Prompt LibraryThe NOVA Prompt Standard

Professional AI prompts

The NOVA Prompt Standard: 12 Quality Checks for Professional AI Prompts

Every NOVA prompt passes 12 quality checks before going on sale. Together, the checks make an AI prompt give the same quality of answer on every run: the right method, the right structure for your situation, no made-up numbers, clear rules for hard cases and a final check of the answer.

That is what we mean by a professional, enterprise-grade AI prompt: the work is checked the way a senior expert’s work is checked inside a large company.

How Each AI Prompt Is Scored

Each check is scored 0, 1 or 2. A NOVA prompt goes on sale only when every check scores 2, and is scored again every time the prompt changes. The same rule applies to the generators and to every prompt pack.

0Missing. The prompt leaves this decision to the AI.
1Partly there. Present in some places, or stated without a rule the AI can follow.
2Working. Written as an instruction the AI follows on every run.

The 12 quality checks

Each check below quotes the instruction as written in the AI PowerPoint generator, and shows the result for Ridgeline Health, the made-up hospital group in that product’s worked example.

01

Before the work · Persona and principle

Thinks like a senior expert

What the check requires. The prompt names a senior expert with a track record, plus the one working rule that changes the result.

Why it matters. The working rule decides what the answer leads with and what it leaves out. A job title alone decides neither.

The instruction, as written in a NOVA prompt

a strategy consultant who has led board engagements for fifteen years ...
Your operating principle: a deck exists to get one decision, so the
recommendation reaches the room by slide 2, every title states a finding,
and any slide that does not prove the argument is cut.

A typical prompt

“Act as an expert” sets a tone, not a standard. Nothing decides what goes first or what gets cut.

In the worked example

The recommendation, operating margin from 0.6% back to 3.2% by FY28, is on slide 2, before any evidence.

02

Before the work · Intake

Asks you for the facts first

What the check requires. The prompt asks for the facts that decide the result, repeats them back and confirms them before work starts. If you have none, the prompt can research them with cited sources, suggest labelled estimates, or leave marked gaps: your choice, and the work never stops.

Why it matters. Most weak AI answers start with a fact the AI filled in silently. Asking first is cheaper than rewriting later.

The instruction, as written in a NOVA prompt

If a decisive input is missing, or I tell you I have none, ask once and offer:
 (a) I SUPPLY IT - you wait for it.
 (b) YOU RESEARCH AND PROPOSE IT - ... cite each one with its source and date ...
 (c) YOU LEAVE A GAP - build the full output with [NEEDS INPUT] placeholders.

A typical prompt

The model starts writing from one line of request and fills every gap itself.

In the worked example

The brief and the FY23–FY25 finance and operations figures, restated in seven lines and confirmed before any analysis.

03

Before the work · Routing

Fits your situation

What the check requires. Each different kind of case gets its own method and structure, from a table inside the prompt.

Why it matters. A board paper, a sales pitch and an investor deck each need a different argument. One method for all of them fits none.

The instruction, as written in a NOVA prompt

ROUTER  (deck type -> spine | analyses | blueprint)
- Strategy/board -> SCR + answer-first | problem-framing, situation,
  competitive, options, business case, risk | Strategy blueprint

A typical prompt

Every deck gets the same title, agenda and bullet slides, whether it is a board paper or a sales pitch.

In the worked example

Matched to the board presentation structure (situation, problem, answer), with six analyses and a 12-slide plan. One of ten presentation types.

04

During the work · Written-out methods

Uses the right method

What the check requires. Textbook methods, with the rule or formula written into the prompt.

Why it matters. When the formula is in the prompt, two runs on the same data give the same answer.

The instruction, as written in a NOVA prompt

Business case: ... ROI = (total benefit - total cost) / total cost;
payback = months until cumulative net benefit turns positive; NPV at my
discount rate, or [NEEDS INPUT] if none is given

A typical prompt

The model picks a method on each run, so two runs on the same numbers disagree.

In the worked example

$9M of one-time investment against $55M a year of recurring operating income, with a 14-month payback, each with its arithmetic in the notes.

05

During the work · Evidence rules

Never makes up numbers

What the check requires. Every figure comes from your material, a cited source or its own arithmetic. Estimates are labelled, gaps are marked, and nothing is passed off as fact.

Why it matters. A number nobody can trace is the fastest way to lose a board, an investor or a client.

The instruction, as written in a NOVA prompt

Never invent a number, a quote, a client, a benchmark or a source ...
A figure the argument needs but my material lacks becomes
[NEEDS INPUT: what, and who could supply it].

A typical prompt

Missing figures are filled with plausible numbers that nobody can trace.

In the worked example

Contract labour at 19.4% of nursing spend, against a peer median of 7.6%: graded Evidence. “Medicare Advantage rates recover at the FY27 renewals”: graded Assumption, and labelled as one.

06

During the work · Output specification

Delivers a ready-to-use format

What the check requires. The prompt spells out the exact structure of every deliverable, so the result arrives in a form your reader can use.

Why it matters. The structure decides whether the result can be used as it is or has to be rewritten.

The instruction, as written in a NOVA prompt

SLIDE [n]
  Action title:    [the single takeaway, as a full sentence]
  Job in argument: [why this slide exists]
  Visual:          [the framework/chart + its encoding - pick via the chart selector]

A typical prompt

Slide titles are topic labels: “Financial Overview”, “Payer Mix”.

In the worked example

“Payer Mix” became “Medicare Advantage grew from 19% to 34% of volume in four years, and is forecast to reach 41% by FY28 on current contracts.”

07

Your control · Audience

Written for your reader

What the check requires. The prompt sets, for each kind of reader, what to lead with and what to cut.

Why it matters. The same finding is presented differently to a CFO, a board and a customer.

The instruction, as written in a NOVA prompt

AUDIENCE OVERLAY  (lead with / cut)
- Board   lead: risk, governance, shareholder value
          cut: execution minutiae

A typical prompt

One deck for every room: the board, the CEO and a customer read the same slides.

In the worked example

Board overlay: the ask names the $9M envelope and three board checkpoints, in FY26 Q2, FY27 Q1 and FY27 Q4.

08

Your control · Edge cases

Warns you about problems

What the check requires. Written rules for the cases where the standard method breaks.

Why it matters. Standard methods break on thin data, contrary evidence and unusual cases. Written rules say what happens then.

The instruction, as written in a NOVA prompt

- Evidence against the objective: say so at STEP 2 and propose the deck
  the evidence supports, including "do not proceed".
- Thin or no raw material: apply NO INPUT? THREE WAYS FORWARD at STEP 0 ...

A typical prompt

Asked for a recovery plan, the model writes one, whether or not the numbers support it.

In the worked example

Checked before building: full data, one decision-maker, 12 slides. No special case applied, and the prompt said so.

09

Your control · Stops for your approval

Checks with you before moving on

What the check requires. The prompt stops wherever you need to decide, instead of producing one long answer.

Why it matters. Approving the direction before the work starts costs minutes. Rewriting a finished result costs hours.

The instruction, as written in a NOVA prompt

Using the ROUTER, name (a) the deck type, (b) the narrative spine,
(c) the analyses to run, (d) the slide blueprint to target ...
Confirm in 6 lines. PAUSE.

A typical prompt

One long answer. A wrong deck type is discovered on slide 12.

In the worked example

Three stops: the route, the graded analysis and the storyline. Nothing was built on a choice the CFO had not approved.

10

Before handover · Quality gate

Checks the answer for you

What the check requires. Named checks the AI runs on its own work, fixing every failure before handover.

Why it matters. An error the prompt catches never reaches your reader.

The instruction, as written in a NOVA prompt

[ ] Every number is sourced (my material, or a cited source with its date) or shows
    its arithmetic; every estimate is tagged ASSUMPTION ...
[ ] Totals, bridges and percentages recompute exactly
[ ] A clear, specific ask/close: the decision, the owner, the date

A typical prompt

No check. The first reviewer is the board.

In the worked example

Eleven checks, nine passed first time. Fixed: a topic-label title, and $55M shown without its arithmetic (2.6 points × $2.1B = $54.6M).

11

Before handover · Build and hand-off

Gives you a finished file

What the check requires. A real Word, Excel or PowerPoint file in Claude, built to a written specification, with a text version in other AI tools and the next step named.

Why it matters. A finished file saves the time spent copying text into a document, a workbook or a deck.

The instruction, as written in a NOVA prompt

Charts as native, editable chart objects with their data inside the
chart - never pictures. ... Then tell me: to present it out loud, paste
the final deck into the NOVA Presentation Script Generator for the
word-for-word talk, timed to the slot.

A typical prompt

Text to paste into PowerPoint by hand, slide by slide.

In the worked example

The twelve slides below: an action title on every slide, seven charts, and the recommendation on slide 2.

12

Before handover · Integrity

Tells you the truth

What the check requires. The prompt says when the evidence does not support your request, and leaves the decision with you.

Why it matters. A prompt that always agrees with the brief cannot tell you when the brief is wrong.

The instruction, as written in a NOVA prompt

If the evidence does not support the objective, say so. Challenge me
when my thinking is weak - the judgment stays mine.

A typical prompt

The model agrees with the brief it was given.

In the worked example

The analysis surfaced it unprompted: the FY24 cost programme cut the appeals team by 30%, and denials overturned on appeal fell from 64% to 41%.

See Every Check in Action Before You Buy

Every product page shows its worked example in full: what the company brought to each prompt, what the prompt returned, and the finished documents. The figures are made up, and every one of them adds up.

AI Prompt Quality: Frequently Asked Questions

What makes an AI prompt professional or enterprise-grade?

The prompt makes the AI work the way a senior expert’s work is checked inside a large company: the method is written out, each case gets the right structure, every figure comes from your material or is marked missing, hard cases follow written rules, and the answer is checked before you see it.

What is the NOVA Prompt Standard?

Our list of 12 quality checks. Each check is scored 0 to 2, and a NOVA prompt goes on sale only at 24 out of 24.

Which NOVA products are built to the standard?

All of them: the AI PowerPoint generator, both voices of the AI speech writer, the AI valuation model generator, and every prompt in the Prompts for Work and Prompts for Life packs.

Do the prompts still work in ChatGPT and Gemini?

Yes. Every NOVA prompt runs in ChatGPT, Claude and Gemini. Where a prompt builds a Word, Excel or PowerPoint file in Claude, ChatGPT and Gemini return the same content as text and tables.