If your AI isn’t delivering gold yet, it’s not lazy. It just doesn’t know what your version of gold looks like.
We humans are great at giving feedback – to interns, employees, even to baristas (“uh, that’s two pumps, not three”). But somehow, when it comes to AI, we go quiet. We just shrug and accept whatever it gives us.
That’s a mistake.
→ AI doesn’t learn through intuition. It learns through iteration. The more you guide it on tone, structure, accuracy, and style, the closer it gets to reflecting your voice and your vision.
And, according to McKinsey, with 71% of organizations now using generative AI regularly, up from 65% from just a year ago, the question isn’t whether we’ll use AI, it’s whether we’ll use it well.
So, if your prompts are clear but your results are meh, it’s time to stop commanding and start coaching.
Why Giving Feedback to AI Is Non-Negotiable
One of the biggest myths about AI is that once you “get good at prompting,” the magic just happens.
Yeah, no. It doesn’t.
Tools like ChatGPT or Claude aren’t mind readers – they’re pattern finders. Skip feedback and your AI will produce what’s probable, not what’s powerful.
That’s how you end up with content that’s perfectly grammatical and emotionally beige. Technically right, but deeply unmemorable – like getting a birthday card that just says “Happy Birthday” with no signature. Sure, it’s correct. But did it matter?
Feedback is calibration. It’s you saying, “Closer, but not quite,” until the output feels right. Like adjusting seasoning in a recipe. Too salty? Too safe? Tell it so.
As Atlassian put it in their AI Collaboration Report 2024, “Mindset matters far more than adoption.”
→ In other words, the real advantage isn’t in the tool; it’s in how you use it. Feedback is where that difference lives.
(And if your feedback style involves a little sighing, eye-rolling, and caffeine-fueled muttering, then welcome to the club.)
The AI Intern Analogy: Train, Don’t Just Tell
Picture this. You hire a new intern. On their first day, you hand them a project:
“Hi ChatGPT. Write me a report about our customers.”
No context. No examples. No feedback.
A week later, they proudly deliver five pages of demographic data when what you really needed was customer sentiment.
Whose fault is that? 👀
Exactly.
Same story with AI. When you say “Write a blog intro” and it gives you something generic, don’t roll your eyes and start over – coach it.
Try this:
“This feels too generic. Make it sound like a confident keynote speaker – short sentences, friendly tone, conversational, a little cheeky.” (Sound like anyone you know? LOL!)
→ Give feedback like that and you’ll watch your AI intern level up in real time.
Five Ways to Give Constructive Feedback to AI
Here’s where I see most people go wrong – they treat AI like a vending machine instead of a team member. Type a request, hit enter, expect brilliance.
But the good stuff happens in the back-and-forth.
Think of these five approaches as your new performance review for your digital teammate.
1. FRAMEWORK FIRST
→ Instead of “make it better,” give it a structure.
This provides Instant focus and less guesswork. Your AI isn’t a mind reader, but it’s great at following a map.
“Apply the PREPARED Prompting Framework to this email. Add more context and examples.”
(Note: Make sure you’ve trained your AI on the PREPARED structure so it knows what you’re talking about.)
2. PERSPECTIVE SWITCH
→ Tell the AI who it is.
That simple shift changes tone, detail, and intent – because context is magic.
For example, if you wanted to shed a few pounds (I think you look amazing, by the way…I wouldn’t change a thing!) and you asked a dietician and a fitness trainer for ideas, you’d get totally different responses.
“Act as my head of marketing. What would you change to make this pop?”
3. THE OPPOSITE GAME
→ Ask for the wrong version first, then fix it together.
The contrast helps your AI understand what “good” looks like in your world. Plus, bad versions are often hilariously entertaining — I once had mine open with ‘Greetings, human consumer!’ Not ideal for a luxury brand.
“Write the worst possible cold email sequence.”
4. STEP BUILDER
→ Go layer by layer.
This slows the AI down, which ironically helps it think faster and better. It’s the same logic behind the 20-60-20 Framework: 20% strategy, 60% execution, 20% refinement.
“List three industries dealing with this challenge. Pick one. Identify its top three pain points. Now write five solutions.”
5. IDEA IMPROVER
→ Treat every draft as Version One.
Iteration isn’t rework; it’s how creativity compounds. And you don’t have to just take my word for it.
A 2025 peer-review study (ICLR) found 27% of reviewers improved their work after receiving AI feedback, and their revisions were more detailed and constructive. Even experts need a coach, and in my experience, feedback from AI can often feel less “emotionally loaded” than from a colleague you have either admiration for or frustration with.
“Give me three versions: basic, better, best.”
What Happens When You Don’t Give Feedback
Quite simply…no feedback = no growth.
Your AI turns into a glorified autocomplete. It will be competent, consistent, and completely forgettable.
And it shows up in the numbers. Only 26% of companies say they can turn AI use into measurable value. The rest? Stuck in pilot mode, drowning in “good enough.”
Worse, sloppy feedback can actually cause harm. Nature Human Behaviour (2025) found human-AI feedback loops can amplify bias when people give vague or one-sided input.
→ Takeaway: Be clear. Be balanced. Don’t abdicate your judgment. Your AI learns from you.
“This is no time to outsource your thinking!”
Feedback Loops That Work (and the 20-60-20 Model)
My 20-60-20 Model breaks collaboration down like this:
- 20% – You set the goalposts: strategy and prompting.
- 60% – AI does the heavy lifting: drafting, building, brainstorming.
- 20% – You review and refine: that’s where feedback lives.
→ That last 20% isn’t optional. It’s your superpower. It’s where “useful” becomes “unforgettable.”
Each time you feed insight back in (“Closer, but make it bolder,” or “Align this with our UPGRADE Framework”), you’re not just fixing one draft; you’re training a collaborator.
And here’s a fun twist: people trust co-created feedback – human + AI together – more than purely human input in some contexts (According to this study from University of Glasgow, 2025).
Turns out teamwork really does make the dream work – even with your digital coworkers.
Just beware! That same study points out that the students trusted the feedback less when they realized AI was collaborating with a human.
→ You have to do your last 20%to ensure that no matter how something is created, it is ultimately owned and operated by a human!
“Don’t just prompt harder. Coach smarter.”
Final Thought
Your AI’s waiting for direction.
→ It doesn’t need perfection. It needs partnership. With YOU!
Coach it like you would a promising intern: specific, kind, consistent. The more feedback you give, the sharper it gets and the sharper you become as a communicator.
If your team’s ready to move from prompts to partnership, check out my article on the importance of Show & Fail . . . or better yet, let’s talk about a talk.
Frequently Asked Questions
Yes. Most generative AI tools adapt during a chat session. When you clarify tone, format, or intent, the model uses that context to refine future outputs. You’re not training it forever — just guiding it in the conversation window.
Be specific. Instead of “make this better,” say “add real-world examples and keep it conversational.” Treat AI like an intern who needs coaching, not a magician who reads your mind.
Feedback bridges the gap between generic and personal. It helps AI understand your goals, audience, and tone so you get results that don’t just sound more human, they sound more you.
You can overwhelm it in one message, but steady, layered feedback works best. Think of it as coaching in rounds: clarify, review, refine.
You’ll get output that’s fine but lackluster. Without feedback, AI defaults to safe, middle-of-the-road responses that sound like everyone else.
It’s Julie Holmes’ approach to AI collaboration: treat AI like a smart, eager intern. Give context, expectations, and feedback so it can deliver work that meets (and hopefully exceeds) your standards.
Spend 20% on clear prompting, let AI do 60% of the work, and use the last 20% for thoughtful review and feedback. That final 20% turns “useful” into “unforgettable.”
Ask it to match tone, add emotion, or use analogies. For example: “Explain this like a friendly colleague chatting over coffee.” Tone coaching matters as much as task coaching.
Yes — if your feedback is balanced. Specific, contextual feedback helps AI avoid assumptions or stereotypes by clarifying what matters and what doesn’t.
Every time you iterate. After each response, highlight what worked and what missed. Feedback isn’t a final step — it’s the heartbeat of collaboration.