Hope you don't mind a longer comment on the nature of this endeavourer. I'm really interested in the phenomenon and like your approach, but uncertain if this methodology will yield results that give evidence on the matter.
With this sentence you'll mix up the results: "summarize what was said if you don't remember exactly, or if the conversation was long and contained a chain of *convincing subpoints*"
To summarize "a speech act" as a conversation muddies the waters on what we're researching here. If a conversation with subpoints counts as "a speech act", then why is this conversation bound to one moment? How about a speech club? If I changed my belief at the end of the semester, surely it was due to "speech acts" on many occasions, but that's not exactly what you're interested in here, if I understand correctly; especially when you're interest was sparked by GPT-2.
So, if I got you, you're referring to a specific act of speech (sentence or sentences) that was/were uttered by one person, to one or many other people, after which they found themselves convinced of a different belief than the one held previously.
Just spontaneously, I'd say my hypothesis would be that there are some four typed of this speech (I used ridiculous examples on purpose):

1) new information that was interpreted as highly probable or with enough probability to tip the scale to the other belief (your trusted friend telling you "I saw your partner cheated on you, I saw it with my own eyes")

2) a summary of different information you already believed in, that gave structure to them, making the model as a whole more probable (your leftist friend at a party telling you "You already know the rich don't pay enough taxes and this has not changed since the financial crisis, even though it was supposed to, because those people pay politicians. Quit supporting a social capitalism and go full communism")

3) the presentation of a view, when the previous belief was held very loosely (your friend at lunch "look, just take the pizza instead of the salad")

4) the presentation of a view you already believed to be true, but where some aspect was diminishing the belief, i.e.. shame of that belief, or guilt (a politician saying "we all know *insert minority here* is actually responsable for the problens you're having").

The most important aspect of these sentences (both in fiction and real life) is their context. That goes for private conversation and political speeches. In order to really analyse the importance of these sentences you'd need more context than the survey supplies. My prediction (65%) would be, that these sentences sound to outsiders either increadably banal (4, 2) or possibly likely, but not without context (1, 3). The same divide applies to what we define as "belief" (I know, I don't wanna get into this either, just hear me out): 4 and 2 are aspects you beliefed already, if the evidence is looked at closely enough; the person just realises in that moment they actually do hold this belief. 1 and 3 are situation where a shift in evidence actually shifts your belief.
Behaviour is another bag altogether, imho. Good luck! I'm interested in the results nonetheless.

I felt so hilariously called out by this post I actually bothered to make an account after years of lurking. I fit that patient description quite well, or did anyway. This is one aspect of why I always went for CBT and never even considered psychoanalasys.

I'd be highly interested in further thoughts on the matter. In case you care for a patient reflection (I'm no psychologist): I think the social connection between identity (for example job) and social relations can be close. I chose to hold onto my identity because I thought all my relationships would crumble if I gave it up, even after the identity was no longer serving my experience. This is also not entirely unreasonable - many relationships do rely on collective identity, for better or worse. It is about preserving the one life that was aquired through the means of this identity, the job etc., and not building a new life with a new identity. This might be why patients rather stay in a job that makes them want to kill themselfs and don't see a way out - they might try to save the life connected to this identity by all means.