Business ·
An AI presentation begins with sources, not slides
AI can assemble copy and imagery quickly, but speed does not make a presentation accurate. A dependable process begins with a question, sources and a chain of evidence. Slides come later.
Define the decision, not the topic
“Make a presentation about the project” says very little about the outcome. One audience needs to approve a budget, another must understand the research, and a third needs to choose between options. Begin by writing down who will see the deck, which decision they face and what they already know.
This produces one working sentence: after the presentation, the audience should understand or do a specific thing. It removes information that interests the author but does not move the story forward. AI becomes useful once this task is clear, rather than when it is asked to invent twenty slides about a broad subject.
Build a source ledger
Collect documents, spreadsheets, interviews, photographs and links in one ledger. For each source, record its title, author or owner, date, location and the claim it supports. If two sources disagree, mark the conflict before it turns into a confident statement on a slide.
A model can extract claims and compare documents, but it does not become a source. Every number, quotation, date or specification should lead back to the original material. For confidential documents, define approved tools, access and retention rules before uploading anything.
Construct the argument
A good presentation is a movement of thought rather than a sequence of facts. It usually needs a starting situation, a tension or question, evidence, a proposed response and a next step. This is not a compulsory template for every genre, but it tests why one slide follows another.
At this stage, AI can propose several structures from the same evidence: one for an executive, one for a working team and one for a public talk. Choose the structure that serves the audience’s decision without depending on invented transitions, rather than the one that merely sounds dramatic.
Map the slides
Before styling, I describe every slide in one line: claim, evidence and purpose. If that line contains two independent claims, split it. If a claim has neither evidence nor a consequence, the slide may not be needed.
The map exposes rhythm: where the story needs a diagram, a comparison, a single image or a pause. It also reveals repetition. Regenerating a short map is cheaper than rebuilding a finished deck after the visual style has hardened a weak structure.
Generate a constrained draft
An AI brief should state the audience, objective, available sources, target length and prohibitions. For example: add no figures without a reference, invent no quotations, distinguish assumptions from facts and preserve the project’s terminology. Clear boundaries reduce the time spent correcting confident fabrication.
Treat the first output as an assembly draft. Test headings for clarity, shorten paragraphs and replace general statements with specific evidence. Add images and diagrams only where they clarify a claim; decorative imagery should never pose as evidence.
Build a visual system
A coherent deck needs only a few rules: a grid, two levels of typography, a limited palette, an approach to imagery and one way to show sources. The rules should survive a long heading, a table, a photograph and an empty state. Consistency comes from repeated logic rather than one identical template imposed on every slide.
Charts should be built from source data rather than copied as images from a model’s answer. Axis labels, units, time periods and comparison bases need to remain visible. If AI generates an illustration, state its role: it is an image or reconstruction, not documentary evidence of an event.
Verify and rehearse
Final review happens in layers. First, check facts against the source ledger. Then review logic, language, contrast, text size and chart legibility. After export, open the file on the device and in the format the audience will use, because wrapping, fonts and video can change.
Reading aloud exposes problems that remain hidden in the editor. If a slide cannot be explained in one sentence, its role is unclear. If the speaker has to read all the text from the screen, the material has not yet become a presentation. Rehearsal returns control to the author, where a generator cannot replace it.
When the process deserves automation
A one-off deck rarely needs its own system. Weekly reports, proposals and project reviews, however, repeat the same sources and rules. Approved data, a template, draft generation and mandatory review can then become one internal tool.
Automation makes sense after the manual process is understood. In the Method, I first describe sources, roles and control points, then build a pilot, and only then move repeatable work into a digital product. AI speeds preparation without replacing evidence or the author’s judgement.
From an idea to a digital product
If the task calls for a service or internal tool, we can define the first complete journey and a path to a working release.