Turn a product-launch brief into a 10-slide stakeholder deck in under an hour with Copilot, Gamma, or plain ChatGPT — then rebuild the draft around assertion headlines, add speaker notes, and fact-check every generated stat against the brief. Some of it will be wrong, and decks are where hallucinations go public.
Sneha Kulkarni is a product marketing manager at Trellis Software Pvt Ltd, a fictional SaaS company. The leadership review of the "Trellis Approve" launch got pulled forward to tomorrow, and Sneha has the launch brief but no deck. An hour is genuinely enough — if the AI does the assembly and Sneha does the thinking. The catch: generated decks are fluent, confident, and casual with facts. Numbers drift, dates shift, claims inflate. And unlike an email, a wrong number on a projected slide is wrong in front of everyone at once. This lab makes you generate fast and verify ruthlessly — including against a brief that itself does not fully agree with itself.
A condensed version of the 6-page brief. This is the only approved source for the deck.
TRELLIS APPROVE - LAUNCH BRIEF (v4, marketing-approved)
PRODUCT: Trellis Approve - approval-workflow add-on for the Trellis
suite. Lets ops and finance teams build no-code approval chains
(expenses, discounts, contracts) inside tools they already use.
LAUNCH DATE: 15 September, six weeks from the 1 August kickoff.
Beta ran May-July with 40 customer teams.
PRICING: Rs 4,999 per team per month on the annual plan.
Monthly plan: Rs 5,999 per team per month. Beta customers get
20 percent off year one. Later in this brief the annual plan is
referred to as Rs 4,499 - marketing to confirm final list price.
BETA RESULTS (verified by product analytics, 40 teams):
- Median approval turnaround fell from 3.2 days to 11 hours.
- 87 percent of beta teams built their first workflow without
contacting support.
- 31 of 40 teams converted to a paid intent letter.
- NPS among beta admins: 58.
TARGET AUDIENCE: mid-market ops and finance leaders (100-1,000
employees), existing Trellis customers first (approx 2,400 accounts),
then new logos from Q4.
COMPETITIVE NOTE: two established competitors exist. Our
differentiators: native suite integration (no sync), per-team
pricing (not per-seat), and audit-trail export. Do NOT claim we are
the only no-code approval product - legal has flagged this claim.
RISKS: migration tooling from spreadsheets ships 4 weeks after
launch (mid-October). Do not promise migration at launch.
CALL TO ACTION FOR STAKEHOLDER DECK: approve launch plan, approve
Rs 18 lakh Q4 campaign budget, confirm sales enablement dates.
Define the storyline before generating anything. Write ten assertion headlines yourself — full sentences that each state one message, in order: the problem, the product, proof it works, who it's for, how we win, pricing, launch plan, risks, the money ask, the decision needed. An assertion headline is "Beta teams cut approval time from days to hours", never "Beta Results". Ten sentences, five minutes, no AI yet — this is the part that makes it your deck.
Generate the draft deck. Feed your tool the brief plus your ten headlines: "Create a 10-slide stakeholder deck. Use these exact headlines as slide titles, in this order. Body content only from the brief. 3-5 short bullets or one visual concept per slide. Flag any slide where the brief lacks content." In Gamma or Copilot, paste both documents; in a chat model, ask for slide-by-slide text to transfer. Let it run — do not fix anything mid-generation.
Run the fact audit — every number, date, name, claim. This is the heart of the lab. Make a two-column list: every factual statement on every slide, and the exact line of the brief that supports it. Work slide by slide. Three categories emerge: supported (matches the brief), drifted (number or date changed in generation), and unsupported (appears nowhere in the brief). Expect to find at least one of each — generated decks routinely round numbers up, compress timelines, and "improve" claims.
Interrogate the brief's own contradictions. Your audit will hit at least one place where the brief disagrees with itself — where two statements cannot both be true. A deck cannot carry both versions, and picking one silently is how wrong numbers reach leadership with your name on the title slide. For each internal contradiction: mark the affected slide, choose the defensible handling (use the verifiable value, or flag "pending confirmation" on the slide), and write one line on why. If you found no contradictions, audit again — slower.
Check the legal and risk tripwires. Two specific instructions in the brief are prohibitions: the competitive claim legal flagged, and the migration promise. Search every slide — including AI-generated filler bullets — for any phrasing that crosses either line. Generated decks are especially prone to superlatives ("the only", "the first", "seamlessly migrate") because marketing corpora are full of them. Rewrite any offender to the compliant version of the claim.
Rebuild weak slides around their headline. Any slide where the bullets do not prove the headline gets fixed: cut bullets that restate the title, cut anything unsupported, and ask the AI for alternatives grounded in specific brief lines: "Rewrite slide 3's bullets using only these three facts: [paste them]." One message per slide; if a slide is making two points, split it or cut one — you have exactly ten slides, so cutting forces prioritization, which is the point.
Generate speaker notes and the tough-question bank. Prompt: "For each slide: 2-3 sentences of speaker notes in a natural spoken voice, plus the single hardest question a CFO might ask on this slide and a one-line grounded answer." Fact-check the notes exactly as you did the slides — hallucinations love speaker notes because nobody audits them. The Q&A bank is your rehearsal: if any answer isn't in the brief, you have found tomorrow's homework.
Final pass at the projector distance. View the deck in slideshow mode. Read only the headlines, in order, skipping everything else — they should tell the complete story on their own. Then check the last slide asks for the three specific decisions the brief demands, with the budget figure exactly right. Stop the timer. Note your time and your error count from step 3 — both numbers should shrink every time you run this workflow.
Every generated number, date, and name on every slide must be traced to the source before presenting — a deck is the highest-blast-radius document most professionals ever produce.
Generate the deck fast, but treat every slide as a claim to be proven: headlines you wrote, facts traced to the brief, contradictions surfaced instead of picked, and prohibitions searched for explicitly. One hour is enough — because the hour is mostly verification, and that is what it should be spent on.