AI Content Creation
0/24 complete

Module 05 · Visual and Copy Production

Building a Content Style Guide for Consistency

15 minfocused lesson6practical steps4grounded questions3source links
Open lesson + course map

On this lesson

Course outline

A content style guide turns creative decisions into observable rules. It tells a writer, designer, or AI what to make, avoid, and verify.

By the end, you will have a complete one-page guide covering audience, voice, language, claims, evidence, visuals, formats, calls to action, and ownership. A new helper should be able to use it without a verbal briefing.

// concept

Write Rules That Can Be Checked

Replace adjectives with behaviours. “Friendly” is open to interpretation; “use everyday words and one direct question; avoid corporate openings” is reviewable. Give each voice attribute a do, a do not, and an example.

Weak instructionObservable rule
“Sound natural”Use first-hand observations only when the author recorded them; label fictional scenarios as samples.
“Keep visuals clean”One headline, one supporting line, one focal image, and no more than two typefaces per card.
“Make people engage”End tutorials with one specific experience question; never request a keyword comment merely to unlock information.
“Mix Urdu naturally”Keep technical terms in English; use Roman Urdu for spoken reactions; use hai, not alternating between hai/he.

Define which claims need evidence. Prices, policies, health statements, and market claims require a current primary link and access date; customer quotes need permission.

// concept

Use This One-Page Style-Guide Template

Copy this into a shared document and keep it scannable on a phone. A shared document is enough; if you transfer rules into Canva Brand Kit, Guidelines, or Controls, verify current plan availability on Canva’s official page.

// template — copy me46 lines
# [Brand] Content Style Guide
Owner: [name/role] | Version: [number] | Last reviewed: [date]
Next quarterly review: [date] | Approval contact: [name/role]

AUDIENCE
- Segment: [location, context, need]
- Prior knowledge: [what they know]
- Desired action: [observable action]

VOICE
- Attributes: [each with do / do not / preferred / rejected line]
- Sentence rules: [length, jargon, person, punctuation]

LANGUAGE
- English/Roman-Urdu ratio: [guideline]
- Keep in English: [terms]
- Spellings and accessibility: [approved Roman Urdu, Urdu font, captions, alt text]

CLAIMS AND EVIDENCE
- Source required for: [claim types]
- Evidence: [primary link, first-party record, consented quote]
- Record as: [URL/file + publisher + access date]
- Unknown: [omit or label]
- AI rule: [never invent facts, quotes, people, or results]

VISUALS AND FORMATS
- Colours/typefaces/logo: [values and usage]
- Card/image rules: [headline, image, margin, contrast, consent]
- Approved formats: [tutorial, checklist, story, comparison]
- Platform adaptations: [Reel, carousel, LinkedIn, YouTube]

CTA RULES
- New / returning / sales audience: [appropriate next step]
- Banned: [engagement bait, false urgency, unsupported promise]

BANNED LIST
- Clichés and claims: [specific banned formulations]
- Owner-approval topics: [list]

PREFERRED / REJECTED EXAMPLES
- Hook / claim / visual / CTA: [preferred] / [rejected + reason]

DELEGATION CHECK
- Helper creates: [one test asset]
- Reviewer checks: [voice, language, evidence, visual, CTA]
- Guide is usable when: [pass condition]

// concept

Make AI Follow the Guide, Then Test the Handoff

Attach the guide and source pack to each request.

// prompt — copy me4 lines
Create one 6-card carousel from the supplied brief and style guide.
Use only facts in <sources>. For each factual card, return its source ID.
Return: card copy, visual direction, caption, CTA, and a rule-check table.
Write NEEDS SOURCE for missing facts. Quote the guide rules used for voice, language, and CTA.

Give a new helper the guide, brief, and sources—but no coaching. Revise any ambiguous rule and add a preferred/rejected pair. Review quarterly and after a positioning or platform-policy change.

// worked_example

Worked Example

This is a fictional sample for “Seedhi Dukaan,” an imaginary WhatsApp-first home-organisation shop.

// template — copy me12 lines
# Seedhi Dukaan Content Style Guide
Owner: Content lead | Version: 1.0 | Last reviewed: 18 July 2026
Next quarterly review: 18 October 2026 | Approval: Store owner
Audience: Urdu-speaking Karachi apartment renters, 24–40, who can shop by phone; action: measure one shelf before choosing.
Voice: practical, calm, short instructions; explain one trade-off; never shame a messy home.
Language: about 70% plain English / 30% Roman Urdu; dimensions stay English; use “hai”; Urdu exports need a phone check.
Claims/evidence: dimensions and prices use the linked, dated product sheet; unknowns are omitted; AI invents nothing; customer words need consent.
Visuals/formats: navy #17324D, cream #FFF8E7; one sans-serif; one consented photo; 32 px margin; captions; tutorials, comparisons, checklists.
CTA: new readers—“Save the three-step checklist”; product posts—“Message for the current price and availability.”
Banned: “every Pakistani home,” miracle claims, fake countdowns, exposed addresses, chai/rishta shorthand; religion/politics require owner approval.
Preferred: “Measure the shelf width before choosing a box.” Rejected: “Every Karachi kitchen needs this.” Reason: unsupported generalisation.
Delegation pass: helper makes a sourced 6-card comparison with all five rule groups passing.

Draft one said, “Transform your home with our amazing organiser.” It has no measurable action or evidence. The fix is: “Measure your shelf depth, then compare it with the dimensions on the product card. Message us for current availability.” It uses approved evidence and the sales CTA rule.

// failure_cases

Failure Cases to Diagnose

6 cases to diagnose

  • Voice labels have no boundaries

    “bold, warm, human” produces inconsistent copy. Add behaviours and paired examples.

  • The language ratio fights the sentence

    forced Roman Urdu enters technical instructions. Specify language roles and spellings.

  • Visual rules omit values

    “use brand colours” cannot guide a helper. Add colour codes, typefaces, margins, and image treatment.

  • Sources lack dates

    an old price or policy is reused. Require publisher, link, and access date.

  • The banned list is abstract

    “avoid clichés” catches nothing. List the actual phrases, claims, and recurring angles reviewers reject.

  • The guide needs live coaching

    repeat the no-coaching test and record each ambiguity.

// pakistan_angle

Pakistan Angle

Define the audience beyond a national label. A Karachi renter, Lahore applicant, and Peshawar family-business buyer may differ in language, references, and buying process. Record the region and context. Do not use chai, rishta, load-shedding, or “desi parents” as automatic relatability; approve only audience-relevant references that avoid stereotypes.

For English/Roman-Urdu content, standardise hai versus he and use a fluent reviewer. Test Urdu or Nastaliq on a low-cost Android export because shaping varies by font and app. Get permission before showing faces or WhatsApp messages; hide phone numbers, addresses, payment screenshots, and CNIC details. Write prices in PKR from a current product sheet.

// hands_on

Hands-On Exercise

6 steps

  1. Paste the full template into one mobile-friendly document.

  2. Fill every field from decisions made in earlier lessons; write owner decision needed where evidence is missing.

  3. Add at least one preferred/rejected pair for voice, claim, visual, and CTA rules.

  4. Run the drafting prompt using a real, authorised brief and source pack.

  5. Give the guide and inputs to a new helper without coaching. Record every question they ask.

  6. Fix ambiguous rules, assign the owner, and set the next quarterly review date. Done means the one-page guide is complete, a helper produced one test asset from it, and the reviewer could mark voice, language, evidence, visual, and CTA compliance without guessing.

// completion_rubric

Completion Rubric

5 checks — tick as you verify

0/5

// sources

Sources

// check_yourself

Check yourself

4 questions · answers and options are taken word-for-word from this course

0/4
  1. 1 / 4 · diagnose

    Your work shows this failure mode: “Voice labels have no boundaries.” The lesson describes it like this: ““bold, warm, human” produces inconsistent copy.” What does the lesson tell you to do about it?