A fast idea sprint separates invention from judgement. ChatGPT first produces more options than you need; you then reject weak options with three practical filters. Mixing those jobs too early usually produces ten safe variations of the same thought.

By the end, you will have 30 ideas and eight shortlisted calendar entries—not 30 posts you must publish. The shortlist does not predict reach, sales, or audience response.

// concept

Seed the Sprint With Useful Constraints

Set a 15-minute timer: four minutes to fill the brief, four minutes to generate and inspect 30 ideas, and seven minutes to cull them to eight. Prepare the brief before opening a new ChatGPT conversation. A seed has four parts:

Seed	What to provide	Weak version
Niche	A narrow subject you can discuss truthfully	“Lifestyle”
Audience	Situation, need, language, and main platform	“Pakistanis”
Formats	Formats you can actually produce this week	“Anything viral”
Past winners	Your own logged posts whose response met your stated goal	A competitor's popular posts

For a first sprint, write none yet under past winners. Later, include only your own logged winners: title, format, audience question, and the metric used. Check the platform analytics or content log rather than relying on memory.

Copy this prompt and replace every bracketed field:

// prompt — copy me24 lines

Act as an idea-room facilitator, not a trend forecaster.

NICHE: [narrow subject]
AUDIENCE: [specific people, situation, language, platform]
GOAL: [teach, start a useful conversation, or explain an offer]
FORMATS I CAN MAKE: [for example: talking-head reel, screen recording,
single graphic, carousel, text post]
FIRST-PARTY QUESTIONS OR OBSERVATIONS:
- [redacted question/observation 1]
- [redacted question/observation 2]
- [redacted question/observation 3]
PAST WINNERS FROM MY OWN LOG:
- [post, format, observed signal and date—or “none yet”]
CAPACITY THIS WEEK: [hours, recording access, editing ability]
BOUNDARIES: [claims I cannot make, topics to avoid, permissions missing]

Generate exactly 30 distinct content ideas. Spread them across five buckets:
how-to, mistake diagnosis, comparison, behind-the-scenes, and audience Q&A.
For each idea return: number, working title, one-sentence angle, format,
evidence needed, and estimated production effort (low/medium/high).

Do not invent trends, customer quotes, results, prices, or facts. Mark an idea
“research required” when the supplied material cannot support it. Avoid duplicate
angles and avoid promises about reach or sales. Output a markdown table only.

The 30 rows are raw material. ChatGPT may repeat ideas under new titles or suggest unavailable evidence. Quantity creates contrast; curation creates the usable bank.

// concept

Cull 30 Ideas to Eight With Three Filters

Apply the filters in this order. A single No kills the idea for this sprint:

Audience fit: Does it answer a supplied question or a clearly stated need of this exact segment?
Evidence available: Can you support every factual point using a link, a product record, or your own first-party observation?
You can make it: Can you produce it with this week's time, device, access, and permissions?

Do not lower the evidence bar to reach eight. If only six pass, request replacements constrained by the rejection reasons. If 14 pass, choose a varied eight—not simply the first eight rows.

Use this correction prompt after the cull:

// prompt — copy me6 lines

These ideas failed my review:
[paste killed idea numbers and exact reasons]

Replace only those rows. Keep the original niche, audience and boundaries.
Every replacement must pass: (1) audience fit, (2) evidence available,
(3) producible within my stated capacity. Do not rewrite ideas that passed.

// concept

Put the Eight Into a Calendar

Use calendar fields for publish date, idea, format, source/evidence, production slot, and status. Attach evidence before drafting.

Refresh after four weeks or when you have useful first-party observations. Feed actual logged winners and misses into the next sprint, naming the measured signal without treating it as proof of future performance. New evidence should change the seed.

// worked_example

Worked Example

This hypothetical sample uses a Lahore home baker teaching beginners on Instagram. They can make two phone videos and two carousels weekly. Sample first-party questions are: “Why did my cake sink?”, “Can I bake without an electric oven?”, and “How do I price one batch?”

The prompt is filled with those inputs, none yet for past winners, four hours of capacity, and no unsupported food-safety or profit claims. ChatGPT's sample output is:

Three reasons a cake centre sinks; 2. Pot-versus-oven setup; 3. Batch-cost worksheet; 4. Measuring without a scale; 5. Buttercream in summer heat; 6. Behind one test bake; 7. Oven thermometer check; 8. Ingredient substitution decision tree; 9. Packing a six-cupcake box; 10. Beginner piping drill; 11. Gas-pressure baking diary; 12. Customer flavour poll; 13. Cake-pan size comparison; 14. Cooling-rack alternatives; 15. One-bowl mixing order; 16. Food-colour test card; 17. Failed sponge diagnosis; 18. Recipe source-check routine; 19. Kitchen cleaning checklist; 20. Photographing crumbs in daylight; 21. “Is margarine always cheaper?”; 22. Three famous Lahore bakery secrets; 23. Guaranteed-profit cake formula; 24. Customer reaction compilation; 25. Industrial-oven tour; 26. Frosting texture close-up; 27. Freezer-storage explainer; 28. Order-form walkthrough; 29. Weekly ingredient prep; 30. Ask-a-baker Q&A.

The first output is not accepted as-is. Here is the required five-idea diagnosis:

Idea	Audience fit	Evidence available	Can make it	Decision
1. Sinking cake	Yes: supplied question	Yes: own test bake plus a verified recipe source	Yes: phone demo	Keep
3. Batch-cost worksheet	Yes	Yes: current ingredient receipts	Yes: carousel	Keep
22. “Famous secrets”	No: not tied to the beginner questions	No: implies inside knowledge	Yes	Kill—unsupported premise
24. Customer reactions	Possibly	No permission or approved quotes	Yes	Kill—consent missing
25. Industrial-oven tour	Weak for home bakers	Possibly	No access	Kill—capacity mismatch

Idea 23 is a performance promise, so it is killed immediately. The correction prompt requests four replacements using only supplied questions, available records, and phone/carousel formats. After all rows are checked, ideas 1, 2, 3, 4, 6, 11, 17, and 28 enter the sample calendar. Each evidence cell names test notes, current receipts, or a primary recipe reference. This 30→8 bank makes no performance claim.

// failure_cases

Failure Cases to Diagnose

6 cases to diagnose

Thirty title variations
several rows share the same teaching point. Group by underlying question and keep one; request replacements from missing buckets.
Trend claims with no trace
“trending now” appears without a source and date. Remove it or verify the trend on the relevant platform before scheduling.
A winner borrowed from a competitor
another creator's result is not your first-party winner. Keep it in a separate ethical swipe file, not this seed.
Evidence named too vaguely
“Google” or “research” cannot support a claim. Attach the exact official page, receipt, test note, or redacted observation.
Production fantasy
the model proposes street interviews when you have two hours and no consent process. Kill it or change the format to one you can make.
Eight nearly identical keeps
the filters pass, but the calendar repeats one format or question. Select a more useful mix from the passing pool.

// pakistan_angle

Pakistan Angle

Define the Pakistani audience below national level. “Karachi university students using mobile data” and “Peshawar shop owners posting in Pashto and Urdu” have different questions, language choices, and production realities. Do not add chai, traffic, wedding, or rishta references as a shortcut to relatability. Seed ChatGPT with redacted questions your audience actually asked, and keep religion and politics outside a brand's idea sprint unless the brand has a deliberate, reviewed policy for them.

Capacity must reflect load-shedding, mobile-data cost, and your real equipment. A phone creator can shortlist screen recordings, voiceovers, text posts, and Canva graphics, then upload when connectivity is stable. Never paste CNICs, phone numbers, customer addresses, unredacted WhatsApp messages, or identifiable payment screenshots into ChatGPT. Check current ChatGPT Data Controls before customer work; redaction and permission remain required.

// hands_on

Hands-On Exercise

6 steps

Choose one narrow niche and describe one Pakistani audience segment in a single sentence.
Collect three redacted first-party questions and list the formats you can produce this week.
Add verified entries from your own content log under past winners, or write none yet.
Run the full ideation prompt with a 15-minute timer and save all 30 rows.
Add three filter columns. Mark every row Yes or No and write a reason for each rejection.
Select eight varied passing ideas and place them in a calendar with evidence and production slots. Done means the original 30-row output and the eight-row calendar both exist, every shortlist row passes all three filters, and no evidence cell is a placeholder.

// completion_rubric

Completion Rubric

5 checks — tick as you verify

0/5

// sources

Sources

3 official sources — check every claim yourself

// check_yourself

Check yourself

4 questions · answers and options are taken word-for-word from this course

0/4

1 / 4 · diagnose
Your work shows this failure mode: “Thirty title variations.” The lesson describes it like this: “Several rows share the same teaching point.” What does the lesson tell you to do about it?