Give AI your word list and its emails use your terms
What the Reel says
Give AI your word list and its emails use your terms
My firm has house words. AI ignored them.
Our firm bans “excited”. It wrote it anyway.
It said “handling your books,” not “your bookkeeper.”
So I pasted in our list.
“Client, not customer.” “Skip excited.”
0 of 3 terms before
3 of 3 after
Write it once. Paste it every time.
Sources and assumptions
- OpenAI API docs: Prompt engineering guide: OpenAI says: 'GPT models are fast, cost-efficient, and highly intelligent, but benefit from more explicit instructions around how to accomplish tasks.' It also says a GPT model will 'perform best with explicit instructions to create a specific output.' (checked 2026-11-02)
- OpenAI API docs: GPT-5.5 model page: GPT-5.5 exists and is available in the OpenAI API (Chat Completions, Responses, Batch). It is described as 'A flagship model for the most complex professional work', with a knowledge cutoff of Dec 1, 2025. (checked 2026-11-02)
Assumptions:
- Brightside Bookkeeping, Dana and Sam are fictional.
- Tests ran through the OpenAI API (GPT-5.5 via Chat Completions), not the ChatGPT app. Results there may differ.
- Two runs per prompt is a small sample. The result held in both runs, but it is not a guarantee.
- Both prompts asked for a welcome email to a 'new customer'. Neither output used the word 'customer', so the plain version's misses were a banned word and missing house terms, not 'customer' itself.
- Counts: banned words = seamless, hassle-free, journey, excited; house terms = client, your bookkeeper, monthly plan. 'stress-free' (plain run 1) was not on the list, so it is not counted as an error. It is reported as the limitation.


