Owners of busy restaurants do not reply to their Google reviews. Not because they don’t care, but because it’s forty minutes at eleven at night and it’s the first thing that goes. So Titanium drafts the replies for them, in the voice of that particular business, and the owner reads them and sends.
Getting a model to write one convincing reply is an afternoon’s work. Getting it to write a hundred that don’t read as though a machine wrote them is a different problem, and it’s the one worth explaining.
Four passes, not one prompt
A reply gets drafted, then a second pass critiques that draft and rewrites it. That second pass matters more than it sounds: a model told to avoid a phrase will satisfy the ban lexically and leave the tell intact, so you need something looking at the output rather than at the instruction. Then it’s trimmed to a word budget that shifts with the star rating, and finally checked against the rules that can’t be broken. It’s fenced off from inventing facts — it cannot name a dish the reviewer never mentioned.
The part I’d actually point at
Google removes owner responses that mirror each other across a profile, and a model has no memory of what it wrote yesterday. So every reply that actually goes out is recorded, and the last thirty are mined for repeated openings, closings and three-word runs, which are handed to the next draft as a list of things it may not say.
Variation is enforced outside the model, because it cannot be trusted to the model. That is the difference between a demo and something you can leave running.
Measured, not guessed
Every version is scored against a fixed set of twenty-five reviews, run before and after every change to the prompt. Without that, you are changing the wording and guessing whether it helped.
Scored against that fixed set, the replies went from 3.5 out of 5 to 4.8.
Most people building on these models are guessing.
