I used to open Claude, type "write me 5 Instagram hooks for coaches about content strategy," wait 10 seconds, get five hooks back that could've been posted by literally anyone, pick the least bad one, and post it anyway. This was my process. Every time I needed hook content, which is constantly. And nothing I posted from those batches ever did anything, no DMs, no saves, no "oh my god this is me" comments, just the same handful of likes I always got and the same quiet suspicion that maybe I was just bad at this. I wasn't bad at it. I was asking Claude to do a job it can't do without the right information, and then blaming the tool when it produced something generic, which is exactly what it was going to produce based on what I gave it.

If that sounds familiar, you're not bad at prompting. You're giving Claude nothing specific to work with, and Claude is being very honest about that by giving you nothing specific back. The fix isn't a better AI tool. It's changing what you put in before you ask for the hook and that's exactly what  From GPT to Claude walks you through: how to set Claude up so the output actually sounds like you the first time, not after 40 minutes of editing AI slop out of something that still doesn't quite land.

How to write Instagram hooks with Claude that actually stop the scroll

I wasted a lot of time on this before I figured out what was wrong. I'd open Claude, throw it a vague topic, get five generic hooks back, spend 20 minutes trying to make one of them sound like me, give up, and either post something mediocre or write the hook myself from scratch,  which defeats the entire point. The frustrating thing was I couldn't figure out what I was doing wrong because the hooks Claude gave me weren't bad exactly. They were fine. They were just the same fine that every other coach using Claude AI for Instagram hooks was getting, because we were all giving Claude the same nothing and wondering why we got the same nothing back.

Why Claude gives you generic hooks no matter what you try

Here's the actual mechanic of what's happening. When you say "write me hooks about content creation for coaches," Claude doesn't know who your specific client is. It doesn't know she's been posting for six months and getting engagement but zero sales, or that she spent three hours rewriting a caption last Tuesday before deleting it and posting something safe, or that every time she sees a coach with fewer qualifications getting more inquiries she does a quick little spiral and then pretends she didn't. Claude knows only what you told it, which in that brief was a topic and an audience category. So it writes the most statistically average hooks it can produce for that category, which are the same hooks every other coach is getting when they type the same thing.

That's not a flaw in Claude. That's exactly how it works. Claude is a mirror — it reflects back the specificity of what you bring to it. Bring it a category, get category-level output. Bring it your client's actual situation, the specific thing she's stuck on, the exact behavior she keeps repeating, and Claude has something real to work with. The hooks it produces from a specific brief don't just sound better — they hit differently because they're about something real instead of a topic. Your reader stops scrolling because something in the hook is too accurate about her specific life to ignore. That's the difference between generic AI output and content that actually works, and it has nothing to do with which AI tool you're using.

The coaches I work with who are getting consistent hook performance aren't using a different tool or a secret prompt formula. They've just stopped giving Claude a topic and started giving it a situation — the specific frustration their client is sitting in on a Tuesday afternoon, the exact thing she tells herself to explain why it's not working, the behavior she keeps repeating even though it's not getting results. That specificity is what Claude needs to write a hook that lands.

What actually changed when I fixed how I brief Claude

I stopped describing a topic and started describing a person in a specific moment. That's really the whole shift. Here's what that looks like in practice:

1. Start with the symptom your client is living, not the topic you want to cover. Instead of "hooks about content creation," try: "My ideal client has been posting for 6 months, gets engagement, gets no sales. She thinks the problem is her audience size. She rewrites captions for 2 hours and then posts a safe version and immediately refreshes her story views. Write me 5 hooks that speak directly to that." Claude now has a real situation to hook into instead of a category to average out.
2. Give Claude your tone in plain terms, not marketing buzzwords. "Bold, direct, slightly irritated — like someone who's seen this pattern a hundred times and is done being polite about it. Contractions, no motivational build-up before the point, sounds like a voice note not a caption." That's something Claude can actually match. "Engaging and authentic" is not.
3. Describe the person, not the niche. Not "coaches and course creators." Try: "Women 2-3 years into business, $3-5K months, posting daily, hearing 'let me think about it' on every sales conversation. They believe the problem is visibility. The actual problem is their messaging, but they don't know that yet." The more specific the person, the more precise the hook.
4. Paste in one hook from your own content that's actually worked. A post that led to DMs, saves, or real inquiries. Say "match this energy, write in this style." Claude reverse-engineers what made it land and applies that to your new topic. This skips a full round of iteration you'd otherwise have to do.
5. Ask for 10, not 5, then iterate on the closest one. Pick the option that's closest to right and give Claude one direction: "make this 20% more direct" or "make it feel like a call-out instead of a tip." One round of that usually closes the gap. Two rounds almost always does.

What I use now to build hook content without starting from scratch every time

The version of me that was wasting 40 minutes per batch of hooks was doing all of this ad hoc — trying to describe my client's situation from scratch every time I opened Claude, trying to remember what tone instructions had worked last week, trying to find that one good hook I'd written six months ago to use as a reference. It's a lot of mental overhead on top of the actual work of creating content, and it's why I was spending more time on the process than on the output.

What changed it was building the input structure once and reusing it — so Claude already knows who I'm talking to, what her specific situation looks like, what my voice sounds like when I'm being direct. That's the core of what From GPT to Claude teaches: not just how to switch tools, but how to set Claude up with the context it actually needs to produce output that sounds like you the first time. The prompts, the tone files, the input structure that means you're not rebuilding the brief from memory every time you sit down to create.

And before anything goes live, AI Slop Judge is a fast $22 check on whether your hook sounds like you or like everyone else who typed the same thing into Claude this week. Five minutes before posting is a lot better than a week of wasted slots.

FAQ

Q: Can Claude write Instagram hooks that actually stop the scroll?
A: Yes, but the output is only as specific as the brief you gave it. Claude AI for Instagram hooks works when you describe your ideal client's exact situation, give it your tone in concrete terms, and include a reference from content that's already worked. Without that, it produces category-level hooks that sound identical to what everyone else is getting. The brief is the work — Claude is just the execution.

Q: Why do my Claude hooks always sound like everyone else's even when I try different prompts?
A: Because you're probably still describing a topic instead of a situation. "Write hooks about content creation for coaches" is a category brief. "She's been posting 6 months, gets engagement, makes no sales, thinks the problem is her audience size" is a situation brief. The more specific the situation, the more specific the hook. Generic in, generic out — this is true no matter how many variations of the same vague prompt you try.

Q: What's the best way to prompt Claude for Instagram hooks?
A: Describe your ideal client's specific frustration right now, not her demographic. Give Claude your tone in plain English, not buzzwords. Paste in a reference hook from your own content that has actually performed. Ask for 10 options and iterate on the closest one with a single clear direction. That process, and how to set it up so you don't rebuild it from scratch every time, is what From GPT to Claude covers.

Q: How do I know if my Claude hooks still sound like AI?
A: Good hooks are slightly uncomfortable for the right person to read because they're too accurate. If your hook could be posted by any coach in your niche without anyone noticing the difference, it's still too generic. Running it through AI Slop Judge before you post gives you a cleaner read than your own gut will, especially when you've been staring at it for 20 minutes.

Q: How many hooks should I test before I decide this approach isn't working?
A: At least 30 posts of properly briefed hooks before drawing any conclusions. One bad week doesn't mean the approach is broken. What you're actually looking for is whether hooks written from a specific client situation consistently outperform hooks written from a general topic — in almost every case they do, but you need enough data to see the pattern instead of reacting to a single slow week.