The Authorship Paradox: AI Writes Better Goals — but You Don't Follow Through

❌ DELEGATION (kills motivation)
  You → write a reflection → AI phrases the goals → you get a finished list
  Result: goals are SMART-correct but psychologically foreign → 47% follow-through

✅ KEEPING AUTHORSHIP (preserves motivation)
  You → write a draft of your goals YOURSELF → AI improves the wording → you recognize your goals
  Result: goals get sharper but stay yours → ~73% follow-through

Both approaches happen in one chat, one conversation. The only difference is who writes the first draft.

Task: It's the end of the year and time to set goals for the next one. The usual temptation is to tell the AI "set my goals based on my reflection" and get a pretty list.

Prompt (the right approach — you first, then AI):

I've written three goals for next year. Help me make them sharper and more specific — 
add measurability, deadlines, success criteria. But keep my wording 
and my meaning — that matters. Don't rewrite from scratch; improve what's there.

My goals:
1. Launch a YouTube channel about personal finance
2. Start delegating day-to-day operations at the company
3. Get healthier — sit less, move more

For each goal, show: my version → your improved version → what you changed and why.

Result: The model shows three pairs: your original wording and an improved version with added specifics. You see the difference — and decide for yourself whether to accept the edit. The improved version stays your goal with sharpened details, not someone else's text you're supposed to accept.

LLMs are great at one thing — turning vague into specific. You say "I want to get healthier" → you get "3 times a week for 30 minutes, report on first results by March 1." Technically better. But that clarity is worth nothing if you don't feel you chose it yourself.

Psychological ownership isn't a metaphor or some fuzzy "feeling." It's a mechanism that moderates every step: whether you remember the goal, whether you come back to it, whether you start acting. In the study it fully explained the difference in outcomes — the objective quality of the goal explained nothing. No magic: the brain defends and pursues more strongly what it has invested its own effort in. It's the same IKEA effect — the bookcase you assembled yourself feels more valuable than one bought ready-made.

The most dangerous trap is asking AI to write your goals precisely when you're struggling to put them into words. The study showed that people with less faith in themselves (low self-efficacy) lose the most ownership under AI authorship. In other words, the people who find it hardest to articulate goals are hurt most by full delegation.

Control levers: - The phrase "improve it, but keep my meaning" — keeps the AI in the role of editor, not author - The format "my version → your version → what changed" — keeps you in the decision loop - Ask the AI to ask clarifying questions instead of handing you finished wording — you reach the answer yourself; the AI only guides

I have a {draft / idea / sketch} on {goal or task topic}.
Here's what I wrote myself:

{your text}

Your job is to improve it, not rewrite it. Specifically:
— Add {what to improve: deadlines / metrics / concrete steps / success criteria}
— Keep my format and my words wherever possible
— Show: my version → your edit → an explanation of what you changed and why

If anything is unclear, ask a clarifying question first — don't guess.

What to fill in: - {draft / idea / sketch} — how raw your material is - {topic} — a goal, a plan, a mission, an About Me text, a strategy - {what to improve} — exactly what you want to tighten up: measurability, deadlines, concreteness of steps

🚀 Quick start — paste into the chat:

I want to set goals for {period}, but I want to stay their author — 
not delegate to AI, but improve my own drafts. 
Help me set up this way of working for my situation: {brief description}.

[paste the template above]

The LLM will ask what exactly you've already written and what you want to tighten up — because without your draft it can't work in editor mode instead of author mode.

⚠️ Only for identity-relevant tasks: The principle applies where the result isn't just an artifact but a guide to action: goals, plans, values, strategy. For "write an email to a client" or "make a summary," full delegation does no harm — there's no psychological ownership involved.

⚠️ The study covered goals, not all AI assistance: The results are strong for personal goals. Whether the same holds for career decisions, business strategy or personal values is a separate question. The principle looks transferable, but there's no direct data.

⚠️ Reflection doesn't save you: Participants in the LLM group wrote the same reflection as the self-authoring group — the only difference was who then wrote the final text. So "I thought it through myself but let AI write it" doesn't protect against losing ownership.

The team recruited 470 Americans and split them into two groups. Everyone wrote the same reflection on the past year. Then came the fork: some wrote three personal goals themselves, others received goals written by GPT-4 based on their reflection. Afterward, each person rated their goals on several scales: how important they are, how committed they feel, how much they feel the goals are theirs. Two weeks later came the follow-up: do you remember your goals, have you taken real steps?

The surprising part: AI goals were objectively better on SMART by a huge margin (d=2.26 — a very large effect). But on every motivational metric they were worse: ownership (d=1.38), commitment (d=1.19), importance (d=1.13). The researchers specifically tested what mediates behavior — it turned out psychological ownership explains everything, and the objective quality of the goal explains nothing. That's the key discovery: quality and motivation can completely diverge.

They also tested moderation by self-efficacy. People with low self-belief — exactly the ones who most often reach for AI help — lost the most sense of ownership. The trap: the person who's struggling uses a tool that weakens them further.

1. The principle extends beyond goals

The same logic applies to any identity-relevant text — text that's supposed to become a guide to action, not just an artifact:

💡 Adaptation for personal branding / positioning:

When you're putting into words "who you are and what you offer," don't let AI write it from scratch. Write it yourself as best you can, and let AI improve the wording. The result stays yours — and you'll actually use it.

💡 Adaptation for career decisions:

"What kind of job should I look for, what matters in my next move" — answer it yourself on paper first, then ask AI to structure it and check the logic. Not the other way around.

2. Socratic mode instead of Author mode

Instead of asking AI to write, ask it to ask the right questions:

🔧 Technique: AI as a coach, not an author

Copy

I want to set goals for next quarter. 
Don't phrase the goals for me. Ask me questions one at a time — 
the kind that lead me to a clear wording on my own.
After each of my answers — either the next question 
or a follow-up if my answer is vague.
Once a goal is phrased, offer a SMART check.

The effect: you arrive at the wording yourself through dialogue → maximum ownership → motivation preserved. The AI did the work, but you're the author.

Optimized but Unowned: How AI-Authored Goals Undermine the Motivation They Are Meant to Drive Preprint, May 2026 | osf.io/cajg2

Authors: Vivienne Bihe Chi, Lyle Ungar, Sharath Chandra Guntuku — University of Pennsylvania; Roman Rietsche — Bern University of Applied Sciences; Andreas Göldi — University of St. Gallen

Related work cited in the study: - Pierce et al. (2001, 2003) — psychological ownership theory - Schimpf et al. (2026a) — an AI chatbot for goal setting (quality goes up, commitment doesn't) - Draxler et al. (2024) — AI ghostwriting reduces the sense of authorship over text