Why does my AI blog content suddenly sound generic after week two

I’ve been using a specific prompt chain with GPT-4 to draft SEO articles for my niche SaaS site. For the first two weeks, the output was surprisingly sharp. It captured the specific pain points of our users and used a conversational tone that felt almost human. I was genuinely impressed by how well it handled the nuance in my detailed briefs. However, starting last Monday, the quality took a nosedive. The same prompts, same system instructions, same temperature settings, yet the output now feels flat and repetitive. Phrases like "unlocking potential" and "streamlining workflows" are popping up everywhere, even when I explicitly instruct the model to avoid clichés. I tried switching to a different model provider just in case, but the issue persisted. I also checked my API version to ensure I wasn't hitting a deprecated endpoint, but everything looks stable. I’m wondering if this is a temporary regression in the model’s behavior or if I’m inadvertently training the context window to repeat its own previous outputs over time. Has anyone else experienced this sudden loss of stylistic consistency? I’m currently testing whether breaking my prompts into smaller, more granular steps helps, but I’d love to hear if there’s a known way to reset the stylistic tone without rewriting my entire prompt architecture. It’s frustrating because I was just about to scale this workflow to cover three more product lines.