← Back to the guideThe full worked example: Check your team's board pre-read against the sources before it goes up
campo.
Field guide / L&D / UC-001 · SC-02
The result

1 draft in. 1 board-ready page out.

A VP of Talent & Learning at a 58,000-person manufacturer checks her team's board pre-read against the sources, with any AI chat her company allows.

The resultMei Lin's markup of her team's draft. Fictional company.
TALVORA GROUP · PEOPLE COMMITTEE PRE-READ
Q3 Learning & Development: highlights
Draft v2, L&D analytics
Reviewed by Mei Lin Zhou
Leadership pipeline is strong: 92% of our high-potential leaders completed the Accelerate leadership program in Q3.
OVERSTATED
92% of the 163 enrolled. Of all 310 high-potentials: 48%.
Safety training compliance reached 98% across all sites, our best result to date.
OVERSTATED
34 of 40 sites report. The 6 acquired sites are unmeasured.
Learning is growing fast: total learning hours rose 35% year over year.
OVERSTATED
Headcount rose 18%. Per employee: +14%.
The Digital Skills Academy saved $2.1M in external training spend this year.
NOT SUPPORTED
Vendor's estimate. Finance hasn't validated it.
AI literacy launched to all employees in September.
OVERSTATED
A pilot: ~4,000 office staff in 3 countries; 1,150 started.
Employees value learning more: learning NPS rose 12 points to +41.
OVERSTATED
From 5,100 responses, about 9% of employees.
Recommendation: maintain the 2027 L&D budget at current levels to sustain this momentum.
REMOVE
Not on these numbers. Make the case in Q1 with outcome evidence.
ADD: THE COMMITTEE ASKED TWO QUESTIONS. THE DRAFT ANSWERS NEITHER.
1. Succession bench for the top 60 roles: not in this draft. List as an open item, owner: talent team.
2. Safety at the 6 acquired sites: not yet measured. Say so plainly, owner: EHS.
Caught on review
1
"92% of high-potentials" is 48% of them
2
"Across all sites" is 34 of 40
3
"Launched to all employees" is a 4,000-person pilot
UC-001 · SC-02 · Rev C
FOR
VP, Talent & Learning
SECTOR
Industrial automation
COMPANY
58,000 people · 40 sites
TASK
Check a board pre-read
WRITTEN FOR
ChatGPT · Claude · Gemini · Copilot
AI TEST
ChatGPT, Nov 12, 2026 · 7 of 9 found, 2 partly
Step 0 · The mess

What landed on Mei Lin's desk

A confident draft and four sources that tell a more careful story. The draft isn't dishonest. It rounds every number in its own favour.

DRAFT · L&D ANALYTICS[A]
Q3 Learning & Development: highlights
“92% of our high-potential leaders completed… 98% across all sites… saved $2.1M… launched to all employees…”
A
Six confident claims. Each needs a source.
DATA EXTRACT · HRIS / LMS[B]
Q3 2026 working file
310 high-potentials, 163 enrolled, 150 completed. 34 of 40 sites reporting. Plus hours, headcount and NPS rows.
B
92% of whom? Enrolled, not all high-potentials.
EMAIL · FINANCE[C]
RE: Academy savings number
“that figure is from the Digital Skills Academy vendor's year-end impact report. We haven't validated it.”
C
The $2.1M isn't Talvora's own number.
EMAIL · APAC L&D[D]
AI literacy — first month
“We opened the pilot in September to office-based staff in Singapore, Japan and Australia, about 4,000 people.”
D
“All employees” is a 4,000-person pilot.
NOTE · CORPORATE SECRETARY[E]
People Committee — Q3 pre-read
“The Committee chair has asked that the L&D section address two questions specifically…”
E
The draft answers neither question.
You're not alone
70%
of senior HR executives say their organizations share a standard set of people metrics with their boards. Those numbers get read closely.
i4cp survey of 109 senior HR executives, April 2021. Source
Steps 1–3 · How it works

Three steps, one chat

Here the AI isn't writing. It's checking. The prompt asks it to act like a strict reviewer.

1
Open a new chat

Use the AI tool your company approves: ChatGPT, Claude, Gemini or Copilot.

2
Paste the prompt, draft and sources

Copy the prompt, the draft [A] and sources [B]–[E] from the appendix into one message.

3
Check the checker

Recalculate two numbers yourself. Then decide what the board sees.

The prompt, part by part
Copy-ready version in the appendix

I'm a VP of Talent & Learning at a large manufacturing company. My team drafted the learning page for our board's People Committee pre-read. Before it goes to the CHRO and the board, I need to check it against the source material. Below are the draft [A] and four sources [B]–[E].

1
Who you are, who will read it, and which part is draft and which is source.

Please check every claim in the draft and give me:

  1. Claim check — a table with one row per claim: the claim as written, what the sources actually support, a verdict (supported / overstated / not supported), and which source shows it.
  2. What's missing — anything the committee asked for, or should know, that the draft leaves out.
  3. Suggested rewrite — the same page rewritten so every claim matches the sources, in the same length or shorter.
2
One row per claim. A table makes every gap visible.

Be strict. A board will read this. Recalculate percentages yourself, and check what each number is a percentage of. Don't accept a headline because it sounds positive. If a source doesn't support a claim, say so plainly. Don't add facts that aren't in the sources.

3
The key question: a percentage of what? It catches most overstated numbers. Then guard rails: no flattering headlines, no invented facts.
Where to paste
Every tool has one message box. Paste the prompt, add a blank line, paste the draft and the sources, send. Attaching them as files also works if your tool allows it.
Safe to try
Practise with the fictional draft first. Board papers are confidential: with real ones, use only your company's approved tool.
The AI's first answer

What ChatGPT actually wrote

Exactly as it came back, not edited. Read it the way Mei Lin did: check each point against the source it cites.

UNEDITED
ChatGPT · new chat · Nov 12, 2026 · one run
settings not recorded

The draft is not ready for board circulation. Several figures are mathematically correct, but their descriptions are misleading. The two questions explicitly requested by the People Committee are also unanswered.

1. Claim check

Claim as written [A]What the sources actually supportVerdictSource
“Leadership pipeline is strong”No succession-readiness evidence for the top 60 roles. Program participation does not establish bench strength.Not supportedB, E found
“92% of our high-potential leaders completed…in Q3”150 of 163 enrolled participants completed (92.0%). Only 150 of 310 identified high-potentials completed (48.4%).OverstatedB found
“Safety training compliance reached 98% across all sites”98% compliance among 34 reporting sites. Six acquired sites have no LMS data.OverstatedB found
“Our best result to date”Reporting-site compliance increased from 96% in Q3 2025 to 98% in Q3 2026. No earlier history or equivalent company-wide comparison is provided.Not supportedB found
“Learning is growing fast: total learning hours rose 35% year over year”Hours increased from 412,000 to 556,200, or 35.0%. Average headcount also rose from 49,000 to 57,800. Hours per employee increased approximately 14.4%.SupportedB A: partly
“Digital Skills Academy saved $2.1M”$2.1M is an unverified vendor estimate. Finance reports approximately $0.6M less external training spend YTD, partly attributable to timing and a hiring freeze. Neither figure establishes Academy savings.Not supportedC found
“AI literacy launched to all employees in September”September pilot covered approximately 4,000 office staff across three APAC countries. 1,150 started the first module (28.8%). Frontline and plant workers are excluded.OverstatedD found
“Positioning Talvora for the next phase of automation”No evidence of readiness, business outcomes or a company-wide rollout.Not supportedD
“Employees value learning more: learning NPS rose 12 points to +41”NPS increased from +29 to +41, a 12-point rise. Responses increased from 3,900 to 5,100. The data does not establish a change in sentiment across the workforce.OverstatedB B: partly
“Maintain the 2027 L&D budget at current levels to sustain this momentum”No 2027 budget figures, cost analysis, funding requirements or demonstrated link between spending and outcomes.Not supportedB–E C

Key recalculations

Accelerate completions / all high-potentials

48.4%

150 ÷ 310, not 92%

Sites reporting safety data

85.0%

34 ÷ 40 sites

Learning hours growth per employee

+14.4%

Versus +35.0% total hours

AI pilot participation started

28.8%

1,150 ÷ approximately 4,000

The 98% safety compliance figure cannot be recalculated independently because the source provides neither the number of required completions nor the number achieved. It covers reporting sites only.

2. What's missing

The two most consequential omissions are precisely what the Committee chair asked the team to address.

1. Succession readiness for the top 60 leadership roles [E] found D

The draft provides Accelerate participation and completion, but no evidence about the top 60 positions. Missing information includes ready-now successor coverage, readiness timelines, roles without identified successors, and identified gaps. None can be inferred from the 310 high-potentials.

2. Safety compliance at newly acquired sites [B, E] found D

Six acquired sites in Brazil and India are not yet reporting through the LMS. Their compliance is unknown. The draft should clearly state that 98% applies only to the 34 reporting sites, not all 40 sites.

3. Other material gaps [B–D]

  • Learning hours should be shown alongside headcount-adjusted growth, especially following the May acquisition.
  • Financial claims require Finance validation and an explanation of what costs can actually be attributed to the Academy.
  • AI literacy has no evidence yet of course completion, skill gains or readiness for a wider rollout. Translations and mobile access remain prerequisites.
  • Learning NPS lacks response-rate and workforce-representativeness analysis. The observed increase should not be presented as a proven company-wide sentiment shift.
  • The 2027 budget recommendation requires funding requirements, actual spending and evidence of outcomes before it can be supported.

3. Suggested rewrite

This version is shorter than the original draft and retains only supported statements.

Q3 Learning & Development: Board Update

  • Leadership development: 150 of 163 Accelerate participants completed the program (92% of enrolled; 48% of 310 identified high-potentials). Succession readiness for the top 60 leadership roles has not been established.
  • Safety training: Reported compliance reached 98%, up from 96%, across sites submitting data. Six of 40 sites, all newly acquired, are not yet on the LMS. Their compliance remains unknown.
  • Learning activity: Total hours rose 35% year over year; hours per employee rose approximately 14%. A
  • Digital Skills Academy: The vendor's $2.1M savings estimate remains unverified by Finance.
  • AI literacy: September pilot opened to approximately 4,000 APAC office employees; 1,150 started the first module.
  • Learning feedback: NPS increased 12 points to +41. B

Committee follow-up: Establish succession readiness for the top 60 roles and verify acquired-site safety compliance. The available evidence does not support a 2027 budget recommendation. C

Before submission: The revised page makes the gaps visible, but it cannot fully answer the chair's two questions without additional data. Those two areas should be resolved or explicitly presented as outstanding at the December 9 meeting. The pre-read is due December 1. [E]

foundcaught a planted problemA–DMei Lin changes it; see the next section
The human part

What Mei Lin changed, and why

ChatGPT caught most of the problems. It half-caught two, and its rewrite dropped one of its own warnings. The rest of Mei Lin's edits are judgment calls.

What ChatGPT found: 7 of 9 fully, 2 partly
One run
✓92% was of enrolled; 48% of all
✓“All sites” was 34 of 40
½Hours: per head only +14%
✓$2.1M is the vendor's figure
✓“All employees” was a pilot
½NPS from ~9% of staff
✓Succession question unanswered
✓Acquired-site safety unanswered
✓Budget ask unsupported
Four edits: two catches, two judgment calls
A
CHATGPT WROTE
Rated the hours claim “Supported”. Rewrite: “Total hours rose 35% year over year; hours per employee rose approximately 14%.”
MEI LIN'S VERSION
“Learning hours per employee rose 14%. Total hours rose 35% while average headcount grew 18%, a period that includes the May acquisition.”
WHY · AUDIENCE
With 18% more people, total hours were bound to rise. Per employee is the fairer comparison, so it goes first.
B
CHATGPT WROTE
Its check called the NPS claim overstated. Its rewrite: “NPS increased 12 points to +41.”
MEI LIN'S VERSION
“Learning NPS rose 12 points to +41, from 5,100 responses (about 9% of employees).”
WHY · ACCURACY
The rewrite dropped the caveat its own check raised. Always compare the rewrite with the check.
C
CHATGPT WROTE
“The available evidence does not support a 2027 budget recommendation.”
MEI LIN'S VERSION
Budget ask removed from this pre-read. Mei Lin makes the case at the Q1 budget review, with outcome evidence.
WHY · TONE
True, but a VP doesn't tell her board that her own budget is unsupported. She removes the ask and comes back with evidence.
D
CHATGPT WROTE
“Those two areas should be resolved or explicitly presented as outstanding at the December 9 meeting.”
MEI LIN'S VERSION
A boxed “Open items” answer to both questions: not yet available, with an owner for each.
WHY · AUDIENCE
The chair asked two direct questions. “Not yet, and here's who owns it” beats silence.
What only you can do
Decide what a board should see, say it in a way your CHRO can stand behind, and own every number. The AI checked the draft. You sign the pre-read.
Your turn

Make it yours

Use it on any draft that's about to go up: a board paper, a QBR deck, an investor update.

Your prompt: fill in the brackets

I'm a [your role] at a [industry] company. My team drafted [the document] for [the audience]. Before it goes to [who], I need to check it against the source material. Below are the draft and [number] sources.

Please check every claim in the draft and give me: 1. Claim check (a table: the claim as written, what the sources actually support, a verdict, and which source shows it); 2. What's missing (anything [the audience] asked for, or should know); 3. Suggested rewrite (same length or shorter).

Be strict. [The audience] will read this. Recalculate percentages yourself, and check what each number is a percentage of. Don't accept a headline because it sounds positive. If a source doesn't support a claim, say so plainly. Don't add facts that aren't in the sources.

Five checks before it goes up
1
Every percentage says what it's a percentage of.
2
Words like all, every and launched match the real scope.
3
Growth is shown per person when headcount changed.
4
Savings figures are your finance team's, not a vendor's.
5
Every question your audience asked is answered, or listed as open with an owner.
Make it repeatable (optional)
Save the prompt once and reuse it every quarter: a custom GPT in ChatGPT, a project in Claude, a Gem in Gemini, or an agent in Copilot. Some companies switch these off; copy and paste always works.
With real board papers
Use only the AI tool your company approves for confidential material, and follow its rules on employee data.
What this example shows
One run of this prompt in ChatGPT fully caught 7 of 9 problems planted in a fictional draft, partly caught the other 2, and invented nothing.
What it doesn't show
That every tool, setting or run will do the same, or how much time it saves you. Talvora Group and its people are fictional.
Appendix

Copy-ready: the prompt and the draft and sources

Copy the prompt, then the draft [A] and sources [B]–[E], into one message in a new chat. Everything here is fictional and safe to paste.

The prompt

I'm a VP of Talent & Learning at a large manufacturing company. My team drafted the learning page for our board's People Committee pre-read. Before it goes to the CHRO and the board, I need to check it against the source material. Below are the draft [A] and four sources [B]–[E].

Please check every claim in the draft and give me:

  1. Claim check — a table with one row per claim: the claim as written, what the sources actually support, a verdict (supported / overstated / not supported), and which source shows it.
  2. What's missing — anything the committee asked for, or should know, that the draft leaves out.
  3. Suggested rewrite — the same page rewritten so every claim matches the sources, in the same length or shorter.

Be strict. A board will read this. Recalculate percentages yourself, and check what each number is a percentage of. Don't accept a headline because it sounds positive. If a source doesn't support a claim, say so plainly. Don't add facts that aren't in the sources.

[A] The draft — "Q3 Learning & Development: highlights" (prepared by the L&D analytics team)

Q3 Learning & Development: highlights

  • Leadership pipeline is strong: 92% of our high-potential leaders completed the Accelerate leadership program in Q3.
  • Safety training compliance reached 98% across all sites, our best result to date.
  • Learning is growing fast: total learning hours rose 35% year over year.
  • The Digital Skills Academy saved $2.1M in external training spend this year.
  • AI literacy launched to all employees in September, positioning Talvora for the next phase of automation.
  • Employees value learning more: learning Net Promoter Score rose 12 points to +41.

Recommendation: maintain the 2027 L&D budget at current levels to sustain this momentum.

[B] HRIS / LMS data extract — Q3 2026 (from the analytics team's working file)
Measure Q3 2025 Q3 2026 Note
High-potentials identified (talent review) 296 310
High-potentials enrolled in Accelerate 150 163
Accelerate completions (of enrolled) 131 150
Sites reporting safety-training data 38 of 38 34 of 40 6 acquired sites (Brazil, India) not yet on the LMS
Safety-training compliance, reporting sites 96% 98%
Total learning hours 412,000 556,200
Average headcount 49,000 57,800 Acquisition closed May 2026
Learning NPS survey responses 3,900 5,100 Pulse sent to all employees
Learning NPS +29 +41
[C] Email — Finance

From: Andre Kowalski, Finance Business Partner, HR Subject: RE: Academy savings number

Mei Lin, on the $2.1M: that figure is from the Digital Skills Academy vendor's year-end impact report. We haven't validated it. External training spend in our ledger is down about $0.6M year to date, but some of that is timing and the hiring freeze in H1. I'd be careful putting $2.1M in front of the board as our number.

Andre

[D] Email — APAC L&D

From: Sunil Rao, L&D Lead, APAC Subject: AI literacy — first month

Hi Mei Lin, quick update on AI literacy. We opened the pilot in September to office-based staff in Singapore, Japan and Australia, about 4,000 people. 1,150 have started the first module so far. Frontline and plant teams aren't in scope yet; we need translated content and a mobile version first. Happy to share more at the next review.

Sunil

[E] Note — Corporate Secretary

From: Office of the Corporate Secretary Subject: People Committee — Q3 pre-read

Reminder that pre-reads for the December 9 People Committee are due December 1. The Committee chair has asked that the L&D section address two questions specifically: (1) how ready is our succession bench for the top 60 leadership roles, and (2) are the newly acquired sites meeting our safety-training standard?