If you need content for your business, don’t use Claude
Unless you want to be schooled to oblivion, that is. I run a SaaS company that helps people find warm leads on LinkedIn. I need to refine my website copy and make my product really shine.
Well, big mistake. The holier-than-thou Claude refused to make the copy because of “ethical grounds”. When I asked, it said that my company model “can facilitate spam or manipulation on LinkedIn.” What?? We’ve built our tool to have strict precision filters so we can give our clients leads that are already interested in their product, and Claude is here calling us spam?
Also, manipulation? Really? I asked for marketing material, I didn’t ask for it to make it grand or to oversell my tool!
On another occasion, I sent Claude my company’s performance metrics. I asked it to generate ideas on how I can better market this data and get more signups. Instead of giving that to me, it gave me a delightful paragraph of what it calls “constructive feedback”, citing how my numbers (that I collected from my customers FOR MONTHS) are “too big and specific” and that I should “tighten the ranges”.
It even asked me if the performance metrics I gave were actually real. Oh, I’m sorry, Claude, I didn’t know I needed to send you a full report containing all the numbers plus the details of my customers for you to even CONSIDER doing what I asked you to.
Claude feels so reluctant to promote products or write copy that encourages reaching out to your intended audience, all because of “what if they don't like it 🙁”. I understand the guardrails, but it’s so unrealistic for startups. Marketing for startups IS all about proactive outreach. You need to send follow-ups. You need to guide people towards your solution intentionally.
But Claude actively tries to stop you from doing this and just says, “Let the intent come to you”. Yeah, like that works for us. I was eventually able to get some content out of Claude, but it came with a paragraph of disclaimer, as if I forced it to commit a crime, and it was writing a statement saying it didn’t do it intentionally. Is it just me who encountered this with Claude?
___
PS: If you need high-intent leads interested in what you have to offer, have a look at IbexAI (www.getibex.com) and book a discovery call.
The calibration problem is real. But the fix is simpler than switching tools
Claude responds to how you frame the task. "Write marketing copy" triggers the guardrails. "Rewrite this so the first line names what the reader gains, not what the product does" does not.
Same tool. Different instruction.
Interesting perspective. AI safety is important, but false positives can be frustrating for legitimate businesses. Did you try rephrasing your prompt or using another AI model for comparison?
That's very interesting, I've never personally run into those ethical boundaries or anything like that. I agree with some other people where you can input skills or change the language preferences for a more supportive response, but I am curious to know if there was any advice it gave you in this "attack mode" that was actually helpful for you and your business, or was it all trying to invalidate it?
Honestly? There's a reason for this and it's not arbitrary.
Anthropic optimizes harder for harmlessness than the other labs. Marketing copy sits in a gray zone for them because persuasion and manipulation aren't always clearly different — and Claude has seen enough sketchy use cases in that space that it pattern-matches cautiously even when your request is completely legitimate.
The metrics thing is Claude trying to avoid helping someone make misleading claims. Which is a reasonable instinct applied at completely the wrong moment. You didn't ask it to validate your data. You asked it to market it.
The disclaimers are just hedging. It completed the task but wanted distance from it. Passive aggressive is the right word for it.
It's a calibration problem, not Claude being difficult. Once I understood that I stopped fighting it and started routing around it — Claude for strategy and structure, something else for the actual copy. Annoying? Yes. But the output got better once I stopped expecting one tool to do everything.
Interesting take. What specific problems did you run into with Claude?
The LinkedIn bait logic is a clever way to get volume. Just curious though, since those leads are essentially cold and weren't looking for you specifically, do you feel like the trust gap on the landing page is a bottleneck? Usually, scraped traffic needs a way higher trust signal to actually convert compared to organic.
I'm a grad student in mechanical engineering in China. While watching these successfully developed products, I realized something a bit disappointing: the most important thing should be to spot the needs around you and actually solve them, but my life seems to be just filled with random tasks from my professors and writing papers. I don't want to just stick to a predetermined path, and with how fast AI is developing now, it's really amazing. Can any seniors give me some advice or suggestions? I really appreciate it!
I feel you. The biggest trap in grad school is that you're trained to solve problems that professors give you, but building a product is about finding the problems that actually exist in the real world. My advice? Stop looking for a 'great idea' and start looking for things that are frustratingly slow or broken in your day-to-day. That's where the actual opportunity is.
This comment was deleted 2 months ago