Strategy and decisions
Use ChatGPT to Turn a Product Brief into a Reviewable Glossary
Cihan's view: Upload a supported product brief, ask ChatGPT to separate stated definitions from proposed wording, and verify each glossary row against the source.
Product briefs often use the same word in three different ways. That is how a small language problem becomes a support problem later.
ChatGPT can work with uploaded documents and compare or analyze their contents. A safer use is to ask it for a glossary that keeps source wording separate from suggestions. You still need to verify every row.
This is a documentation-based workflow using fictional text. It does not claim a personal test or a customer result.
Reader and decision
You are a product operations lead preparing a launch brief for Support. Your decision is whether the glossary is safe to circulate as a review draft, or whether product owners need to resolve the open terms first.
The artifact is a five-column glossary:
- Term
- Definition stated in the brief
- Plain-English explanation
- Source quote
- Open question
The decision is not “is this polished?” It is “which terms are actually defined, and which ones still need an owner?”
Before you start
You need:
- ChatGPT with file uploads available in your account.
- A supported file such as DOCX, PDF, TXT, or CSV. The practice input below is plain text.
- Permission to upload the document. Do not use confidential product or customer data in a practice run.
OpenAI’s file upload documentation says ChatGPT can summarize, compare, and analyze uploaded files. It also says availability and limits depend on plan and account settings. Treat the current interface and your account permissions as the source of truth.
Fictional input
Create a file named northstar-brief.txt with this content:
Northstar workspace brief
Northstar is a shared workspace for reviewing supplier onboarding requests. A request is "ready" when the required company details and a named internal reviewer are present. The brief does not define which company details are required.
The review queue shows requests that need attention. A request can be "paused" when the supplier has not provided a document. Paused requests remain visible, but the brief does not state whether the review timer continues.
The workspace includes a reviewer note. Reviewer notes are internal comments attached to a request. A note may mention a missing document, but the brief does not explain who can read notes.
The launch team wants to describe Northstar as a "single source of truth." The brief does not define this phrase or say which system remains authoritative when data conflicts.Three steps
1. Upload the brief and set the evidence rule
Start a new ChatGPT conversation and attach northstar-brief.txt. Use this prompt:
Read the attached product brief and create a glossary with exactly these columns:
1. Term
2. Definition stated in the brief
3. Plain-English explanation
4. Source quote
5. Open question
Rules:
- Include only terms that appear in the brief.
- Copy source quotes exactly, using short quotes that prove the row.
- Do not turn a proposed marketing phrase into a product definition.
- If the brief does not provide a detail, write "Not stated in brief."
- Put unresolved ownership, permissions, timing, or system-of-record questions in Open question.
- Do not invent a launch date, SLA, permission model, or required field list.
Return Markdown only.The prompt asks for a structured artifact, but it does not make the result authoritative.
2. Inspect the rows, not only the prose
A useful illustrative output could look like this:
| Term | Definition stated in the brief | Plain-English explanation | Source quote | Open question |
|---|---|---|---|---|
| Ready | Required company details and a named internal reviewer are present. | The request has enough information for review, but the required details are not listed. | “A request is “ready” when the required company details and a named internal reviewer are present.” | Which company details are required? |
| Paused | A request can be paused when the supplier has not provided a document, and it remains visible. | The request stays in the queue while waiting for a supplier document. | “A request can be “paused” when the supplier has not provided a document.” | Does the review timer continue while paused? |
| Reviewer note | An internal comment attached to a request. | A note is a comment for internal review, but the audience is not defined. | “Reviewer notes are internal comments attached to a request.” | Who can read and edit notes? |
| Single source of truth | Not stated in brief. | A proposed phrase that needs a concrete system and conflict rule. | “single source of truth” | Which system is authoritative when data conflicts? |
This table is illustrative. Replace " with normal quotation marks in your actual result. The placeholder only keeps the example readable inside the Markdown table.
3. Resolve the gaps with owners
Ask a follow-up question that does not rewrite the product brief:
Using only the attached brief and the glossary you created, return an unresolved-question list with columns: question, why it matters for Support, proposed owner, and evidence. Set proposed owner to "Not assigned" when the brief does not name one. Do not answer the questions.A reviewable result should leave proposed owners as “Not assigned” when the source does not provide them. Product can fill them in later.
Acceptance checks
Before circulating the glossary:
- Every term appears in the fictional brief.
- Every definition has a source quote from the brief.
- The glossary does not define the required company fields, timer behavior, permissions, or authoritative system by guesswork.
- “Single source of truth” is marked as proposed wording, not a product capability.
- Open questions are specific enough for a product owner to answer.
- A human checks the quotes against the uploaded file.
If the result is wrong
ChatGPT adds plausible details. Ask it to return only source quotes and the exact phrase “Not stated in brief” for missing details. If it keeps adding details, do the extraction manually.
The file will not upload. Check the file type and account availability. OpenAI lists DOCX, PDF, TXT, CSV, XLSX, XLS, and TSV among supported formats, but plan limits and settings apply. Export a Google Doc to PDF or DOCX before uploading.
The source quote is inaccurate. Ask for shorter quotes and compare them with the file. A glossary row without a verifiable quote is a draft, not evidence.
The output becomes a long summary. Re-send the exact column list and ask for Markdown table output. Keep the glossary separate from the narrative summary.
The useful limit
This workflow is good at finding where a brief is vague. It cannot decide the product policy behind the vague sentence. Use the glossary to create questions for the people who own the answers.
CTA
For more practical AI workflows for business decisions and execution, visit cihanuygur.com.
Sources
- OpenAI Help Center, File Uploads FAQ: documents supported uploads and using files to summarize, compare, and analyze information.
Evidence label: Documentation-based. The product brief, glossary rows, and outputs are fictional teaching examples. No personal test, screenshot, performance result, or customer outcome is claimed.