Beginner · Module 07

What Just Happened? Reading AI Output

After this, you will be able to read an AI response, decide whether it answered what you actually asked, spot the signs that something went wrong, and know what to do next.

Intro

The AI's first response is a draft. You are allowed to push back, ask for changes, or start over. Reading it carefully before you react is the skill.

What is happening

Many people assume the AI's first response is the final answer. It is not. It is an attempt — based on what it understood from your message. Sometimes that attempt is exactly right. Sometimes it missed the point. Sometimes it got the right topic but the wrong angle.

Reading the response before reacting — really reading it — is what separates a useful AI interaction from a frustrating one. If you ask for changes without reading carefully, you might ask for the wrong fix. If you start over without reading carefully, you might abandon something that was actually mostly good.

There are four signals that the AI may have misunderstood or reached its limits:

Generic language where you needed specifics. If you asked about your team's meeting agenda and the response sounds like it could apply to any team meeting in any company, the AI did not have enough to work with.
Wrong framing. If you asked for a casual message to a friend and got a formal letter, the AI misread the tone signal. If you asked for a summary and got an essay, the stopping condition did not register.
Made-up details. If the response includes names, dates, figures, or facts that you did not provide and that the AI could not know, treat those with caution. This is the AI filling in gaps you left open. It is not lying — it is guessing. Check anything that matters.
Outdated information. AI tools are trained on data up to a certain date. If the response gives you information about something that changes frequently — laws, prices, current events, recent news — it may be out of date. Check any time-sensitive facts from a current source before acting on them.

The exercise

Send a real task — something from your actual work or life. Use any message style you have practised so far. Then use the checklist below before doing anything with the response.

The output review checklist

Work through these four questions after every response, until it becomes automatic:

1. Did it answer what I actually asked?
Not a related question, not a broader version of the question — the specific thing you asked. If not, what did it answer instead?

2. Is anything generic where it should be specific?
If you gave the AI details about your situation, those details should show up in the response. A response that ignores your context is usually a sign the message needed more specific information.

3. Did it invent anything I did not give it?
Names, statistics, quotes, specific recommendations — if you did not provide them and the AI cannot look them up, verify before using.

4. Is the length about right?
If you asked for three bullet points and got twelve, something did not land. If you got one sentence when you needed a paragraph, the message may have been too short.

Keep this checklist. You will use it for every module from here on.

COPY — PERSONALISE — USE

Once you have read the response and identified what needs changing, here is how to ask for a specific change:

The response is [what it got right — e.g. "mostly good" / "the right length" / "correct in tone"]. Please change [the specific thing that needs changing]. [New instruction for that part — e.g. "Make the second paragraph shorter" / "Remove the introduction" / "Use a more casual tone throughout"].

How to personalise it:

[what it got right] — acknowledge what to keep so the AI does not rewrite everything. "The bullet points are good" or "the structure is right" or even "most of this is fine."

[the specific thing that needs changing] — be precise. Not "make it better" but "the third paragraph is too formal" or "the summary at the end is not needed" or "the second bullet point missed the point."

[new instruction for that part] — tell the AI exactly what to do with it. "Remove it." "Shorten it to one sentence." "Rewrite it in plain language."

What good looks like

After using the checklist, you should be able to make a clear decision: use it as-is, ask for one specific change, or start over with a better message. Any of these is a valid outcome. The checklist helps you avoid the worst outcome — using a response that missed the mark without noticing.

If this did not work

If you asked for a specific change and the AI rewrote everything instead, your change request may have been too broad. "Make it better" will trigger a full rewrite. "Change the third sentence to be less formal" will not. Be as specific as possible about what to change and what to leave alone.

If the response was so far off that no single change would fix it, B-08 covers what to do when you need to start a conversation over from scratch.

Your take-home reference

Output review checklist:

Did it answer what I actually asked?
Is anything generic where it should be specific?
Did it invent anything I did not give it?
Is the length about right?

Now that you can read a response critically, B-08 covers what to do when reading reveals a bigger problem — when the AI has gone far enough off track that a small correction will not fix it.