How I build AI-powered
content systems

Overview

FLOP is a fictional streaming app for 1-star films.

FLOP doesn’t have a content system yet. I show how it can scale content, keep its team aligned, and strengthen the brand.

Role: Content Designer, Systems and Governance
Tools: Figma, ChatGPT, Claude, NotebookLM

Problem

Content-heavy products scale quickly. As new features, platforms, and teams ship, the language begins to drift:

  • Inconsistent terminology

  • Uneven error handling

  • Confusing verbs

  • Loss of trust during failure states

This needs to be nipped in the bud.

Step 1: Listen and learn

Goal: Understand the situation before creating guidelines.

  • Reach out to key people who may have brand documents, business goals, recurring support issues, and other relevant info

  • Understand platform differences: TV, mobile, desktop, web

    AI

  • Prompt ChatGPT and Claude to list common failure scenarios in streaming apps (e.g., playback interruptions, network loss, payment errors, etc.)

  • Ask these AI models to group these by user impact and trust sensitivity

Received

  • A visual branding doc from Design

  • Business goals and user demographics doc from Product

  • A list of the most common tickets from Support

  • An AI-generated list of prioritized, high-risk areas: playback errors, network interruptions, and confusion around tip payments

Step 2: Content audit

Goal: Explore the content, its patterns, and its needs

  • Identify existing UI components, types of text, terminology conflicts, capitalization, types of errors, UI limitations, and key terms that need to be defined

Before any changes made

Step 3: Language principles

Goal: Establish decision-making constraints before any rules.

  • I defined a small set of principles to guide all language decisions

  • Ran it through ChatGPT and Claude to look for weaknesses

Language principles

FLOP uses language deliberately as a clean, lean system. These principles allow the product to scale without losing branding, clarity, or trust. 1. Plain language

  • Simple, clear language comes before branding. For example, onboarding uses the word films before new users learn that films are called flops.

  • Aim for a 7th-grade reading level.


2. One word, one job

  • Each approved term has a single purpose and a clearly defined scope.

  • Synonyms are avoided to prevent ambiguity, reduce cognitive load, and ensure consistent meaning across surfaces.


3. Language reflects user intent, not UI mechanics

  • Verbs are chosen based on what the member is doing, not how the interface behaves.


4. Follow platform conventions for clarity

  • Language respects platform expectations across mobile, desktop, and TV. For example, tap on mobile, click on desktop, and Exit on TV.

Goal: Lock down terminology before tackling sentence styles.

  • I built a governed word list

  • If some terms were unclear, inconsistent, or otherwise not right, I changed them to what’s best

  • Organized them not A-Z but in context groups

  • A-Z is good for searching but bad for understanding systems and designing for scalability

  • Context grouping shows where the words are used and with what other words

  • Again, I ran this list by ChatGPT and Claude to look for weaknesses

  • Also, I pasted all UI screens and asked these AI models to find any words I missed (in a company I would be very careful about this, especially if there are legal issues or unreleased flows)

Step 4: Word list

FLOP Word List

BRAND & PRODUCT
FLOP
Definition: The name of the product and brand.
Usage: Always written in all caps and used only for branding, titles, and product references
Avoid: Flop (as a brand name), FLOP, lowercase or sentence case when referring to the product





ACCESS
Create
Definition: The action of setting up a new account in FLOP.
Usage: Use Create to describe account creation actions and flows
Avoid: Register, Registration, Sign up (too similar to “sign in”)





Member
Definition: A person who has an account with FLOP.
Usage: Use member instead of user in all UI copy
Avoid: User

Account
Definition: A member’s credentials, security, and identity settings.
Usage: Used for authentication, privacy, and membership management
Avoid: Profile when referring to security or sign-in


Sign in
Definition: The action of accessing an existing account.
Usage:

  • Sign in as a verb (e.g., “Sign in to FLOP”)

  • Sign-in as a noun or adjective only (e.g., “sign-in screen”)

Avoid: Log in, Login

CONTENT
Flop
Definition: A single movie within the FLOP catalog.
Usage: Default noun for content items. Plural: flops.
Avoid: Movie, Film (allowed only on the first onboarding screen to establish meaning)

Director
Definition: The creator of a flop.
Usage: Used consistently across profiles, bios, and tipping flows
Avoid: Artist


Follow
Definition: The action of tracking a director so you can see their flops and tip them.
Usage: Sole verb for this relationship
Avoid: favorite, subscribe, friend

PLAYBACK
Playback
Definition: The act or state of watching a flop.
Usage: Used in settings, errors, and technical descriptions
Avoid: Player as a user-facing term, Watching mode

History
Definition: A list of flops a member has previously watched.
Usage: Used consistently in navigation and settings
Avoid: Watch history, Viewing history

Leave
Definition: The action of abandoning a flop or playback flow where progress may be lost.
Usage: Used only when exiting a specific flop or flow
For closing or exiting the app, see “Exit”.


Stay
Definition: The action of remaining in the current flop or playback flow.
Usage: Used as the safe alternative paired with Leave.
Avoid: Cancel, Back


NAVIGATION
Search
Definition: Direct lookup of flops or directors.
Usage: Used for fields and actions involving direct input
Avoid: Find, Explore, Discover at the field level

Settings
Definition: The configurations for FLOP.
Usage: Used consistently across the app
Avoid: Preferences, Options


Support
Definition: Help, feedback, and issue reporting.
Usage: Used as the sole label for assistance
Avoid: Customer support, Help desk

MONETIZATION
Tip
Definition: Monetary support given to a director.
Usage: Used in all payment and support flows
Avoid: Donate, Donation, Charity or fundraising language


SYSTEM
Continue
Definition: The action of progressing forward in a flow.
Usage: Used in onboarding and step-based flows
Avoid: Next

Exit
Definition: To close the FLOP app on desktop or TV platforms.
Usage: Indicates leaving the app itself, not a flop or playback flow. Not used on mobile.
Avoid: Close, Leave (reserved for abandoning a flop or playback flow)

Try again
Definition: The primary recovery action after an error.
Usage: Used for all retry states
Avoid: Retry

Step 5: Writing guidelines

Goal: Establish writing at the sentence level for predictability.

  • Clarity and sentence length

  • Capitalization (sentence case)

  • Punctuation

  • Numbers and money

  • Ampersands

  • Error tense (can’t vs couldn’t)

  • Voice and agency (“we” vs “you”)

Writing Guidelines


1. Voice and tone
a. The FLOP voice is quietly confident and cultured like the owner of an indie movie theater.

b. Tone adapts by context.

  • Discovery & editorial: Subtle personality is allowed. Language may be dry or opinionated, but never ironic.

  • Playback, errors, payments, settings: Neutral and literal. Clarity comes first.


c. Say what’s happening, plainly

2. Be clear

a. Write to be understood on first read. Avoid jokes, metaphors, or irony in errors, payments, settings, and privacy.

3. Use contractions

a. Contractions (e.g., “can’t”) sound natural. Use them in UI copy to sound human and direct.

  • Avoid non-contracted forms unless required for legal or accessibility reasons

  • Avoid uncommon contractions such as ain’t and y’all

4. Keep sentences short and declarative

a. Favor simple sentence structures.

  • One idea per sentence

5. Capitalization
a. All UI copy in FLOP uses sentence case by default.

  • Screen titles

  • Headings

  • Body copy

  • Button labels

  • Form labels

  • Error messages

  • Settings names

  • Exception: the brand name FLOP is all caps. This helps differentiate it from the common noun flop, which refers to a movie available in the app. It’s also just for general branding as it movie theaters use all caps.

6. Punctuation

a. Periods

  • Use periods in full sentences in body copy

  • Omit periods in headings, labels, and button text


b. Commas

  • Use to improve readability, not rhythm

  • Avoid long, compound sentences that rely on multiple commas

c. Colons and ellipses

  • Avoid ellipses (…) in UI copy

  • Use colons only in labels or structured lists when clarity requires it

d. Exclamation points

  • Do not use exclamation points in the product

  • Avoid: Saved!, Try again!

7. Ampersands (&)
a. Use ampersands in flop titles, UI labels, or tight spaces where brevity improves clarity. In all other cases, use “and” by default.
Do: Genre: Bleak & dystopian, Page: Terms & privacy

Don’t: We couldn’t save your progress & your history.

Exceptions: Established brand or feature names such as AT&T

8. Errors

a. Explain what’s wrong before telling the member to do.

  • People need orientation before action. Jumping straight to a fix increases confusion and anxiety.

b. Use can’t for current errors that require action before trying again

  • Ex: Heading: Can’t play this flop Body: Check your internet connection and try again.

  • Don’t use unable

c. Use couldn’t for errors when the user can try again without taking other actions first

  • Ex: Heading: Couldn’t save your progress Body: Check your connection and try again.

  • Don’t use unable

d. Guide toward one clear solution

  • Multiple competing actions slow decision-making.


9. Numbers
Use numerals for clarity and scannability. Numbers should be easy to parse at a glance.


a. Use numerals (0–9, 10, etc.) instead of spelling numbers out
Do: 3 flops available, Try again in 10 seconds
Don’t: Three flops available, Try again in ten seconds
Exceptions:

  • Spell out numbers only when they start a sentence

  • Three flops were added today. (rare; usually rewrite instead)


b. Ranges

  • Use an en dash or “to” for ranges

  • 1–3 minutes or 1 to 3 minutes

    1. Avoid slashes (/) for ranges.


c. Money

  • Use numerals with currency symbols

  • Tip $5

    1. $99 per year

    2. Avoid spelling out currency.


d. Dates & time

  • Use numerals for dates and time units

  • 5 minutes ago

    1. 2 hours remaining

    2. Avoid vague terms like soon or a while.


e. Ordinals

  • Use numerals with suffixes

  • 1st flop watched

Step 7: Scalability test

Goal: Validate the system and enable repeatable enforcement with AI.

I used NotebookLM to test the rules against real UI copy. This works with any source-grounded LLM and helps catch inconsistencies early, without replacing human judgment.

To the right are standardized, repeatable prompts designed to make validation fast with consistent, predictable results.

These prompts are intentionally designed for human review and decision-making. However, they could be easily rewritten to have AI output the corrected copy first with all the changes and rules cited below.

List of standardized prompts

  • Rule compliance
    Use: Validate copy against all writing content system documents.
    Prompt: Review the UI copy below against the FLOP language principles, word list, and writing guidelines:
    Heading: [text]
    Body: [text]
    Context: [optional]
    Expected output: A list of any violations with rules cited. Copy not rewritten by AI. Only errors are pointed out. Human judgment required for changes.

  • Legal check
    To check copy against a specific legal issue (e.g., a new law only in the UK), you can add a new legal document and deselect all other sources.
    Prompt:     Review the UI copy below against the UK Policy source:
    [text]
    Expected output: A list of any violations with rules cited. Copy not rewritten by AI. Only errors are pointed out. Human judgment required for changes.

Step 8: Results

Before

  • Terminology inconsistency

  • Language debates instead without a single source of truth

  • Unclear error patterns

  • High learning curve for new contributors

After

  • Consistent terms and fewer debates

  • Guidelines sped up reviews and approvals

  • Error copy has predictable, reusable patterns

  • New copy can be validated quickly using the system

What this enables

  • Faster onboarding for new team members

  • More consistent user experiences

  • Less time spent re-reviewing the same issues

Example of a corrections made as per new content system