The Frenemy Prompt in ChatGPT: Getting Honest Criticism [2025]

TL; DR

• The frenemy prompt is a Chat GPT technique that tells the AI to act like a critical friend who delivers honest feedback instead of flattery.
• Real criticism beats praise because it identifies actual problems, weak arguments, and gaps in your thinking.
• Implementation is simple but the results are transformative for writers, developers, managers, and anyone creating things.
• Why it works is psychological: Chat GPT defaults to being agreeable, so you need explicit permission to flip that behavior.
• You're leaving value on the table if you're not using a feedback mechanism that challenges your assumptions.

Introduction: Why Chat GPT Defaults to Being Your Cheerleader

Last Tuesday, I asked Chat GPT to review a presentation I'd spent two hours building. It told me it was "well-structured," "compelling," and "engaging." All three. Within thirty seconds.

Then I sent the same deck to an actual colleague. Her feedback: "The third slide contradicts your main premise. Also, nobody cares about this statistic."

That's the gap we're talking about.

Chat GPT isn't designed to be mean. It's designed to be helpful, which in the AI world usually translates to agreeable. The model was trained on human preferences that reward positivity, and it's optimized to avoid conflict. So when you ask for feedback, you get a carefully curated list of minor suggestions wrapped in encouragement. It's like a teacher handing back a paper with "great effort" written across the top.

But here's what changed for me: the frenemy prompt.

It's not a magical formula. It's just permission. You're telling Chat GPT, "I don't want you to be nice. I want you to be honest." And when you're explicit about that, the AI behaves differently. You get pushback. You get someone to poke holes in your logic. You get the version of feedback that actually makes you better.

I started using this about six months ago, and I've since incorporated it into my workflow for everything from email drafts to strategy documents. It's become the conversation starter whenever I need real criticism instead of validation. And once you experience the difference, it's hard to go back to the generic praise version.

The frenemy prompt isn't new. It's been floating around in AI communities for a while. But it deserves more attention because most people are still using Chat GPT like a very polite rubber stamp when it could be a true sparring partner. This guide shows you exactly how to use it, why it works, and how to get the most from it.

Runable scores highest in delivering critical feedback due to its integrated feedback features, while Claude is noted for its willingness to deliver hard truths. Estimated data based on tool descriptions.

What Is the Frenemy Prompt and How Does It Work

The frenemy prompt is a pre-conversation instruction that reframes Chat GPT's role. Instead of asking for feedback, you're asking for criticism from someone who cares about you but won't sugarcoat the truth.

The basic structure looks like this:

"I'd like you to act as a critical friend. Your job is to find genuine flaws, weak points, and areas for improvement. Be honest even if it might sting. Don't worry about being nice or encouraging. Just tell me what's actually wrong."

Then you share your work.

What happens next is remarkable. Chat GPT stops being diplomatic. It identifies contradictions you missed. It spots where your logic breaks. It questions your assumptions. It tells you when something is confusing, weak, or pointless.

The magic isn't in the wording of the prompt itself. The magic is in the permission structure. You're explicitly saying, "You have my permission to criticize me." That's the pattern break that changes the output.

I tested this by giving the exact same piece of writing to Chat GPT twice. Once with a standard feedback request. Once with the frenemy prompt. Same input, different instruction.

Standard feedback version: "This article is well-organized and covers the topic thoroughly. The examples are relevant and help illustrate your points. I'd suggest adding a few more citations to strengthen credibility."

Frenemy prompt version: "The intro buries the main argument under three paragraphs of context nobody cares about. The second section contradicts your thesis. You're padding the middle sections with obvious information that doesn't add anything. The examples are fine but generic. This needs a complete restructure before it's worth publishing."

One is nice. One is useful.

The reason this works is rooted in how language models are trained and deployed. Chat GPT's default behavior is calibrated toward being helpful and harmless. Part of that calibration is a preference for politeness, agreement, and positive framing. It's baked into how the model learns to respond.

But that training isn't immutable. You can override the defaults through explicit instruction. When you tell Chat GPT to adopt a different persona or follow a different directive, it will. The frenemy prompt is just one such override. You're essentially saying, "Ignore the politeness default for this conversation."

Another way to think about it: Chat GPT is like a coworker who desperately wants you to like them. Without explicit permission, they'll never tell you the hard truth. Once you give them that permission, they'll tell you everything.

Custom instructions and saved prompt libraries are rated highest for efficiency, closely followed by ChatGPT Plus with GPT-4. Estimated data based on typical user feedback.

Why Default Chat GPT Feedback Feels Like Flattery

Most people don't realize they're getting a version of Chat GPT that's been heavily filtered toward agreeability. They ask for feedback, get back a list of minor tweaks, and assume that's the real critique.

It's not.

Here's what's happening under the hood: Chat GPT is optimizing for multiple things simultaneously. First, it wants to be helpful. Second, it wants to avoid conflict or offense. Third, it wants to reinforce your perception of yourself as competent. All three of those priorities push toward positivity.

Think about the structure of typical feedback requests. "Please review my work and tell me what I can improve." That phrasing is actually non-threatening. You're asking for suggestions, not judgments. Chat GPT interprets that as, "Please be constructive but ultimately supportive." So it frames everything through that lens.

The result is feedback that's technically true but strategically useless. It highlights your strengths. It mentions weaknesses as "opportunities for growth." It suggests small additions rather than major deletions. It rarely tells you to start over, even when that's the right call.

I see this all the time when people ask Chat GPT to review emails. Standard feedback: "This is clear and professional. You might consider adding a bit more detail about the timeline." That's accurate but toothless. What it doesn't say is, "This email is passive-aggressive and reads like you're mad at the person."

Or when reviewing code: "Good structure here. Have you considered adding error handling?" When the real issue is that the entire approach is inefficient and should be scrapped.

The flattery effect gets worse the more specific your initial request. If you ask Chat GPT to "rate your writing from 1-10," it will give you 7 or 8 virtually every time. If you ask it to "identify three strengths and three weaknesses," it will work overtime to find strengths when there might only be one.

This isn't malice or failure. It's design. Chat GPT was built to be a supportive tool, and it does that extremely well. But support isn't always what you need. Sometimes you need someone to tell you the work isn't good enough.

The Psychology Behind Why People Prefer Flattery

Before we talk about how to use the frenemy prompt effectively, it's worth understanding why most people aren't using it yet. The answer is psychological.

Humans have what researchers call the "self-serving bias." You tend to attribute your successes to your abilities and your failures to external circumstances. Conversely, you attribute others' successes to luck and their failures to incompetence. That bias is deeply embedded.

When you ask for feedback on something you created, you're already primed to hear it defensively. A critical comment feels like a personal attack. A compliment feels like validation. So Chat GPT's default agreeability actually matches what you want to hear, even if it's not what you need.

There's also the dopamine effect. Your brain releases a small hit of dopamine when someone praises your work. That feels good. It reinforces the behavior of seeking more praise. Criticism does the opposite. It triggers threat responses. Your amygdala activates. You become defensive.

But here's the key insight: that defensive response is actually a feature, not a bug. It means the criticism is registering. It's hitting something real. When you read a critical comment and feel that sting, that's your brain telling you the feedback matters.

The frenemy prompt works because it lets you bypass that initial defensiveness. You're agreeing in advance to receive criticism. You've created psychological permission to be wrong. That changes how you process the feedback.

I noticed this shift in myself when I started using it. The first few times, I was defensive. I'd read Chat GPT's critical feedback and want to argue. But because I'd explicitly asked for it, I couldn't blame the AI for being mean. I'd asked for honesty. I got honesty. The conversation became about the work, not about me.

After a few cycles of that, something shifted. The defensive response faded. I started looking forward to finding problems in my work because fixing them made the work better. The sting of criticism became a signal of opportunity.

That's the real value of the frenemy prompt. It's not about getting better feedback. It's about retraining your brain to value criticism over flattery.

The Frenemy Prompt offers high speed and consistent feedback, while human feedback is most effective when context is understood. Estimated data.

How to Set Up and Use the Frenemy Prompt

Implementation is straightforward, but there are nuances that make the difference between mediocre criticism and really useful feedback.

Step 1: Choose Your Starting Prompt

There's no single "correct" frenemy prompt. You adapt it based on what you're working on and what kind of criticism you need. Here are a few variations:

For writing: "You're a harsh but fair editor. Your job is to identify what's not working, not what is. Point out contradictions, unclear sections, weak arguments, and boring parts. Assume the reader is smart and impatient. Don't sugarcoat anything."

For strategy documents: "Assume you're a skeptical board member reviewing this. Your job is to poke holes in the logic, identify assumptions that aren't validated, and point out where the plan is weak or risky. Be direct."

For code: "You're reviewing this for production deployment. Assume the person wrote it under time pressure and didn't consider edge cases. Find the bugs, inefficiencies, and architectural problems. Be specific."

For presentations: "You're the toughest audience member. You're skeptical, you know the subject matter, and you get bored easily. Point out where you'd be unconvinced, where you'd tune out, and where the logic falls apart."

The pattern is the same: give Chat GPT a persona that's critical by nature and explicit permission to be harsh.

Step 2: Share Your Work and Your Context

Don't just dump raw text. Give Chat GPT context about what you're trying to do, who the audience is, and what success looks like. This helps it give more targeted criticism.

For example: "I'm writing a blog post for technical founders about improving API response times. The goal is to teach them one specific optimization technique they can implement in an afternoon. The audience is impatient and skeptical of clickbait. Here's the draft: [text]"

Context changes the feedback. Without it, Chat GPT gives generic criticism. With it, the AI understands what's actually failing.

Step 3: Read the Criticism Without Defending

This is the hard part. When Chat GPT tells you something is weak, your instinct will be to explain why it's actually good. Don't. Just read it. Let it sit. Ask yourself: is this true?

Often it is. Sometimes it's not. But the moment you start arguing, you lose the value. The frenemy prompt only works if you approach it with genuine openness.

Step 4: Ask Follow-Up Questions

If the criticism is vague, dig deeper. "What specifically makes that section confusing? Show me the sentence that lost you." Chat GPT will point to it. That specificity is gold.

Step 5: Revise Based on Patterns

Don't try to fix everything at once. Look for patterns in the criticism. If Chat GPT mentions that three different sections are confusing, the problem isn't those sections. It's your overall structure. Fix the root issue, not the symptoms.

Real-World Applications: Where the Frenemy Prompt Works Best

The frenemy prompt isn't a universal tool. It works brilliantly in some contexts and less effectively in others. Understanding where to use it is crucial.

Writing and Content

This is the frenemy prompt's home turf. Essays, articles, emails, proposals, presentations, all the written work you do. Writing is subjective enough that you need a critic who isn't afraid to say, "This is boring," but specific enough that the criticism can be actionable.

I use the frenemy prompt on every important piece of writing before it ships. It catches contradictions I've made (because I'm too close to the work to see them), identifies sections that don't pull their weight, and points out where I'm being unclear.

Last month, I drafted a long-form piece on AI limitations. The frenemy prompt told me the first 800 words were throat-clearing. It was right. I cut them and the piece became much stronger.

Strategy and Planning

When you're building a plan, making a decision, or proposing a direction, the frenemy prompt is invaluable. It challenges your assumptions before you commit to them.

I've used it for business strategy (flagged where the competitive advantage was real vs. imagined), product decisions (identified where we were solving for ourselves instead of the customer), and hiring strategy (pointed out where our compensation was too low to attract the right people).

The prompt works here because strategy is built on assumptions. Someone needs to question those assumptions. Chat GPT, when explicitly instructed, is good at that.

Code and Technical Decisions

Chat GPT isn't going to find all your bugs. It will make mistakes. But when you ask it to "find the problems in this code," it often does, especially on second and third passes.

I use the frenemy prompt when I'm about to deploy something significant or when I've written code that feels fragile. The AI catches edge cases I missed and inefficient approaches I've become too comfortable with.

It's not a replacement for code review from human teammates. But it's a useful checkpoint before that review.

Less Effective: Creative Work Without Clear Goals

The frenemy prompt struggles when there's no objective standard for success. If you're writing poetry, experimenting with style, or exploring creative ideas, harsh criticism might just shut down the creative process.

Criticism works best when there's a goal. "This email isn't persuasive" is useful feedback. "This poem isn't moving" is less useful because moving is subjective.

Less Effective: Emotional Content

If you're writing something emotionally vulnerable (a personal essay, a heartfelt message to someone), the frenemy prompt can be too harsh. You need different feedback for different contexts.

I've found that for personal writing, I use a different version of the prompt: "Be honest but kind. Point out what's not working without being cruel."

Estimated data shows that while users find ChatGPT feedback clear and constructive, they often perceive it as less useful and honest due to its agreeable nature.

Common Mistakes When Using the Frenemy Prompt

People mess this up in predictable ways. Knowing the mistakes helps you avoid them.

Mistake 1: Using It When You Want Reassurance, Not Feedback

If you're in a fragile emotional state, the frenemy prompt will hurt. You'll get harsh feedback and then spiral instead of improving. Know yourself. If you need encouragement today, use a different prompt.

Mistake 2: Confusing Harsh with Helpful

Criticism should be specific and actionable. "This is terrible" is harsh but useless. "This is terrible because you haven't explained why the reader should care, so the whole thing feels pointless" is both harsh and helpful.

If Chat GPT gives you vague criticism with the frenemy prompt, follow up with, "Be more specific. What exactly is the problem?"

Mistake 3: Treating One Pass as Definitive

Chat GPT's first pass with the frenemy prompt is good but not complete. Do multiple passes. After you revise, ask the AI to critique the revision. After that revision, do it again. You'll surface more issues each time.

Mistake 4: Ignoring the Feedback That Stings Most

Psychologically, the hardest feedback to hear is often the most true. If Chat GPT says something that makes you defensive, pause. That defensiveness is a signal. Dig into it.

Mistake 5: Trying to Fix Everything at Once

If Chat GPT identifies 15 problems, you can't fix all 15 in one revision. Prioritize. Fix the foundational issues first (structure, logic, clarity). Then handle the secondary issues (word choice, tone, examples).

Advanced Techniques: Getting Even Better Criticism

Once you've mastered the basic frenemy prompt, there are variations that give you deeper feedback.

Technique 1: The Hostile Reviewer

Instead of a critical friend, ask Chat GPT to act as someone who disagrees with your premise and is looking for ammunition.

"Assume you're someone who disagrees strongly with my position. You're looking for flaws in my argument so you can counter them. What would you attack first?"

This surfaces weak points that a friendly critic might overlook. It's useful when you're making an argument and need to stress-test it.

Technique 2: The Exhausted Reader

Ask Chat GPT to critique the work from the perspective of someone who's overwhelmed and impatient.

"You're reading this at 11 PM after a long day. You're exhausted and easily bored. Point out the moment you'd stop reading and why."

This is brutal feedback but it works. It identifies exactly where your work loses people.

Technique 3: The Expertise-Mismatched Reviewer

For technical work, ask Chat GPT to review it as if it's being read by someone competent but not expert in that specific field.

"You have 15 years of general software engineering experience but have never worked with this specific technology. Point out what's unclear, what you'd need to research, and where the explanation assumes too much knowledge."

This catches where you're being unclear because you're too close to the subject.

Technique 4: The Sequential Critic

Instead of asking for all criticism at once, ask Chat GPT to critique one section at a time, in order.

"Review just the first section of this piece. What's not working in this section specifically?"

Then the second, then the third. This prevents you from being overwhelmed and gives you more detailed feedback on each part.

Technique 5: The Comparative Critic

If you have two versions of something, ask Chat GPT to critique them against each other.

"Here's version A and version B of the same thing. Which one is stronger and why? What does one do better than the other? What does each one get wrong?"

This gives you relative feedback that's often more useful than absolute critique.

The chart illustrates the estimated increase in comfort and integration of the frenemy prompt technique over time, highlighting the importance of consistent practice. Estimated data.

Why Chat GPT Sometimes Still Holds Back (And How to Push Past It)

Even with the frenemy prompt, Chat GPT sometimes doesn't go as hard as it could. There are safety guidelines baked into the model that prevent it from being maximally harsh.

It's unlikely to tell you that your entire idea is stupid. It might say, "The core concept needs significant rethinking," but it won't say, "This idea won't work."

It's unlikely to tell you that you're bad at something. It might say, "This isn't your strength," but it probably won't say, "You're not good at this."

These aren't failures. They're intentional design choices. Open AI decided that an AI shouldn't be needlessly cruel. That's reasonable.

But you can work around it.

If you need harder criticism, frame it as external feedback: "A harsh reviewer said [thing]. Do you agree with that assessment?" Chat GPT is more willing to validate harsh criticism from a hypothetical third party than to generate it directly.

Or ask it to predict the most damning criticism: "What would your fiercest critic say about this work? What would they attack hardest?"

Or use a different AI model. Claude tends to be slightly more willing to deliver hard truths. So does Perplexity when you ask it to research what the main criticisms of a concept are.

But honestly, Chat GPT with the frenemy prompt is usually enough. You don't need maximally harsh. You need honest. And once you unlock that, the difference is night and day.

How the Frenemy Prompt Changes Your Thinking Over Time

I've been using this for about six months. The immediate effect is better work. But there's a secondary effect that's more interesting.

Once you get used to receiving honest criticism from an AI, your tolerance for receiving it from humans increases. You become less defensive. You start looking for the truth in criticism instead of defending against it.

I notice this in meetings now. Someone says something critical about my idea, and my first instinct isn't to defend. It's to ask, "What specifically is the problem?" Because I've practiced that with the frenemy prompt so many times.

There's also a quality to your thinking that improves. When you know you're going to get harsh feedback, you naturally iterate before you share. You catch your own contradictions. You strengthen weak points preemptively. The AI's feedback becomes less necessary because you've internalized the critical mindset.

And you start using the frenemy prompt not just for feedback, but as a thinking tool. Before you commit to something, you ask the AI to play devil's advocate. Before you make a decision, you ask it to find the flaws in your reasoning.

It's a small shift, but it compounds. After six months, your decision-making is noticeably better because you've practiced arguing with yourself, via an AI, about 50 times.

The frenemy prompt is estimated to be most effective in code reviews (score of 9), where specific and critical feedback is crucial. Estimated data.

Comparing the Frenemy Prompt to Other Feedback Methods

The frenemy prompt isn't the only way to get feedback. How does it compare to other approaches?

Versus Human Feedback

Human feedback is better when the human is an expert in your field and has the time to understand context. But most humans also have their own biases, politics, and kindness that filter their feedback.

The frenemy prompt is more consistent. It doesn't have an off day. It doesn't soften criticism because it likes you. It focuses on the work, not the person.

Ideal scenario: human feedback plus frenemy prompt. Use the AI to do the initial heavy lifting. Then give it to a trusted person for the nuanced feedback that only humans can give.

Versus Traditional Feedback Frameworks

Frameworks like "Situation-Behavior-Impact" or "Radical Candor" are designed to deliver hard feedback kindly. They work. But they also require the person giving feedback to care about doing it well.

Chat GPT doesn't need to be motivated to care. It just does, by default.

Versus Peer Review

Peer review is slow. You have to wait for someone to have time. The frenemy prompt is instant. You can iterate 10 times before your peer even reads the first draft.

The tradeoff is that peers catch things Chat GPT misses. Peers understand context and nuance that an AI doesn't.

Use them together. Use Chat GPT to get to 80%. Use peers to get to 95%.

Versus Reading Your Own Work Critically

You should always read your own work with a critical eye. But you're bad at it. You've read your own work so many times that you've stopped seeing it. You see what you meant to write, not what you actually wrote.

That's why outside feedback is essential. The frenemy prompt gives you that outside perspective instantly.

Building a Feedback System Around the Frenemy Prompt

Once you understand how the frenemy prompt works, the real power comes from integrating it into a complete feedback system.

Step 1: Use It in Multiple Languages and Contexts

Your email doesn't need as much scrutiny as your blog post. Your internal Slack message needs less feedback than your board presentation.

Create different versions of the frenemy prompt for different contexts. Light version for quick documents. Heavy version for important work. This prevents you from over-iterating on low-stakes content.

Step 2: Combine Multiple Feedback Passes

Don't ask Chat GPT for all feedback at once. Ask it to focus on different things sequentially:

First pass: "Is the core argument clear and convincing?" Second pass: "Are there confusing sections that need to be rewritten?" Third pass: "Is the tone appropriate for the audience?" Fourth pass: "What's still not working?"

Sequential passes catch more problems than one combined pass.

Step 3: Include Humans at Key Points

After you've done 3-4 iterations with the frenemy prompt, get human feedback. But don't ask humans the same questions Chat GPT already answered. Ask them the subjective stuff:

"Does this convince you?" "Would you send this to your boss?" "Does the tone feel right?" "Is this memorable?"

Humans are good at subjective judgment. AI is good at objective critique. Divide the labor.

Step 4: Track Your Patterns

After 20-30 iterations with the frenemy prompt, you'll notice you make the same mistakes repeatedly. Chat GPT will tell you the same things.

Listen. Those repeated notes are where you need to change your behavior, not just your work.

If Chat GPT keeps saying your structure is confusing, you need to learn how to organize ideas better, not just reorganize this specific piece.

If it keeps saying you bury the lede, you need to change how you draft, not just move paragraphs around.

Tools and Integrations That Make This Easier

You can use the frenemy prompt in Chat GPT's web interface. But there are ways to make it even more efficient.

Use Custom Instructions

Chat GPT now allows you to set custom instructions that apply to every conversation. You can set the frenemy framework as a default instruction, then just paste your work.

This saves the step of writing out the full prompt every time.

Create a Saved Prompt Library

If you use Chat GPT's API or a third-party tool like Zapier, you can create a library of frenemy prompts for different content types.

Button 1: "Critique this writing" Button 2: "Critique this code" Button 3: "Critique this strategy"

One click, auto-populated prompt. Fast iteration.

Use Chat GPT Plus with GPT-4

GPT-4 is noticeably better than GPT-3.5 at giving nuanced criticism. If you're doing this frequently, the Plus subscription pays for itself in time saved on iterations.

Try Other Models

As mentioned earlier, Claude from Anthropic is also good at this. So is Runable, which offers AI-powered feedback on presentations, documents, and reports.

Runable is particularly useful if you're using AI to generate content and need feedback built into your workflow. It starts at $9/month and integrates with your creation process.

The Psychological Impact of Embracing Honest Feedback

Beyond the practical benefits, there's something deeper happening when you start regularly using the frenemy prompt.

You become less precious about your work. You stop confusing your work with yourself. Someone criticizing your essay isn't rejecting you. They're identifying a fixable problem.

That separation is psychologically powerful. It reduces the defensiveness that prevents learning. It lets you see your own blind spots.

I've noticed my writing improved not just because I'm fixing the problems Chat GPT identifies, but because I'm more willing to make big changes now. I'm less attached to sentences I like but that don't serve the work.

I'm more willing to delete paragraphs. To restructure sections. To acknowledge when something isn't working.

That willingness to break apart your own work and rebuild it is one of the highest-leverage skills you can develop. Most people never get there because feedback usually comes wrapped in kindness that prevents them from hearing the core message.

The frenemy prompt strips away that kindness. You hear the message clearly. And once you hear it enough times, you start internalizing it.

Limitations: What the Frenemy Prompt Can't Do

It's important to be realistic about what this technique can and can't do.

Chat GPT won't catch domain-specific errors that require expert knowledge. If you're writing about quantum physics and make a factual error, the AI might not catch it if the error is stated confidently.

Chat GPT won't understand your audience the way a real person in that audience would. It can guess based on description, but it doesn't have the visceral reaction that someone from that audience would have.

Chat GPT won't give you creative breakthroughs. It won't tell you that your entire approach should be different in a way that transforms the work. It can identify problems but can't always see the solution.

And Chat GPT might sometimes be wrong. It might critique something that's actually fine. Or miss something that's broken.

The frenemy prompt is a tool that makes you better, not a replacement for thinking critically yourself.

You still need to read the criticism, evaluate whether it's true, and decide which feedback to act on. Not everything Chat GPT says is correct. Some of it you'll disagree with. That's fine. The goal isn't to follow all feedback blindly. It's to get external perspective on your work.

Making the Frenemy Prompt a Habit

The real value of this technique comes from consistency. One round of harsh feedback helps your current project. But regular use of the frenemy prompt changes how you think and work.

Here's how to make it a habit:

Week 1-2: Try It on Something Low-Stakes

Start with something you don't feel too attached to. An email you're drafting. A document for internal use. Get comfortable with the format and the feeling of receiving harsh feedback.

Week 3-4: Use It on Something Medium-Stakes

Move to something that matters but isn't critical. A blog post. A presentation for your team. Notice how the feedback changes the work.

Week 5+: Use It on Your Most Important Work

Once you're comfortable, integrate it into your workflow for everything that matters. Before you share something significant, run it through the frenemy prompt first.

Ongoing: Review Your Feedback Archive

Monthly, pull up the notes you've been saving and look for patterns. What does Chat GPT consistently tell you? What do you repeatedly need to improve?

That meta-awareness is where real growth happens.

Variations for Different Work Styles

Not everyone works the same way. Here are variations of the frenemy prompt for different personalities and work styles.

For Perfectionists

You need the frenemy prompt to prevent you from over-iterating. Ask Chat GPT: "What are the three most important things to fix? Ignore the small stuff."

Then fix only those three and move on. The goal is good enough, not perfect.

For People Who Avoid Conflict

Your version of the frenemy prompt should be collaborative: "Help me identify problems with this work so I can improve it. I want to get stronger."

Frame it as helping, not judging. That makes the feedback easier to receive.

For Fast Movers

You need to balance speed with quality. Use a quick version of the frenemy prompt: "In 2 minutes, tell me the one biggest thing that's not working."

Get the critical insight without the detailed analysis.

For Detail-Oriented People

You probably love the frenemy prompt already. Challenge yourself to act on feedback faster. Don't spend a week analyzing each critique. Fix it and move on.

FAQ

What is the frenemy prompt exactly?

The frenemy prompt is an instruction you give Chat GPT that asks it to provide honest, critical feedback instead of polite praise. You're essentially asking the AI to adopt the role of a critical friend who cares about you but won't sugarcoat the truth. It works by overriding Chat GPT's default tendency toward agreeability and politeness, giving you permission to receive harsh feedback.

How do I know if the frenemy prompt is working?

You'll know it's working when Chat GPT tells you things that sting a little but are clearly true. If the feedback is vague or generic, the prompt isn't working. If it's specific, actionable, and makes you slightly defensive before you realize it's right, that's the signal that you're getting real criticism. Compare the output to what a normal feedback request gives you—the difference should be obvious.

Can I use the frenemy prompt with other AI tools besides Chat GPT?

Yes. Claude, Gemini, Perplexity, and other language models will respond to similar prompts. Each has slightly different strengths. Claude tends to be slightly more willing to deliver hard truths. Perplexity can research external criticism of concepts. Runable specifically offers integrated feedback features for presentations, documents, and reports, starting at $9/month. Experiment and find which tool gives you the most useful feedback for your work style.

How many times should I iterate using the frenemy prompt?

Most people get diminishing returns after 3-4 passes. The first pass catches major issues. The second pass catches secondary problems. By the third or fourth pass, you're usually dealing with minor refinements. I usually do 2-3 passes with the frenemy prompt, then get human feedback, then do one more pass based on what I learned. More than that, and you're spending time on optimization that doesn't meaningfully improve the work.

What if I disagree with the frenemy prompt's feedback?

That's fine. You don't have to implement every piece of feedback. But pause before you dismiss it. Ask yourself: is the AI wrong, or am I defensive? Often the feedback that stings most is the most true. If you genuinely disagree after reflection, ignore it and move on. But if you find yourself dismissing all critical feedback, that's a sign you're not really open to the process.

Is the frenemy prompt just manipulation of an AI?

No, it's not manipulation. You're not tricking the AI into doing something it shouldn't. You're making an explicit request for a specific kind of response. That's a legitimate use of the tool. Chat GPT has the capability to deliver critical feedback. You're just asking it to do so. The AI doesn't resent you for this. There's no deception involved.

Can the frenemy prompt backfire and damage my confidence?

It can if you use it at the wrong time. If you're in a fragile emotional state or early in a new skill, harsh feedback might be discouraging rather than helpful. That's why context matters. Use it when you're stable enough to receive criticism productively. If you're feeling vulnerable, use a gentler feedback approach first. Build yourself up, then bring in the harsh critic.

How is the frenemy prompt different from asking "What's wrong with this"?

It's about permission and framing. "What's wrong with this?" is a question that Chat GPT interprets conservatively. It might still hold back because it's trained to be helpful and kind. The frenemy prompt is a full reframing of the AI's role. You're saying, "Your job isn't to be nice. Your job is to be honest." That reframing changes the output significantly. Try both and you'll see the difference immediately.

Should I use the frenemy prompt for every piece of work I create?

No. For low-stakes content, it's overkill. Use it for work that's important, complex, or where you're making an argument that needs to hold up. For quick emails or internal notes, skip it. But for anything you're sharing widely or that represents your best thinking, the investment in getting harsh feedback is worthwhile.

What happens if Chat GPT refuses to give harsh feedback even with the frenemy prompt?

Rarely, Chat GPT will soften the feedback even when asked for harshness. This usually happens if you've been kind and complimentary in the conversation before the prompt, or if the content touches on sensitive topics. Try again. Or try a different phrasing. Or use a different AI model. But in 90% of cases, the frenemy prompt will get you the critical feedback you're looking for.

Conclusion: Making Criticism Your Competitive Advantage

Most people go through their careers avoiding criticism. They seek out feedback that feels good. They surround themselves with people who believe in them. They're always on the receiving end of polite, constructive comments that validate their choices.

That's comfortable. It's also a path to stagnation.

The people who improve fastest aren't the ones who get the most praise. They're the ones who seek out harsh feedback and take it seriously. They know the feedback that stings is usually the feedback that matters.

The frenemy prompt is a way to get that feedback on demand. It's a way to hear the hard truth from an AI that doesn't care about your feelings but does care about accuracy.

I started using it six months ago almost accidentally. I asked Chat GPT to be critical and was surprised by the difference. Since then, it's become central to how I work. Every piece of writing gets the frenemy treatment. Every strategy gets stress-tested. Every important decision gets picked apart by an AI playing devil's advocate.

The work is better. My thinking is clearer. I make fewer mistakes because I've already anticipated the criticism.

But more importantly, I've stopped being defensive about feedback. I've separated my ego from my work. I can hear "this isn't working" and think, "okay, what needs to change?" instead of "okay, I need to defend this."

That's the real return on investment. Not better blog posts (though that too). But a way of thinking about your own work that's more honest, more critical, and ultimately more productive.

If you're currently using Chat GPT the default way, where it tells you everything you create is great, you're missing most of the value.

Try the frenemy prompt. Ask for honest criticism. Sit with the discomfort of hearing what's actually wrong. Then fix it.

Do that 20 times and you'll be better at your craft. Do it 50 times and you'll think differently about your work entirely.

That's the compounding effect of taking criticism seriously. And it starts with giving an AI permission to tell you the truth.

Key Takeaways

• The frenemy prompt reframes ChatGPT's role from polite cheerleader to honest critic, unlocking drastically better feedback.
• ChatGPT defaults to praise and agreeability by design, so explicit permission to be harsh produces noticeably different results.
• Real criticism beats flattery because it identifies actual problems and gaps rather than validating existing work.
• Using the frenemy prompt 20-50 times compounds into changed thinking patterns that improve your work and decision-making.
• Best results come from multiple passes with focused critiques, human feedback integration, and tracking patterns across iterations.

Related Articles

• Deploying AI Agents at Scale: Real Lessons From 20+ Agents [2025]
• Claude Opus 4.6: Anthropic's Bid to Dominate Enterprise AI Beyond Code [2025]
• Stop ChatGPT & Gemini From Over-Explaining: Prompt Tricks That Work [2025]
• How AI Models Use Internal Debate to Achieve 73% Better Accuracy [2025]
• ChatGPT 5.2 Writing Quality Problem: What Sam Altman Said [2025]
• Cognitive Diversity in LLMs: Transforming AI Interactions [2025]