← All AI ROI products
Founder pricing
AI product · AI and automation

Prompt Library

Turn the AI tools you already pay for into consistent results: a tested library of prompts for the jobs your team does every week, written in your house style.

About 4–8 hours of your time, plus short tests by your team (estimate)Your team's recurring tasksRun it yourself or with Geoff
What it answers

Four questions it answers

  1. 1

    Which jobs are worth a prompt?The frequent ones where quality matters.

  2. 2

    What does a good prompt look like here?Written for your business and your style.

  3. 3

    Does it work?Each prompt tested on real examples by the people who do the work.

  4. 4

    How do we keep it up to date?A simple owner and review routine.

Founder pricing. The method, the calculation spreadsheet with 14 built-in checks, and a complete worked example are built and tested. The first client for this product gets the founder price, in return for permission to measure the results and, if they are good, write them up as a case study. After that, the price goes up. Book a call to take the founder place.

What you get

What you get

A prompt library and its register

Named, numbered prompts for your recurring tasks, each with when to use it and an example, and a register of every prompt's owner, version and review date.

Your house style

Tone, words to use and avoid, and formats, built into the prompts.

Test results

Each prompt tried on real examples, with what good output looks like.

Safe-use notes

What must never be pasted into an AI tool, prompt by prompt.

A team guide

A short guide so new starters can use the library on day one.

A library agent

On guided and done for you: only prompts that passed testing served to your team, which work logged week by week in your own accounts, and fixes drafted for each prompt's owner to test. Nothing leaves your business, and no decision changes, without one of your people approving it.

Worked example

A Bristol cleaning and facilities firm

A fictional 62-person contract cleaning business whose office team writes quotations, complaint replies, client reports, incident write-ups and job adverts every week. 12 written tasks were listed and ranked; disciplinary letters were kept out of AI by decision.

9 prompts passed testing on real examples and went live. The complaint reply failed its first test (it offered a credit nobody had agreed) and was fixed; the tender answer invented an accreditation and stays a draft.

Read the full example

Fernmoor Facilities LtdFictional · Worked example
Live promptspassed testing on real examples
9
Usabletests of live prompts: ready or light edits
100%
Time saveda month, counted
28 h
Valuea year of time released
£9,236
IllustrativeAll names and figures are invented. The bigger gains are consistency, fewer mistakes and safe use; the value counts time only.
How it works

Five stages

  1. Collect

    List the written tasks your team repeats every week.

  2. Write

    The AI drafts a prompt for each, in your house style.

  3. Test

    Try each one on real examples and refine it.

  4. Share

    Publish the library and agree who keeps it current.

  5. Keep

    Your AI assistant serves the tested prompts, logs which work and drafts fixes for each owner to test.

Three ways to buy

Do it yourself, with Geoff, or done for you

The same method and the same spreadsheet in every option. You choose how much of your own time goes in.

Self-implement kit

Run it yourself

For owners comfortable with a spreadsheet and an AI assistant.

  • The AI skill
  • The calculation spreadsheet
  • A worked example
  • The buyer's guide
Book a call
Guided implementation

Run it with Geoff

For owners who want a second pair of eyes on the data and the analysis.

  • Everything in the kit
  • Two working sessions
  • Geoff reviews and tests your prompts
  • Your library, reviewed
  • The weekly agent, set up with you in your own accounts
Book a call
Done for you

Geoff runs it

For owners who would rather hand over the data and get the answer.

  • Geoff builds and tests the library in your own systems, from your examples
  • House style guide and team guide
  • A one-hour handover session
  • The agent's first four weekly runs in your account, each note reviewed with you
Book a call
What you'll need

Data you already have

  • A list of the documents, emails and reports your team writes regularly
  • Two or three good examples of each
  • Access to an AI assistant
  • About 4–8 hours of your time over a week or two, plus about 15 minutes per test from your team (estimate)
  • For the agent (guided and done for you): an AI assistant your business already uses or approves, that can run a scheduled task and read your library's folder and use log; we confirm this with your IT before you commit
Which AI?

Designed for Claude, in an open format

It is built on the open Agent Skills standard and designed for Claude (any plan, with code execution switched on). A paste-in version is included for ChatGPT, Gemini and Microsoft Copilot; it has not yet been tested on them.

The calculations live in the spreadsheet, not in the AI, so the numbers don't depend on which assistant reads them.

Get consistent results from AI

Book a short call to talk through the tasks your team repeats every week.

Book a call
Geoff@fox.coach · 07770 683883 · fox.coach