Turning one good prompt into everyone's
The gap between two colleagues
In every team that has used these tools for a few months, someone is getting far more out of them than everyone else. It is almost never because they know a secret model. It is because they have accumulated twenty or thirty briefs that work, and they reuse them without thinking about it.
That accumulation is an asset and it is trapped in one person's head. Getting it out is the highest-return organisational move in this whole course, and it costs one shared document.
What an entry looks like
Not a prompt on its own. A prompt with the three things that make it usable by somebody else.
Use: first draft of a refusal letter for a grant application.
When to use it: routine refusals only. Not for appeals, not where the applicant has complained.
Prompt: [the brief, with blanks for the specifics]
What it gets wrong: invents a named contact if you do not supply one; over-apologises; sometimes states an appeal window — always replace with the real one from the scheme document.
Last checked: March, on our approved tool.
The "what it gets wrong" line is the one that matters most and the one nobody writes. It converts the entry from a shortcut into teaching. A colleague using it now knows where to look, which is the difference between a faster draft and a faster mistake.
The date matters for a reason people underestimate: models change under you. Providers update them, and a brief tuned to last year's behaviour can quietly stop working. A dated entry can be re-tested; an undated one just erodes until people stop trusting the file.
Where to keep it
Wherever your team already looks. A shared document, a wiki page, a channel with pinned messages. The tool matters far less than the habit, and a plain text file that people actually open beats a beautiful system that they do not.
If your tools support it, promote the best entries into the product itself — saved prompts, custom instructions, a project or workspace with your standing context, a "gem" or custom assistant with your house style and reference documents attached. These are convenient and they have one real drawback worth knowing: they are invisible. A colleague using a shared assistant cannot see the instructions shaping its answers, and when those instructions go stale nobody notices. Keep the plain text version as the source of truth and treat the built-in feature as a convenience layer over it.
The three entries worth writing first
Start narrow. A library of two hundred entries that nobody reads is worse than five that everybody uses.
- Your most repeated document. Whatever your team writes weekly. One good brief here saves more hours than everything else combined.
- Your house style. Two or three of your best real documents, stripped of identifying detail, as examples. The technique from module two, stored once for everyone.
- Your checking pass. The list of what must be verified before anything in your field leaves the building. Not a prompt at all — a checklist — and it belongs in the same file because that is where people will be at the moment they need it.
Two rules that keep it honest
Nothing confidential goes in the library. Examples get stripped before they are stored. A shared prompt file is read by more people than the original document ever was, and it tends to outlive the person who wrote the entry.
Anyone may add, someone must prune. Twenty minutes a quarter deleting entries nobody has used and re-testing the ones people rely on. Without pruning, the file grows, the average quality falls, people stop trusting it, and the knowledge goes back into individual heads where it started.
Why this beats training
Most organisations respond to AI by running a session. Attendance is good, nothing changes, and six months later the same three people are the only ones getting value.
A library changes what a new colleague can do on their first day. It captures what actually worked here, in this job, with these documents, and it improves every time somebody finds a better version. That is the difference between an event and an institution, and it is available to a team of four with a shared file and no budget at all.
The one thing to keep
A prompt that worked is an organisational asset, and writing it down with its context and its failure notes is what turns one person's fluency into a floor everybody stands on.
Before you move on
A team keeps a shared file of good prompts. Six months later most entries no longer produce good results, and people have stopped using it. What was most likely missing?
Pick the one you would defend. Nobody sees your answer.