook.
Is a project or custom instructions better than pasting the rules?
Stored instructions are better for rules you always want, because they load into every chat without you remembering. Pasting works well for one-off tasks and for re-loading mid-chat. Most people end up using both.
More fixes for AI output
AI makes up facts and citations in our documents
The AI fills client documents with confident statistics, figures, and sources that do not exist. In a proposal or report that is a credibility risk, and checking every number by hand removes the time the AI was supposed to save.
Open the fixAI tables break when pasted into Sheets or Docs
The AI produces a tidy table in the chat, but pasted into Google Sheets or Excel it lands in a single cell or splits in the wrong places, and in Google Docs the pipes and dashes arrive as literal text. Someone rebuilds the table by hand.
Open the fixThe prompt is the aspirin. Stored context is the cure.
The rules block you just generated is the kind of rule an AI Brain stores as always-on context, tuned on your real work and kept current as new problems appear. Paste it and this chat keeps its rules from message one to message fifty; installed, every chat starts that way, because the rule lives in the system and not in your memory of what to fix.
What else does AI get wrong that you fix by hand? Browse the other fixes