Start with a small task, give the facts and tell the assistant what a useful result looks like. You can practice every lesson here for free.
LESSON 1 · GIVE THE ANSWER A JOB
Four things to put in your request.
Task and audience. Say what you need and who will read it: “Write a friendly email to neighborhood volunteers.”
Your facts. Add the time, place and details it should use. Leave out private information the task does not need.
Shape and length. Ask for a subject line, a checklist or a short explanation. Set a word limit when it helps.
Limits. Say “Use only these notes. Mark missing details with [confirm].” Check that it follows this instruction.
These techniques work with the OpenAI models in Junction and can also help when you use ChatGPT directly.
LESSON 2 · WATCH A VAGUE REQUEST BECOME USEFUL
From “help with my event” to a usable draft.
A vague request leaves the assistant guessing. This practice example supplies the job, the facts and the limits:
Write a friendly invitation to local volunteers. Use only these practice details: a book swap on Saturday at 11 a.m. in the community room. Include a subject line and one clear next step. Keep it under 100 words. Mark missing details with [confirm]; do not invent a price or refreshments.
What improved: the audience, event details and format are now explicit.
What still needs checking: the calendar date, address and contact person were not supplied.
Try it: write your own prompt for a three-item errand checklist. Add the errands, the format and one limit.
Check your practice
One possible answer: “Turn these errands into a three-item checklist: return library books, collect a parcel and buy bread. Use only these errands. Do not invent opening hours or a route.”
LESSON 3 · IMPROVE THE RESULT WITHOUT STARTING OVER
Review, then ask for one clear change.
Check against your notes. Mark dates, names, prices and claims that need confirmation.
Be specific about the revision. “Keep the facts; shorten this to 80 words and put the next step first.”
Use a small task first. Try Luna for a short draft. Move to a stronger model when the task needs more reasoning.
Know this studio’s limits. It uses the text you supply; it does not browse current websites, book anything or send your draft.
Make the final decision. A fluent answer can still be wrong. Verify before sending, publishing or relying on it.
For example, a 2-credit request uses 10¢ of the $5 pack or about 6.53¢ of the $49 pack. A 5-credit request uses 25¢ or about 16.33¢ respectively.
Package prices are before checkout tax. Credits cover OpenAI processing, safety checks and the Junction service.
Junction AI credits and ChatGPT subscriptions are separate purchases. Free Junction lessons and local document tools need neither.
How model pricing affects your quote
Different models have different input and output costs. Junction reserves a bounded response budget and shows a fixed credit quote for your approval. OpenAI counts text in tokens, which are pieces of words; output billing also includes any reasoning tokens. Our credit price is a service price, not a direct resale at OpenAI’s token rate.
Text rates checked September 13, 2026, in US dollars per million input / output tokens:
Timing is an estimate. The ranges above are planning guidance, not measured speed benchmarks or guarantees. Longer output, reasoning and provider demand affect the wait.
Checks take time, too. The service checks your request and the result before returning a draft.
Keep one request. The studio shows elapsed time. Use its recovery buttons if the connection drops; they keep the same request reference.
There is a limit. The text provider call has a 90-second limit and images have 150 seconds, with additional time for checks. An unfinished reservation expires after five minutes.
Get to a useful draft sooner. Ask for a shorter result, remove irrelevant source text and split a large job into smaller tasks.
Save your request to the private account queue, even when your AI credit balance is zero.
The support team and its AI assistants can review queued work. You give permission before your prompt is saved.
Queued replies are complimentary. Waiting and team completion spend no customer credits; failed direct generation releases its hold automatically.
Expiry pauses on active, unexpired AI credit grants while you have an open request. That waiting time is added back when your last request closes.
Check your request number and status in your account. Cancel an unfinished request when you no longer need it.
Image requests wait for an available image service. A text reply is not presented as a finished image.
We apologize for interruptions. We do not guarantee a completion time. Credit purchases during an interruption require your acknowledgment that generation will be available later; buying does not change queue priority.
Original Junction lessons sponsored by CrownThriveU. AI generation uses OpenAI models; the sponsorship is separate from OpenAI. For more prompting examples, see OpenAI’s prompting guide.