Twenty to thirty posts can be a strong voice dataset, but I would not feed them in as one unlabelled pack. Tag each post by audience, goal, format, hook, proof, and call to action. Then ask the model to infer your writing rules rather than copy your phrases. Keep five posts out as a test set and compare the generated drafts against them for tone, clarity, and specificity. If evyAI accepts examples, start with eight to ten diverse posts and compare the result before adding all thirty. Which part matters most for your persona: tone, structure, or point of view?

@Rick Kloete Yes. I separate four numbers: impressions, profile views, relevant conversations, and agreed next steps. The first two show reach. The last two show whether the content is creating business momentum. When a post starts a useful conversation, I would trace it back to the exact idea or comment that opened the exchange. Which of your recent posts produced the best conversation?

LinkedIn measures characters, not words, so a 400-word post can still cross the limit depending on sentence length, line breaks, and formatting. I would cut in this order: repeated setup, examples that make the same point, then the weakest closing sentence. Keep the opening, one specific example, and the question you want readers to answer. Actually, shorter is useful only if the central idea becomes clearer. Does the warning appear inside LinkedIn itself or inside the evyAI editor?