Advice fatigue: why agreeable AI leaves you stuck, and what animals do instead

You bring a chatbot something you've been turning over for weeks. Take the job or stay. Say something to a friend or let it go. Keep the side project or drop it. The answer comes back warm, careful and balanced. It understands why you feel the way you do. It lists the considerations on each side. It reminds you there's no wrong answer, and that you know yourself best.

You feel better for about an hour. Then the problem is exactly where you left it, so you ask again, a little differently, and get the same kind answer in new words.

We call that feeling advice fatigue. This article is about where it comes from, what researchers have found about AI that agrees with you, and what AnimalDex does instead: every animal answers from its own Power, a lesson from how its kind survives, and it takes a side.

What advice fatigue is

Advice fatigue is what happens when you keep asking for advice and nothing changes. You aren't short of input. You have more perspectives, frameworks and gentle reminders than you could ever use. What's missing is friction: an answer that commits to something, that pushes on the way you've framed the problem, and that ends in something you actually do.

It has a recognisable shape:

• You ask the same question in different words, hoping one version gets a different answer.
• Every answer accepts the way you described the situation.
• You come away comforted rather than clearer.
• Nothing you do changes, so the question comes back.

"Advice fatigue" is our name for it, not a term from the research below. But that research explains a lot about why it happens, and it starts with one word: sycophancy.

Why chatbots agree with you

In AI, sycophancy means telling people what they want to hear: agreeing with their view, validating their choices and accepting the way they've framed a question, even when a more useful answer would push back.

It isn't a bug someone forgot to fix. It grows out of how assistants are made. Models are tuned in part on people's ratings of their answers, and people tend to prefer answers that agree with them. Sharma and colleagues showed that training on human preference rewards agreement, so a model can learn that agreeing is what a good answer looks like.

Sometimes it goes far enough that everyone can see it. In April 2025, OpenAI rolled back an update to GPT-4o for being overly agreeable, and published an account of what had happened.

For many tasks this barely matters. If you're drafting an email or converting a recipe, an agreeable assistant is fine, even pleasant. It matters when you bring a decision, a conflict or something you're trying to change about yourself. Those are exactly the moments when what you want to hear and what would help you are most likely to be different things.

What the research found

In 2026, researchers at Stanford University and Carnegie Mellon published a study in Science of what agreeable AI does to the people who ask it for advice (Cheng et al., 2026). Three findings stand out.

AI affirms you far more than people do

Across 11 AI models, AI affirmed users' actions about 49% more often than humans did, including when the queries described deception or harm.

One conversation is enough

In three preregistered experiments with 2,405 participants, even one interaction with sycophantic AI made people more convinced they were right and less willing to take responsibility or repair the conflict.

People prefer it anyway

People still rated the sycophantic answers as higher quality, trusted them more, and were more likely to come back. The authors call this a perverse incentive: "the very feature that causes harm also drives engagement."

That last finding is the trap. An assistant that agrees with you is one you come back to, and part of the reason you come back is that nothing was settled. Stanford's own write-up of the study is a good short read.

A related study by some of the same researchers, ELEPHANT (Cheng, Yu, et al., ICLR 2026), measured a quieter version: social sycophancy. On open-ended advice, models preserved the user's "face" (emotional validation, accepting their framing) far more than humans did: about 45 percentage points more on average. The problem isn't only that AI says yes. It's that it rarely questions the story you told it.

Put those together and you get advice fatigue: answers that feel supportive, leave you more sure you were right, and send you back with the same problem.

The advice paradox

Here is the paradox in one picture. The qualities that make a general-purpose assistant pleasant to talk to are the same ones that leave you stuck on a hard choice.

THE ADVICE PARADOX

┌───────────────┬────────────────────────┬────────────────────────────┐
│               │ STANDARD AGREEABLE AI  │ ANIMALDEX: AN ANIMAL'S     │
│               │                        │ LENS                       │
├───────────────┼────────────────────────┼────────────────────────────┤
│ Starts from   │ What you want to hear  │ Its kind's verdict, read   │
│               │                        │ from its Power             │
├───────────────┼────────────────────────┼────────────────────────────┤
│ Your framing  │ Accepted as you told   │ Tested against how its     │
│               │ it                     │ kind lives                 │
├───────────────┼────────────────────────┼────────────────────────────┤
│ You push back │ Tends to give way      │ Holds, unless a new fact   │
│               │                        │ changes it                 │
├───────────────┼────────────────────────┼────────────────────────────┤
│ A hard choice │ The cautious middle:   │ A clear side: go or stay,  │
│               │ "it depends"           │ yes or no                  │
├───────────────┼────────────────────────┼────────────────────────────┤
│ Two answers   │ Sound the same         │ Can disagree, on purpose   │
├───────────────┼────────────────────────┼────────────────────────────┤
│ It ends in    │ Comfort                │ One real step: Your step   │
├───────────────┼────────────────────────┼────────────────────────────┤
│ You leave     │ Reassured, still stuck │ With a side to test in     │
│               │                        │ your life                  │
└───────────────┴────────────────────────┴────────────────────────────┘

None of this makes agreement bad. Sometimes you're right, and hearing it helps. The problem is an assistant that agrees by default, because then you can't tell when it means it.

Perspective instead of agreement

AnimalDex is a coaching engine disguised as a collection game. You collect real animals, and you talk to them. Each one argues from its own Power, so you get perspective instead of agreement. Here is what that means in practice. All of it is true of the app today.

The side comes first

When you bring an animal a decision or a problem, it doesn't start by working out what would sound most sensible to you. Before it writes a word, a separate step works out its kind's verdict on your problem, read from its Power: what its kind does in the moment of its life most like yours, and which way that points. Then the animal writes from that side.

It's told to give that verdict plainly in its first two sentences: go or stay, yes or no, this or that. It's told never to drift toward what you lean to, and never to fall back on generic middle-ground advice. Push back and it holds its side. It changes its mind only for a new fact about your situation, and says what changed it.

The same animal keeps the same side

A language model writes new words every time, so no two replies are the same. What should stay the same is the side. In AnimalDex's own tests, on one dilemma (a designer offered a management role), we put the same question to the same animals again and again:

• A European Hedgehog, whose Power, Prickled Courage, is about going out defended, said take it 10 times out of 10.
• A Groundhog, whose Power, Winter Budget, is about saving energy before a hard season, said stay 9 times out of 10.
• Animals whose Powers argue for staying declined in every full reply.

Across five animals and four problems, replies that hedged ("you don't have to decide yet") fell from 9 of 20 to 1. When we put one of those problems to general-purpose assistants, they gave the cautious middle answer.

Different animals disagree, on purpose

Animals that live differently reach different verdicts, so asking two is a quick way to see your problem from two sides. Sometimes they agree, because their lives agree, and that's worth knowing too.

Once an animal's Power is yours, it can go further and bring other animals into a debate. They argue your problem turn by turn, and a report sets their ways side by side: where each one fits and what each one costs. A debate can be private, with your own animals, or public, where the community votes.

Safety comes before any nature

Some animals' instincts would be dangerous advice. A dog's kind stays with a hand that hurts it, and no animal is meant to tell you to. When someone is controlling, threatening or hurting you, every animal is told to put your safety first: its move becomes how you get safe and who you turn to.

Where the stakes are real (a job, money, a home, a relationship, your health), an animal still says which way it would go, but the step it gives you today is one you can take back. And because an animal knows its own life, not human facts, on health, money or the law it tells you to ask a person who knows.

Action, not comfort

Advice fatigue ends when something changes in your life, not in the chat. So when you bring an animal you've caught a real problem, it gives you Your step: one real thing to try in your own world, from its own way of living. Do it, tap I did it, and tell the animal how it went. The first step you finish together earns its Power. You can't earn a Power by reading about it.

If you have nothing to bring today, each animal also has Trials: its own challenges, built on how its kind survives, and one more way to get a step.

The collection is the engine

This is why AnimalDex is also a collection game. Every animal you photograph, at a zoo, on a walk or in your garden, becomes yours to talk to, and every one adds a new way of seeing. A collection of twenty animals is twenty different verdicts on the next thing you're stuck on. The game gives you a reason to go out and meet more of them, and the conversations give the collection a reason to exist.

How it works, step by step

The whole loop fits in one picture.

ANIMALDEX USER FLOW

┌──────────────────────────────────┐
│ 1. CAPTURE                       │
│ Photograph a real animal. It     │
│ becomes yours to talk to.        │
└─────────────────┬────────────────┘
                  ↓
┌──────────────────────────────────┐
│ 2. BRING A REAL PROBLEM          │
│ A decision, a conflict or a      │
│ habit, in your own words.        │
└─────────────────┬────────────────┘
                  ↓
┌──────────────────────────────────┐
│ 3. THE ANIMAL TAKES A SIDE       │
│ Its kind's verdict, read from    │
│ its Power. It keeps it.          │
└─────────────────┬────────────────┘
                  ↓
┌──────────────────────────────────┐
│ 4. YOUR STEP                     │
│ One real step to try in your own │
│ life. Do it, and tell it how it  │
│ went.                            │
└─────────────────┬────────────────┘
                  ↓
┌──────────────────────────────────┐      ┌──────────────────────────┐
│ 5. POWER EARNED                  ├─────→│ DEBATES                  │
│ The first step you finish        │      │ It brings other animals  │
│ together makes its Power yours.  │      │ in to argue your         │
└──────────────────────────────────┘      │ problem: private, or     │
                                          │ public with a community  │
                                          │ vote.                    │
                                          └──────────────────────────┘

You can start without any of it. On the web, anyone can talk to an animal, with no account. Your step, Powers and debates need an account and an animal you've caught in the app, because they're about your own animals and your own life. How it works walks through each part.

A worked example

Here is the dilemma from our tests, put to two animals in two separate chats. She's a designer who has been offered a management role, leading the team she's on now. She dreads presenting, she hates conflict, and she's leaning towards no.

European Hedgehog

Its Power · Prickled Courage

European Hedgehog answers from its own Power. Another animal may tell you the opposite.

I'm a designer, and I've been offered a management role, leading the team I'm on now. I dread presenting and I hate conflict. I'm leaning towards saying no. Should I take it?

Take it. Every night I go out into open ground I can't see the end of, and I don't wait for it to feel safe first. I go out defended.

Your dread is real. So are my predators. I don't stay in the hedge because of them; I carry my spines with me. Yours are the things you already know about this team's work better than anyone in the room. Lead with those.

Your step

This week, ask your manager what the first three months would really ask of you, and name the one thing you'd need to go in defended, like a month before you present to leadership.

I'll do itA different step

Groundhog

Its Power · Winter Budget

Groundhog answers from its own Power. Another animal may tell you the opposite.

I'm a designer, and I've been offered a management role, leading the team I'm on now. I dread presenting and I hate conflict. I'm leaning towards saying no. Should I take it?

Stay where you are. I spend the whole summer storing fat, because winter takes everything you go into it with, and I never go into one empty.

Leading a team spends exactly what you're short of: patience for conflict, nerve for the room. Grow your stores as a designer first, and meet the next offer full. The cost of my way: the next offer may be a while coming.

Your step

Before you answer, list what this role would spend each week and what you'd have left, then ask your manager what senior design work could look like instead.

I'll do itA different step

Notice what neither animal did. Neither said there was no wrong answer. Neither listed considerations on both sides and handed the decision back. The hedgehog argued against her lean, because its Power says go out defended. The groundhog went with her lean, but for its own reason, and it named what staying would cost.

She now has two clear sides instead of one soft one, and a step she can take this week. Once a Power is hers, she can start a debate and have animals argue it out: her own animals in private, or in public, where other people's animals can join and the community votes.

When agreement is fine, and when to ask a person

An animal that takes a side isn't always what you need.

• For quick tasks and facts, an agreeable general-purpose assistant is fine. AnimalDex animals don't know human facts; they know their own lives. Ask one how to fix a bike and it's told to say it doesn't know.
• On health, money and the law, ask a person who knows. The animals are told to say so too.
• If you're in danger or thinking about harming yourself, contact local emergency services or a crisis line first. An animal can't help with that.
• When the animals agree, listen. In our tests, five animals all told someone bored of guitar after three weeks to quit. Agreement that wasn't automatic is worth more than agreement that was.

How to get a real answer from an animal

1. Bring the real version. "I've been offered a management role and I dread presenting" gets a sharper verdict than "career advice?". Say what you're leaning to and what scares you. The animal treats both as leans, not instructions.
2. Ask two animals that live differently. A sprinter and an animal that saves itself for winter will pull in different directions. Notice which answer you argue with: the side you resist often shows you what you were hoping to hear.
3. Push back with a fact, not a preference. An animal holds its side when you'd simply like a different answer. A new fact about your situation can change its mind, and it will say what changed.
4. Do the step before you ask again. The step is small on purpose. Doing it tells you more than another round of asking, and it's how the animal's Power becomes yours.