Today's AI Tip
Report to slides? One message per slide, then ask what's cut.
What most of us type
Tested in GPT-5.5 (OpenAI API), Oct 2026
This is the vague version, so don't copy it. Most of us paste the report and type something like it. Our sample is a fictional bakery's Q3 report with five caveats tucked inside.
Twelve slides, titled by topic
Real output · GPT-5.5 (OpenAI API) · Oct 2026
Asked only to turn the report into slides, GPT-5.5 gave 12 slides titled by topic, like "Q4 Plans". Some had a takeaway line, but none tied to the decision. This one adds a reason the report never gives.
A label names it. A message says it.
A topic label says what a slide is about. A message ties the slide to the decision. Give the audience and the decision, then ask for a sentence and a figure.
Source: General prompting advice: Anthropic, Prompting best practices (Claude docs), checked Oct 2026
The prompt that changed the deck
Tested in GPT-5.5 (OpenAI API), Oct 2026
This is the full prompt. Change the audience and the decision line to fit your own deck.
Six slides with real messages
Real output · GPT-5.5 (OpenAI API) · Oct 2026
Six slides instead of twelve, with full-sentence headlines. Not perfect: Slide B's figure is a fragment, and Slide D's headline leans toward pricing action. Another headline just restated the decision.
The short deck dropped every caveat
Real output · GPT-5.5 (OpenAI API) · Oct 2026
None of the report's 5 caveats made the 6 slides, including the one that matters most for the price decision. The 12-slide version had kept them. The prompt's last line exposed the gap.
Read the list, then fix the deck
Before pasting a real report, remove names, account numbers and other personal details, and check the tool's data policy or your employer's approved tools. Then put the untested-price caveat back on its slide.
Try it: Paste your next report with this prompt; change the audience and decision.
Sources and assumptions
- Anthropic, Prompting best practices (Claude docs), 'Be clear and direct' and 'Add context to improve performance' sections: Verbatim: 'Being specific about the output you want can help enhance results.' 'Be specific about the output format and constraints you want.' 'Providing context or motivation behind your instructions, such as explaining to Claude why such behavior is important, can give Claude more information about your goals and help it deliver more targeted responses.' 'Think of Claude as a brilliant but new employee who lacks context on your norms and workflows.' This is Anthropic's guidance written for Claude; the post's test ran on GPT-5.5, so slides must present it as general prompting guidance, not a claim about GPT-5.5. (checked 2026-10-24)
- Real test runs, engine/ai-test.js (record.tests ids before-gpt and after-gpt): GPT-5.5 (OpenAI API) outputs dated 2026-10-25 UTC (2026-10-24 local): before = 12 slides with topic headlines, an invented rationale ('Intended to offset flour and butter cost increases') and derived numbers (~11%, ~56%, ~$17,280). After = 6 one-sentence headlines with verbatim figures, plus a list of all 5 report caveats that no slide mentions. (checked 2026-10-24)
Assumptions:
- The sample report (Harbor Lane Bakery Co.) is fictional and was written for this test; all of its numbers are invented.
- The tests ran on GPT-5.5 through the OpenAI API, not the ChatGPT app. The Claude runs (ids before, after, after-claude) failed to execute in this research session and have no output, so the post must not claim anything about Claude's result.
- Each prompt was run once; another run may give a different slide count or wording.
- The before output's derived figures are arithmetically correct (41,000 / 371,000 = 11.05%; 230 / 410 = 56.1%; 640 × $27 = $17,280), but none of them appear in the report.
- The report states no reason for the price rise; the before output's 'Intended to offset flour and butter cost increases' is the model's inference.
The short version
- A topic label names a slide; a message states it.
- Give the audience, the time and the decision.
- Ask for one sentence and one exact figure.
- Ask for the caveats no slide mentions.
