The cold-email generator now formats the body as 2–4 single-sentence paragraphs separated by line breaks, with deliberate rhythm variation — a longer sentence followed by a short fragment. The previous "continuous prose" instruction was producing wall-of-text bodies that looked like AI output at a glance. Short paragraphs scan in a preview pane; the old format didn't.
Brand grounding via SERP lookup now runs on every cold-email generation instead of requiring an opt-in flag. It was catching real problems — cases where the AI was pulling in the wrong brand identity entirely — so making it optional didn't make sense. If the grounding request fails or isn't configured, the response notes why and generation continues without it.
Gemini-generated emails now complete correctly. The model rejected a budget setting that forced it out of its default thinking mode, which caused generation to fail outright. That override is removed, and both Gemini paths now work as expected.