Other generators provide you with instructions and leave you alone in front of the document folder. Here, you leave with the list of files, their purpose, their internal headers, and their name already decided.
Your Custom GPT doesn't respond poorly because it lacks instructions: it responds poorly because you've uploaded three huge PDFs and it can't find what it needs inside.
When you generate your custom GPT's profile, this tool doesn't stop at the text of the instructions: it adds a knowledge files list. It's a list of between 0 and 6 documents, and each one comes with four things resolved: what it is for (a sentence), in what format it should be (md, csv, json, or txt), a skeleton of internal headers, and the file name with which you will save it.
The skeleton of headers is the part that changes the outcome. Each section addresses a single concept because the GPT's internal search doesn't read the entire document: it retrieves the fragment that best fits the question. If a section mixes prices, schedules, and return policies, the GPT will return that entire block for any of the three questions. One concept per section is what ensures it retrieves the correct piece.
File names are always constructed using the convention NN_topic_vN.ext: the number orders by priority, the topic is in lowercase without accents or spaces, and the version avoids confusion between revisions. You can copy each name with a click, mark the documents you have prepared, and track progress (“3 out of 5 prepared”); the missing ones also appear as tasks in the GPT's publication checklist.
And if your GPT doesn't need external knowledge, it tells you so: it puts it in writing instead of inventing documents to seem more complete. Uploading files that don't add anything doesn't leave the GPT the same; it leaves it worse.
Use it free without an account. If you want to save your result, you can sign up afterwards.
Save / Sign upFrom 0 to 6 documents, never a filler list
if everything the GPT needs is already in its instructions, it clearly states so instead of recommending papers that only add noise.
One purpose per file, in one sentence
you know exactly what each document needs to contain before you start writing it, and what it should not contain.
Ready-to-copy header skeleton
3 to 6 short headers per document, one concept per section, so that the GPT's semantic search retrieves the correct snippet and not the entire chapter.
File name already decided
NN_tema_vN.ext convention applied automatically and copyable with one click. No more 'final definitive document.pdf' that helps neither you nor the GPT.
Deliberately chosen format
md for structured text, csv for tables, json for data, and txt for plain text. Each document also includes an estimated length so you don't go over.
From plan to publication without jumps
each document becomes an item on the GPT's publication checklist, with its status saved, and you can check off the ones you have ready.
Made with ChatGPT by hand, not in the abstract
the plan comes from the same generation as the system instructions, candidate names, and conversation starters, so the documents fit with the GPT you just defined.
Define your custom GPT and the file will be generated without an account. The complete knowledge files list, with headers and names, is unlocked with the free account.
In the features that use AI, responses are generated by artificial intelligence and are not a substitute for professional advice.