Brainstorming: Divergent vs Convergent Prompt Modes
Willowpath Pet Memorial handles 220 cremations a month at an average of 310 dollars, and its founder, Marcus Bell, has a structural problem: a grieving family rarely becomes a repeat customer. Revenue cannot come from loyalty, so it has to come from services that a family wants in the worst week of their year, and from referrals that follow good care. Marcus needs a new service line, and he is about to make the classic mistake of asking one prompt to both invent ideas and pick the winner.
Brainstorming has two modes, and ChatGPT behaves differently in each. Divergent mode widens: it generates options with no judgement. Convergent mode narrows: it weighs options against criteria and returns a shortlist. Ask for both in one prompt and the model quietly filters while it generates, which is how you get ten safe ideas instead of thirty. This guide is a decision table for seven brainstorming goals, the mode each one fits, a copy-paste prompt per mode, and the cost of switching at the wrong moment. To keep a multi-step reasoning trace visible, see the chain of thought prompt, and to give the model a defined point of view, use role prompting.
Brainstorm Mode Decision Table
| Brainstorm goal | Mode to use | Why it works | What it costs |
|---|---|---|---|
| Fill the board with raw options | Divergent, no filtering | Produces volume and unexpected angles | Plenty of duds; looks messy on purpose |
| Choose three to prototype | Convergent with criteria | Turns a long list into a decision | Can close a promising idea too early |
| Learn what families actually want | Convergent on evidence | Grounds ideas in real interviews | Slow; needs access to customers |
| Invent a service nobody offers | Divergent with constraints | Pushes past the obvious competitors | Risky; some ideas will not fit the brand |
| Kill weak ideas fast | Convergent, adversarial | Protects attention from pet concepts | Can discard a slow-burn winner |
| Price and package the shortlist | Convergent with a price frame | Converts ideas into sellable offers | Feels uncreative; the fun part ends |
| Name and position the new line | Divergent on language | Finds words that match a gentle tone | Can drift into vague marketing language |
One rule keeps the two modes from contaminating each other: never let the same prompt both create and judge. Open wide, then close hard, in that order. If a session starts narrowing in the first message, the model is guessing your taste instead of exploring the space, and the safe ideas it returns are the ideas you already had.
Choosing Divergent or Convergent
Go divergent when the space is empty. If Willowpath has no clear list of add-on services, or the existing list is three items long, that is a divergent problem. Ask for quantity, forbid judgement, and allow weird ideas to stay in. Ten of the thirty will be unusable and that is the point.
Go convergent when you have to defend a choice. The moment Marcus has to budget for one service, the question changes from “what could we offer” to “which one, and why”. That is convergent work, and it needs explicit criteria so the reasoning is auditable later.
Go divergent with constraints for novelty. “A memorial service a family could gift to a friend” is a constrained divergent brief. The constraint does the creative work; without it the model returns the same urn options every competitor already sells.
Go convergent on evidence before you build. The cheapest way to avoid building the wrong service is to run a convergent pass against real interview notes from fifteen families, scoring each idea by how often the need came up and how emotional the moment was.
Prompt Blocks for Each Mode
Divergent pass. Volume first, judgement later, no price talk yet.
You are brainstorming value-added services for Willowpath Pet Memorial, a pet cremation and memorial company handling 220 cremations a month.
Mode: divergent. Generate 30 distinct service ideas.
Rules:
- Do not rank, filter, or comment on any idea.
- Group them into five themes and label each theme.
- Include at least five ideas that could work as gifts between friends.
- Include three ideas a competitor is unlikely to offer.
- One line per idea, no pricing, no marketing language.
After the list, state which single constraint would most change your output.Convergent pass. Score the list against criteria that matter to the business.
Score the 30 Willowpath service ideas against five criteria:
1. Fits the family's emotional moment.
2. Gross margin above 50 percent.
3. Can be delivered by a team of six without new licences.
4. Generates referrals or gifting.
5. Feasible to launch within 90 days.
For each idea, give a score from 1 to 5 per criterion and a total.
Return the top five and, for each, one sentence on why it won.
State the two ideas you rejected despite a high score, and why.Adversarial kill pass. The convergent step that removes the comfortable choices.
You are a skeptical operations lead at Willowpath. Take the top five ideas and attack each one.
For every idea, list:
- The single most likely reason it fails in practice.
- The hidden cost the pitch omits.
- The one family it would alienate.
End with a verdict: keep, reshape, or drop, and the cheapest experiment that would prove you wrong.Pricing frame pass. Turns an idea into an offer with a number attached.
For the two surviving Willowpath ideas, design a product page outline.
Include:
- A plain name and a one-line description a grieving family could read quickly.
- What is included, and what is not.
- Three price tiers and the reasoning behind each tier.
- When it is offered in the current 11-step customer journey.
Keep the language gentle and factual; avoid the word "upsell".Naming pass. A short divergent round on language only.
Generate 20 candidate names for one Willowpath service: a memorial keepsake made from a pet's collar.
Rules:
- No puns, no words that imply replacement of the pet.
- Each name under three words.
- Group by tone: warm, plain, ceremonial, nature.
- Do not score or recommend; just produce the list and stop.Switching Modes: The Costs
Divergence has a billable cost: thirty ideas take longer to read than five, and the team will want to debate the strange ones. Cap the divergent round at one pass and a fixed count, then move on. The other cost is emotional; for a company this close to grief, an unconstrained brainstorm will surface ideas that are technically clever and clinically cold, so a human should read the raw list before anyone puts it in a deck.
Convergence has a different cost: premature closure. A scoring prompt rewards ideas that fit the criteria and quietly buries the one idea that would have needed a new criterion to shine. The repair is to always ask the model to name one high-scoring rejection and one low-scoring idea worth a second look, so the shortlist stays honest. For a service line, the reliable loop is two divergent passes on separate days, one convergent (scoring) pass, one adversarial pass, then a pricing pass. To reuse the same set-up across many sessions, keep the brief inside a custom GPT persona so the brand’s gentle tone is applied every time.
Sources & Notes
Prompts tested with ChatGPT on the GPT-4o and GPT-4.1 models in 2026. Willowpath Pet Memorial, Marcus Bell, the 220 monthly cremations, and the 310 dollar average are an illustrative scenario, not a real business. Any service touching bereavement should be reviewed against local consumer-protection and funeral-services rules before launch, and priced only after real cost accounting.
The core discipline is sequence, not cleverness. Divergent prompts raise the ceiling of what is possible; convergent prompts decide what is affordable, deliverable, and kind. Run them in the wrong order and the model will happily hand you a tidy set of the most obvious ideas. Run them in the right order and it will occasionally hand you one that a grieving family would genuinely want. Category mapping follows the guunn.com AI Prompts plan (Prompt Engineering = 961).
