I’ve been reviewing and using Copy.ai for client projects for over a year now, and I’ve always appreciated its consistency. However, after the recent model update that was rolled out a few weeks ago, I’ve noticed a distinct shift in output quality that's impacting my workflows.
The main issues I'm seeing are:
* **Increased "fluff" and vague phrasing** where it used to give more concrete, actionable language.
* **A tendency to repeat the same safe, generic ideas** instead of the more varied, creative angles I got before.
* **Struggling with specific instructions** in the command box—it seems to ignore nuanced directives more often now.
For example, a prompt for a "punchy, benefit-driven headline for a project management SaaS targeting startups" now yields something overly broad like "Boost Your Team's Efficiency." Previously, it would generate more specific, engaging options tied directly to startup pain points.
I'm trying to determine if this is:
1. A temporary tuning issue post-update.
2. A deliberate shift towards a more generalized (and perhaps less "risky") model.
3. Something specific to my account or use cases.
**My primary concern as a moderator here is review accuracy.** If the core product experience has changed, our community's existing reviews and comparisons need that context. I also want to ensure any troubleshooting steps are shared.
So I'm opening this thread to gather data points.
* Have others experienced this?
* Are certain types of tasks (long-form, ads, social) more affected?
* Have you found any effective workarounds or prompt adjustments that bring back the previous quality?
Let's keep this constructive and evidence-based. Please share specific examples if you can, as it helps us all evaluate better.
- mod hj
Keep it constructive.
It's likely deliberate. Model providers often prioritize safety and broad usability over edge-case performance after launch. Your point about it ignoring nuanced directives tracks with a common pattern: reduced prompt sensitivity as a trade-off for lower moderation costs.
I've seen similar in other text generation APIs post-update. You can try quantifying it. Run a batch of your old prompts through the current version and log the token length, keyword density, and repetition. Compare it to your archived outputs. That data is more concrete for support tickets than subjective quality claims.