Beginner · Module 08
Recovery — When It Does Not Work
After this, you will know when to correct within a conversation, when to restate from scratch, and when to start a new conversation entirely.
Intro
When a response misses the mark, there is a right response and a wrong one. The wrong response is to pile corrections on top of a bad start. This module teaches the right ones.
What is happening
B-07 taught you to read a response before reacting. This module covers what to do when reading reveals a real problem — not a minor adjustment, but a response that went in the wrong direction.
There are three tools available to you, in order of how much work they require:
A correction in the same conversation works when the AI understood the task but got one detail wrong. "Make it shorter." "Change the tone." "Remove the last paragraph." These are adjustments, not changes of direction.
A restate from scratch in the same conversation works when the AI misunderstood the task itself. You write a new, clearer version of your original message — not a correction of what it produced, but a fresh attempt to explain what you actually wanted. Keep the same conversation, but write a new first message as if the previous exchange had not happened.
A new conversation works when the current conversation has drifted too far. If you have tried corrections and restates and the responses keep circling the same wrong interpretation, the AI has accumulated too many conflicting signals in the current conversation. Starting fresh clears all of that. You bring only the essential information forward.
The reason layering corrections often makes things worse: each correction adds to the context the AI is working from. It tries to reconcile all of it — your original message, what it produced, your correction, what it produced next, your next correction. By the third or fourth correction, it may be as confused as it is capable of being. Starting fresh is faster than fixing a tangled thread.
The exercise
Find a response from your own use that missed the mark. It can be from this course or from any time you have used an AI tool. If you do not have one handy, use a message you send in Step 1 below.
Step 1: Create a situation to recover from
Send this message to the AI:
Help me write something.
Read the response. Notice how vague and generic it is. This is what happens when a message gives the AI nothing to work with.
Step 2: Try a correction in the same conversation
Send this follow-up in the same conversation:
Actually I need it to be more professional and shorter.
Read the result. Notice whether it improved, and whether the AI is still guessing at what "something" means.
Step 3: Restate from scratch in the same conversation
Still in the same conversation, copy this message and personalise every bracket before sending:
I need to start this over. I'm a [your role] and I need to [specific task — e.g. "draft a short email declining a meeting invite"]. [Stopping condition — e.g. "Keep it under 80 words"]. [One constraint — e.g. "Professional but not stiff" or "No excuses or explanations"].
Read the result. Notice whether it is closer to what you wanted.
Step 4: Start a new conversation with a better opening message
Open a new conversation. Take what you learned from Steps 2 and 3 — what the AI needed to know, what you actually wanted — and write a clear first message from scratch. Use the restate template from Step 3 as your starting point.
Compare the results from all three steps. Which approach produced the most useful response?
COPY — PERSONALISE — USE
The restate template:
I need to start this over. I'm a [your role] and I need to [specific task]. [Stopping condition]. [Constraint].
How to personalise it:
[your role] — who you are, in a few words. This gives the AI the context it lacked before.
[specific task] — start with a verb. "Draft." "Summarise." "Explain." Then say exactly what you want that verb applied to. Be more specific than you were in the original message.
[stopping condition] — how you will know it is done. "Under 100 words." "3 bullet points." "One paragraph only."
[constraint] — what you do not want, or the boundaries of the task. "No jargon." "Professional but not formal." "Do not suggest things I did not ask for." One constraint is enough.
What good looks like
The restate in the same conversation, or the clean start in a new conversation, should produce a noticeably better response than the correction attempt. If you find yourself on your third or fourth correction of the same response, the restate or the new conversation will almost always be faster.
If this did not work
If the restate in the same conversation still felt like it was fighting the earlier context, trust that instinct — move to a new conversation. There is no benefit to staying in a conversation that has drifted. The AI has no preference. Every new conversation starts completely fresh.
If you cannot identify what was wrong with your original message, go back to the output review checklist from B-07. What did the AI miss? Was the task unclear? Was the context missing? Was there no stopping condition? The answer tells you what to add to the restate.
Next
Now that you can recover when things go wrong, B-09 shows you how to set up standing instructions that apply to every conversation — so you spend less time repeating yourself each time you start a new one.