It is not the model
When AI outreach reads like filler, the instinct is to blame the model. It is almost always the wrong culprit. The same model that writes a flat email will write a sharp one when it has something specific to say.
The problem is upstream. Most tools give the model a thin brief and a name to insert, and a thin brief produces a thin email.
Structured context beats a prompt
A single prompt field is where generic outbound comes from. It flattens everything you know about your product into one paragraph that nobody maintains.
The alternative is to hold brand and product as structured records: positioning, the problems you solve, the proof you can point to, the objections you hear, and the tone you use. Records can be reused, updated and checked. A prompt cannot.
Research is what makes it specific
Context tells the system who you are. Research tells it who they are. Recent news, a role change, a public update: these are the details that let the same product be pitched in the way that matters to each person.
Without that, personalisation is just a merge field. With it, the first line has a reason to exist.
What good input looks like
Good input is specific and honest. Name the two or three things you are genuinely better at, and the situations where you are not the right fit. Give real proof rather than adjectives, and write the objections down as you actually hear them.
Do that, and the output stops sounding like AI. Not because the model changed, but because it finally had something to work from.