I've heard that this is one of the many "ticks" that LLMs develop, e.g. in www.youtube.com/watch?v=ugvH... there is a discussion that model was over-using the construct "Certainly, ...".
I wonder if this is an artifact of RL from AI Feedback arxiv.org/pdf/2212.08073 that somehow prefers this answer

Special shout out to @aborg.dev, @kristen-fm.bsky.social, and @darkscience@mastodonsweden.se! ☺️