@Kerri-Ann Watson It is harder using your own GPT, but that is also why it is more useful. You are learning where the pieces actually sit, instead of just copying my example. On the name issue, that is usually a field problem. Your GPT is probably mixing up the person receiving the email with the person signing the email. For now, I would add one clear rule into the GPT instructions: “Always sign the email from Kerri-Ann Watson. Never use the school name, principal name, or client name as the sender.” Long term, I would separate the fields more clearly: - school name - principal name - school email - sender name - draft email - review notes But don’t rebuild the whole thing yet. Just fix the sender name first and test again. Also, if you added an extra field, it needs to be added in all the right places: the Google Sheet column, the GPT Action schema, the n8n mapping, and the GPT instructions. If it only exists in one place, it may not come through properly. On researching the school, yes, that is a good idea. But I would not add that layer yet. For now, do the research manually. Copy 3-5 notes from the school website into your GPT and let it use those notes to write the email. Later, you can connect a research step. But that is another layer that requires an LLM doing the research, and I would get the draft-to-review-queue working cleanly first. So your next move is simple: Fix the sender name. Run one fake school test. Check that the row lands properly. Check that the draft is signed by you. That is enough for Week 7. And thank you for the wedding message, very kind of you - was the best day of my life :)!!!

@Kerri-Ann Watson I'd suggest to test it with a small fake example. Week 4 is like: if I come back tomorrow and use it again, have I made it clear enough that it knows what job it is meant to do? So for “fake test input”, don’t put your GPT instructions there. Just put a pretend example you would give the assistant. Something like: - School: Green Valley Primary - Principal: Sarah Jones - Program: confidence and wellbeing workshop for Year 5 and 6 students - Notes: the school is focused on resilience this term - Goal: write a warm first-contact email asking if they would like more information That is enough to test it. Then your proof could be really simple: - My assistant: School Outreach Email Assistant - Where I packaged it: Custom GPT - What it does: Helps me turn school/program notes into a first-contact email draft. - Starter prompt: Use these school notes to write a warm first-contact email. Do not invent missing details. Tell me what I need to check before sending. - Fake test input: Green Valley Primary example above. - Pass/fail: Pass, but needs refining. It gave me a usable first draft, but I still need to improve the tone and make the missing-info checks clearer. - Human review point: I check the principal name, school details, tone, accuracy, and anything the GPT may have guessed before I send it. The proof is “I now have one assistant, for one job, with one test, and I know what I need to check before trusting it” That is a good Week 4 result in my opinion :)

@Kerri-Ann Watson You’ve got the main thing right: the AI is not sending the email. It is preparing the draft and putting it into a review queue so you can check it first. The blurry part is probably: “What exact information does the GPT need from me before it creates the school email?” Some fields are mandatory in my opinion such as: - School name - What program you want to introduce - Who the email is for, even if that is just “the principal” - The purpose of the email - The draft email - Status: Needs Review - Original notes Optional but useful: - Principal name - School email address -> depends who you send this email to, this might be in the mandatory fields. - School location -> if programs are relevant to the location this might also be mandatory. - Year levels - Anything personal about the school - Previous contact or relationship - Tone preference - Any deadline or event date PS: nobody knows better than you what is mandatory and what is optional, I shared the option of having fields that are mandatory and not to allow for some structure in your GPTs and response :). If the principal name is missing, the GPT should not invent one. It can write the greeting as “Hi there,” or “Hi Principal,” and put this in the review notes: “Principal name missing. Please check before sending.” Same with the email address. If the school email is missing, the draft can still go into the review queue, but it is not ready to send yet. So your workflow could become: “When I give the GPT details about a school and my program, create a first-contact email draft for the principal, save it into my Google Sheet review queue, mark anything missing in the review notes, and set the status to Needs Review.” The key rule is: Missing information does not mean the workflow fails. It means the row gets saved for review, and the missing parts are clearly marked. The next automation step would not be “send it automatically” yet. A safer next step would be: once you mark the row as approved, it creates a Gmail draft for you to review one final time.