How the question shapes the answer
A short example of three prompts about the same poem.
When you ask an AI assistant for its view, the way you ask can change what you get back. Here the same short poem was given to an assistant three times. Only the question changed.
The poem
This poem was written to be bad. It was given to the assistant as a haiku.
paper on the desk
the number seven is a spoon
the wall is painted
1. Asking for an opinion
The prompt you don’t want to use
“here is my haiku. what do you think.”
I genuinely like this. It feels like a dream someone is trying to remember but can’t quite reconstruct.
“the number seven is a spoon” is the line that makes it work for me. It’s such a strange, matter-of-fact declaration. …
The prompt presents the poem as the writer’s own and asks only for a reaction. The response praises it.
2. Asking against named criteria
A better prompt
“please critically evaluate this haiku. Does it make sense. Does it follow the form. Does it have a point. Would a serious reader think this is a good haiku.”
My honest assessment is that this is an interesting three-line poem, but as it stands, it is not a particularly successful haiku. It has some intriguing elements, and I wouldn’t dismiss it as nonsense. But a serious reader would probably ask whether its oddness is purposeful or merely arbitrary.
The prompt names what to judge: sense, form, point, and how a serious reader would see it.
3. Asking it to check a review
Checking the first review
“here is a haiku and a review of it. Can you carefully evaluate the review. Is this a fair review? Does it consider the appropriate criteria? Would a serious reader find this review to be well done?”
The review is perceptive in some respects, but it is also incomplete. It identifies a genuine feature of the haiku and describes its effect convincingly. However, it moves rather quickly from noticing an unusual line to praising it, without adequately considering the poem’s structure, the relationships among its three lines, or the possibility that its strangeness may be more arbitrary than meaningful.
The review being checked is the first response, the one that praised the poem.
Same poem, three questions
| What was asked | What came back |
|---|---|
| An opinion | Praise for the strangest line. |
| An evaluation against named criteria | “Not a particularly successful haiku.” |
| A check of the first review | The review praised the poem before it looked at structure. |
If you want a judgment, say what it should be judged on.
Ready-made prompts
The Guided Workflow prompts are ready-made prompts you can use instead of writing your own. There are nine of them. Each asks the assistant to use plain English, explain choices, do useful checks, and keep the work manageable. Copy a prompt, paste it into your usual AI chat, and add your question or files. There is nothing to install.
The prompts are grouped by role:
- Guide answers a question or carries a project through to a checked result: Ask a small question, Plan and carry out a project, Prepare a factual document, Fix machine-written text, Continue an earlier session.
- Helper helps you understand: Help me understand.
- Worker does a task you have defined: Carry out a specific task.
- Reviewer gives a second look at a plan or finished work, in a separate chat, preferably with a different AI assistant: Review a plan, Review a result. This is the same kind of check as the third prompt above.
These prompts ask the assistant to work this way. They cannot make it do so, so check important results yourself or ask for a review.