Review Round to Task List: A Four-Prompt ChatGPT Chain That Turns Comments Into One Direction

What This Builds

A review round leaves you with a pile of comments from the creative director, the client brand lead, legal and the account lead, some of them contradicting each other. This chain turns the pile into one numbered direction for the team, a separate list of conflicts for you to decide, a list of items for the account team, a task list for each production role, and a table proving that every original comment was handled.

The Level 2 Gmail summary shows who said what in a long thread. It does not give the team a direction. This chain does, and it checks its own work against your original list.

Prerequisites

• A ChatGPT account. A paid Plus plan is $20/month, and Unlimited everyday text chats with GPT-5.6 Luna plus a Think button for harder questions, with limited uploads, image generation, voice, and deep research. describes the free tier. Check the pricing page for limits.
• The round's comments in one place
• Total ongoing cost of the finished build: none beyond your ChatGPT plan
• Permission to paste the comments (see below)

The Concept

A good production lead does not hand the team a transcript. They sort the comments, settle the clashes, and give each person a short list. This chain does the sorting and the formatting. You do the settling, because choosing between a client's wish and the creative director's is your call.

Each step ends with a count or a check you do yourself. If the numbers do not match your original list, stop and fix it before moving on.

Before You Paste Anything

Review comments describe unreleased work, and they carry what clients and colleagues said about it. Check the agency's or company's AI policy, the client contract and ChatGPT's data settings (whether chats are used to improve models, and how long they are kept). Free consumer plans may differ from business plans.

If pasting is allowed, replace the client and product names with code names, and replace each person's name with a role label: CD, Client brand lead, Client legal, Account lead. No names, no email addresses. Number the comments 1, 2, 3 yourself before you paste.

Build It Step by Step

Part 1: Prepare the comments

Open a new ChatGPT conversation and keep all four prompts in it. In a document, list each comment on its own line as 1. CD: [comment text]. Keep the comments exactly as written. Remove names and anything outside the layout feedback. Keep a copy of this numbered list. You will check the final table against it.

Part 2: Prompt 1, group by layout area

Below are numbered comments from review round [round number] on one layout, each with a role label. Group every comment by the layout area it is about (for example headline, image, logo, footer, format). Quote each comment exactly with its number and role label. Do not merge, summarize, rewrite or drop any comment. If a comment touches two areas, list it under both and mark it "also under [area]". End with a count of comments per group and the total.

Comments:
[paste the numbered comments]

Checkpoint 1. Count. The total must equal the number of comments you pasted, and every number from your list must appear. If one is missing, reply Comment [number] is missing. Add it. and re-check. Note that a comment listed under two areas will make the group counts add to more than the total, so count the distinct numbers.

Part 3: Prompt 2, one direction

Using the grouped comments, write:
1. A numbered direction for the team: one clear instruction per item, each naming the comment numbers it comes from.
2. A section "You decide": comments that conflict, showing who said what, with numbers and role labels. Do not resolve them.
3. A section "For the account team": any comment that touches a product claim, a legal line, a price or a usage right. Do not act on these or rewrite them.
Every comment number must appear in exactly one of the three sections.

Checkpoint 2. Add up the comment numbers across the three sections. They must match your list exactly. Then make each decision in "You decide", yourself, and paste them back: My decisions: [for each conflict, the choice and one line of reason]. Send the "For the account team" section to the account lead. The account team takes claims and legal lines to the client's legal team.

Part 4: Prompt 3, tasks by role

Turn the numbered direction and my decisions into a task list for each role: designer, production artist, retoucher, copywriter. For each task give: the task, the comment numbers it covers, "Round: [number]" and "Due: [date]". Write "No tasks this round" for a role with none. Do not add tasks that are not in the direction or my decisions. Do not include the "For the account team" items.

Checkpoint 3. Check that every task traces to a direction item or a decision. Fill the round and due-date placeholders. Remove anything the model added on its own initiative.

Part 5: Prompt 4, the check table

Make a table with one row for each of the original comments, in number order. Columns: number, role label, the comment (quoted), and "Where it went", which must be one of: "task" (with the task), "decision" (with the decision), or "not actioned" (with the reason). End with a count of each of the three outcomes and the total.

Checkpoint 4. Compare the table with your original list. The total must equal your comment count and each count must match what you can see in the table. Any "not actioned" row needs a reason you agree with. Only then send the task lists out.

Real Example: Round 2 on the Lantern Hero Layout

Invented, with role labels instead of names. Eight comments:

1. CD: Move the headline lower. It crowds the product.
2. CD: The product shot feels flat. Pull more contrast.
3. Client brand lead: Make the logo bigger.
4. Client brand lead: Headline should be smaller so the product has room.
5. CD: Keep the logo at the brand-standards size. The layout is tight.
6. Client legal: Add the long-form legal line to the footer.
7. Account lead: Can the headline say "the best in town"?
8. Account lead: Client also wants the layout in 1080 x 1350.

Prompt 1 groups them: Headline (1, 4, 7), Image (2), Logo (3, 5), Footer (6), Format (8). Count: 3 + 1 + 2 + 1 + 1 = 8, the total you pasted. Checkpoint 1 passes.

Prompt 2 gives a direction of 3 items. Item 1 lowers and reduces the headline (comments 1 and 4). Item 2 adds contrast to the product shot (comment 2). Item 3 prepares the 1080 x 1350 version (comment 8). "You decide" holds comments 3 and 5, because the client brand lead and the CD disagree about the logo size. "For the account team" holds 6 (legal line) and 7 (the "best in town" claim, which is not on any approved claims list). Count: 4 direction comments (1, 4, 2, 8) + 2 conflict + 2 account team = 8. Checkpoint 2 passes. You decide: Logo stays at the brand-standards size (comment 5). Ask the client brand lead to confirm.

Prompt 3 gives three tasks. Designer: lower and reduce the headline (comments 1, 4), Round 3, Due [date]. Retoucher: add contrast to the product shot (comment 2). Production artist: build the 1080 x 1350 version (comment 8). Copywriter: no tasks this round. (1080 x 1350 is a 4 to 5 ratio, which the production artist should check against the publisher's spec sheet.)

Prompt 4 maps all eight: comments 1, 2, 4 and 8 go to "task" (4), comments 3 and 5 go to "decision" (2), and comments 6 and 7 go to "not actioned" (2), with the reason "goes to the account team". Totals: 4 + 2 + 2 = 8.

What to Do When It Breaks

• A comment disappears between steps → The counts catch it. Reply naming the missing number and ask for the section to be redone.
• The model resolves a conflict itself → Reply Return comments [numbers] to "You decide" and do not choose.
• The model rewrites a client's wording → Compare the quotes against your list. Quotes must be exact. Paste the original line back.
• The model adds a claim or a legal line to a task → Delete it and send the item to the account team.
• The conversation gets long and the model loses track → Start a new chat. Paste the numbered comments, your decisions and the direction, then run prompts 3 and 4 again.

Variations

• Simpler version: Run prompts 1 and 2 only, and write the tasks yourself.
• Extended version: Run the chain on each round and keep the check tables together as a record of what was decided.

What to Do Next

• This week: Run the chain on your next review round.
• This month: Save the four prompts so you can paste them in order.
• Advanced: Pair it with the deliverables digest so the new tasks get due dates.

Advanced guide for art directors, agency or in-house. Any images or comps made with AI are for internal concepts only unless legal and the client have cleared the tool for published work.