AI Slide Style Guide Reverse Engineering
Convert existing branded decks into reusable rules for AI presentation design
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 6
- Confidence
- 98%
This method treats a company’s existing presentations as evidence from which a model can infer a reusable slide design system. Select polished decks that genuinely represent the brand, then run a research task asking the model to identify repeated typography, hierarchy, colors, layouts, imagery, iconography, tone, and voice. Go beyond a conventional brand guide by specifying the structure of recurring slide formats, such as agendas, problem statements, solution overviews, data visualizations, and summaries. Review the generated rules against the original presentations and correct unsupported inferences. The resulting document becomes a portable constraint set that can accompany future AI presentation prompts, helping generated decks look intentional and recognizable rather than merely attractive.
Origin
Extracted from Marketing Against The Grain during the preparation of a HubSpot pitch-deck style guide.
Core principles
- 01Use demonstrated brand behavior as the source of truth
- 02Document slide systems rather than isolated visual attributes
- 03Capture both visual identity and communication style
- 04Make implicit design conventions explicit for generators
- 05Choose the strongest existing decks as references
How to run it
- 1
Choose Canonical Decks
Collect presentations that represent the organization’s best and most current visual work.
Pro tip Prioritize flagship event decks and polished customer-facing materials.
Watch out Low-quality or inconsistent examples will teach the model conflicting rules.
- 2
Provide Usage Context
Tell the model whether the new presentations are internal, external, event-based, sales-focused, or intended as leave-behinds.
Pro tip Identify any specific deck families that should carry more weight.
Watch out The same company may use different design systems for different audiences.
- 3
Extract Core Visual Rules
Ask for explicit specifications covering typefaces, weight, size hierarchy, color palette, layout, structure, and text styles.
Pro tip Require concrete rules rather than adjectives such as modern or clean.
Watch out Check whether inferred fonts and colors actually appear in the source material.
- 4
Define Slide Archetypes
Document how recurring slide types should arrange headlines, body copy, evidence, visuals, and whitespace.
Pro tip Include agenda, problem, solution, data, proof, summary, and call-to-action formats.
Watch out A list of brand colors alone does not tell an AI how to compose a slide.
- 5
Capture Supporting Conventions
Add imagery, iconography, tone, voice, and any restrictions governing charts, screenshots, or product visuals.
Pro tip Note whether slides are designed for live narration or independent reading.
Watch out Unspecified image conventions can lead to stylistically inconsistent or fabricated assets.
- 6
Validate and Reuse
Compare the guide with the reference decks, correct errors, and attach the approved version to future slide-generation requests.
Pro tip Version the guide when the presentation system changes.
Watch out Do not assume the first research output is an authoritative brand document.
In the wild
ChatGPT researched existing HubSpot slide decks, with special attention to polished Inbound event presentations. It returned typography, hierarchy, colors, layouts, text styles, imagery, iconography, tone, voice, and designs for several recurring slide types. The guide was then pasted into GenSpark to constrain presentation generation.
→ Generated slides used recognizable fonts, colors, and structural conventions despite requiring some layout fixes.
Common mistakes
Extracting Only Brand Colors
Colors and fonts do not describe how information should be structured. The guide must also cover layout and recurring slide archetypes.
Using Unrepresentative Decks
Old, improvised, or low-quality presentations can cause the model to formalize conventions the organization does not actually want.
Trusting Inferred Details Blindly
A research model may confidently infer an incorrect font, hierarchy, or design rule. Validate the guide against the source material.
Is it for you?
Best for
It is best for organizations with strong existing decks but undocumented presentation conventions.
Not ideal for
It is not ideal when the available decks are outdated, inconsistent, or unauthorized examples of the brand.
From the transcript
“And I had it just do a deep research project on existing HubSpot slide decks out on the internet.”
“So it didn't just give the brand style elements, it actually did a written style guide for the different formats and types of slides that…”
“Talks about imagery and iconography, tone and voice, all the things you need to build a great slide deck.”
From the episode
Your Boss Will LOVE This Pitch Deck Hack (FREE AI)