Give the model your symbol table
Macros and a short notation list so variables stay stable.
Give the model your symbol table
Ask a model to draft or revise mathematical text and the prose usually comes back fine while the notation drifts. Your vector \vect{x} returns as \mathbf{x} in one paragraph and \vec{x} in another; the expectation you write as \E[X] comes back as \mathbb{E}(X). Each variant compiles, so nothing errors, and the inconsistency only surfaces when a reader notices the same object dressed three different ways. The cause is straightforward: the model has seen every notation convention in existence and, without instructions, samples from all of them. The fix is to hand it yours explicitly.
Macros make notation enforceable
The groundwork is independent of AI and worth doing anyway: route every recurring symbol through a macro, so the source says \vect{x} rather than raw \boldsymbol{x}. This gives each concept exactly one name, lets you restyle the whole paper by editing one definition, and, for our purposes here, gives the model a small vocabulary to imitate instead of an open choice. Defining macros is covered in Custom commands.
What to paste into the prompt
Alongside your actual request, include a compact notation block: the macro definitions plus one comment per line saying what each is for.
% vectors: bold italic \vect{x}% matrices: bold upright \mat{A}% expectation: \E{X}Then state the constraint directly: “Use only these macros for these concepts. Do not introduce new notation for anything listed, and if you need a symbol that is not listed, say so instead of inventing one.” Models follow this kind of concrete, checkable instruction far better than a vague “match my style”. For a full paper, maintain the block as a short notation section in a comment near the preamble, so it is always current and always ready to paste.
Verify mechanically
The macro layer makes checking cheap. If every vector must be \vect{...}, then any raw \mathbf{ or \vec{ in model output is a violation you can find with a text search rather than by rereading the mathematics. Run that search over anything you paste in, fix the stragglers, and the notation stays uniform no matter how many model-assisted edits the paper absorbs.
Inside Oleafly
The assistant in Oleafly reads project files, including the preamble where your macros live, so it usually picks up defined commands without being told. The explicit instruction still helps, because seeing a macro exists is weaker than being told it is mandatory. A one-line addition to your request, “use the notation macros from the preamble”, combined with the red/green diff review before changes land, keeps drift out of the source.