You are trapped in the AI productivity paradox.
And the weird part is you feel like you’re winning.
You prompt.
AI spits out text in 6 seconds.
You get that little dopamine hit.
Then you read it and think…
Not quite.
So you prompt again.
And again.
“You’ll know it when you see it.”
That’s not a workflow.
That’s a slot machine.
You’re waiting for an outcome you can’t define.
If your thinking is fuzzy, your draft will stay generic.
AI isn’t confused.
Your thinking is.
In an RCT of 16 software developers, they expected AI to speed them up.
Even after using it, they still believed they were faster. But objectively, tasks took longer (19%).
A perception gap that big should make every researcher pause.
Here’s the fix:
Backwards design is key to break this paradox.
Most people start here:
“AI, write my Discussion.”
That’s the lazy approach.
Backwards design flips it.
It forces you to stop “vibing” with the model.
And start acting like the showrunner.
You set the vision.
AI becomes your production crew.
The “Design Backwards” framework in commonly used education and curriculum design:
Start with the outcome, define what counts as evidence, then plan the activities.
Here’s how we can apply it to research when working with AI:
1) Desired Results
Define what “good” looks like before you start working with AI.
Not “900 words.”
Not “make it sound academic.”
Instead, define the job of the Discussion:
- Defend 2–3 key findings
- Explain how they close a specific gap
- Acknowledge study-specific limitations
- Land real implications of your results
If you can’t define “good,” AI will default to generic.
Because at its core, an LLM is optimizing for “likely next words,” not “your dataset’s truth.”
And without your constraints, it will drift toward the safe middle. The average. The bland. The thing that sounds right to everyone and commits to nothing.
2) Determine Evidence
How will you know the draft is good?
This is the step most people skip.
They don’t set a standard.
So they iterate forever.
Instead, build a simple pass/fail checklist.
Not vibes.
Evidence.
Create pass/fail evidence:
- At least 3 studies that support and refute each key finding
- Every major claim maps to one of the 2–3 findings
- The core message is one sentence and repeats (on purpose)
- How the key finding fills the gap is explicit
- Implications are actionable (clinical, research, policy) and sound like you
This is what turns “I’ll know it when I see it” into “I’ll know it because it passed the rubric.”
And it protects you from the most common AI failure mode:
Confident filler (popularly referred as “AI slop”).
3) The Plan
Now design the inputs and workflow that get you there.
Context ≠ a paste-bin.
Notes. Results. Random paragraphs. “Stuff I might cite later.”
That’s static.
AI is a mirror.
It won’t magically create structure.
It reveals the lack of it.
So give it structure.
Before prompting, fill these 5 manuscript pillars:
- Key findings (2–3):
- Core message (1 sentence + Yet/However tension):
- Research gap (knowledge, thinking, or practice):
- Research problem (your long-term mission, not the topic):
- Implications (clinical / research / policy):
Strong pillars in → crisp draft out.
One more thing that matters here.
Control the context.
You meander, the model meanders.
You argue for 30 turns, and now it’s stuck in a weird frame you accidentally created.
Sometimes the best move is a fresh thread with a tight prompt and only the pillars it needs.
Less memory.
More precision.
That’s backwards design in practice.
You’re not “chatting.”
You’re directing.
You’re managing.
You’re building a draft on purpose.
💬 If your AI drafts feel generic, which step are you skipping most:
Desired Results, Evidence, or Planning?
PROMPT OF THE WEEK
Improve Clarity and Readability
Prompt: You are an expert editor specializing in clear, concise, and easy-to-read writing.
Your task is to edit the text below to maximize clarity, logical flow, and readability while preserving the original meaning and intent.
Guidelines:
Remove unnecessary words, filler, and repetition
Tighten sentences without oversimplifying ideas
Improve logical flow and transitions between sentences
Break up long or complex sentences where needed
Replace vague or confusing phrasing with clear, concrete language
Maintain the original tone and voice
Do not add new ideas or opinions
[Paste text here. If not pasted ask the user for the text.]
P.S. Research Boost walks you through the manuscript pillars step by step with your desired result in mind, so the draft comes from your work, not AI guesswork. Researcher-first, always.
Sign up for a free trial here: http://researchboost.com/