Documents, slides and images

Thursday afternoon: Priya owes the leadership team a quarterly operations review — twelve slides, due Monday. The old workflow starts with opening the slide tool and staring at a blank title slide, which is exactly backwards. Slides are a presentation of thinking; the thinking is text; do the text work first, with help, and the deck assembles itself at the end. This lesson is that workflow, plus what you can and can't expect from AI-generated images along the way.

The deck workflow: brief, outline, then slides

Start with the same five-part ask from lesson 3, aimed at an outline rather than a document:

I present a quarterly ops review to leadership — 12 slides, 15 minutes.
Audience: CEO, CFO, COO. They care about cost, reliability, and risk.
The quarter: on-time delivery recovered 89% to 94%. Fleet costs up 6%
(fuel). Nashik warehouse delayed a month. Driver attrition improved.
Two client escalations, both closed. Next quarter's ask: budget for
5 new vehicles.

Give me a slide-by-slide outline: title, the one point the slide makes,
and 2-3 supporting bullets. End with the budget ask as its own slide.

Two things about what comes back. The outline will follow the oldest rule of presenting — one point per slide — because the ask demanded it, and that structure is most of what separates good decks from slide-shaped documents. And the model will sequence for the audience it was told about: numbers early, the money ask at the end after the case is made. Then iterate the outline the usual way — "merge slides 3 and 4", "the attrition story deserves its own slide" — and only then open the slide tool. From a settled outline, asking for speaker notes is one more message: "for each slide, two or three sentences of what I should actually say, conversational, not reading the bullets."

Reading files directly

Everything above pastes text in, but assistants as of mid-2026 also accept whole files — drop in a PDF, a document, a spreadsheet export, even a photo of a whiteboard, and ask against it. This changes the summarize-and-extract workflows from lesson 5 too: no more copy-pasting forty pages. Three practical notes. Long files hit the same working-memory limits as long chats, so for a 200-page document, ask per-section rather than trusting one global pass. Scanned PDFs sometimes read badly — if answers seem oddly blind to a section, the text may not have survived the scan. And a file you upload is data you've shared: the lesson 11 rules apply to attachments exactly as they apply to pasted text.

Images: where they help and where they embarrass

AI image generation is genuinely useful for slides — and genuinely risky in specific, predictable places. It is good at conceptual imagery: a clean illustration of a warehouse at dusk, an abstract visual for "growth" that isn't a stock-photo arrow. It remains unreliable at text inside images (labels come out subtly mangled), at precise diagrams (an org chart or process flow with exact boxes and arrows — build those in your slide tool, where they're editable anyway), and at specific real things — your actual product, your actual office, a real person. The practical rule for business use: AI images for mood and concept, real tools for diagrams, real photos for real things. And if an image could be mistaken for a photograph of something that happened, label it or don't use it — lesson 12 of the marketing course covers the disclosure norms if your work is public-facing.

Do this today: take the next presentation on your calendar — however far off — and generate the slide-by-slide outline now, while the brief is easy. Ten minutes today converts Monday's blank-page hour into an editing pass.