When AI designs for you, your brand needs a checklist

This is moving fast. Not in the vague "AI changes everything" way. In a much more ordinary way: you are already sitting in a chat, asking for a campaign idea, and the same flow can now start making an actual Canva design.

That is useful. It is also where small teams can get messy fast. The important question is not whether AI can make a decent-looking image. The question is whether that image sounds like you, follows the offer, and can be published without someone regretting it later.

Today's AI signal: design is moving into chat tools

Canva says its Canva MCP Server now brings design creation directly into ChatGPT and Codex. In ChatGPT, users can create, preview, edit, and translate Canva designs without leaving the conversation. Canva also says Brand Kits can be used in the workflow, so colors, fonts, and brand elements are applied from the start.

MCP, Model Context Protocol, is a standard for connecting AI assistants to external tools and data. In plain terms, the AI chat does not just answer with text. It can talk to a tool like Canva and create something useful there.

Source: Canva Newsroom: Create with Canva inside ChatGPT and Codex

The same week, Canva announced a Perplexity integration. That one has a slightly different shape: Perplexity is used for research, analysis, and summaries, and Canva becomes the step that turns the material into presentations, campaign assets, infographics, or proposals. Canva says content creation is now the top reason small businesses use AI, and that 74 percent of small businesses rely on digital platforms to compete with larger players.

Source: Canva Newsroom: Canva comes to Perplexity Computer

Canva's own marketing and AI report adds the other half of the picture. Ninety-seven percent of marketing leaders in the survey use AI in daily creative work. At the same time, 70 percent of consumers say AI-generated ads often feel like something is missing, and 87 percent say the best advertising still needs a human touch.

Source: Canva marketing AI report

This is not only a Canva update. It is a signal about where everyday work is going. Design, research, copy, translation, and publishing are getting closer to the same conversation.

For small businesses, schools, and solo operators, that is attractive. It is also exactly where the work can lose its shape.

The risk is not an ugly AI image

An ugly image is easy to throw away.

The bigger problem is an image that looks almost right. The color is roughly right. The tone is almost yours. The offer is almost true. It only misses the sentence that helps the customer understand what is included, or the careful wording a school needs when sending information to parents.

That is how AI slop enters the workflow. Not as disaster. As small compromises.

If AI design moves into ChatGPT, Perplexity, Gemini, Claude, and other tools, the brand cannot only live in someone's head. It needs to exist as a small working routine:

• Which colors and formats are allowed
• Which words you often use
• Which words you avoid
• What promise you are actually allowed to make
• Who approves before anything is published
• Where the final version is saved
• How you disclose AI assistance when that matters

This does not need to become a 40-page brand manual. For many small teams, one page is enough.

Do this in 45 minutes: build an AI design checklist

Pick one recurring asset. Not the whole marketing function. Just one thing that often gets stuck.

It could be a lunch post for a restaurant, an open-house graphic for a school, a course announcement, a client-meeting deck, or a simple campaign for a new service.

Set a timer for 45 minutes and write the checklist before you start prompting.

1. Write what the material should do

One sentence is enough. For example: "Get previous customers to book autumn service" or "Explain the open house, so parents know the time, place, and next step."

If the purpose is unclear, the AI will fill the gap with generic marketing tone. It often sounds polished. It rarely helps.

2. Give the AI your fixed building blocks

Write down colors, fonts, image style, logo rules, and three examples of wording that sound like you. If you have a Canva Brand Kit, use it. If you do not, start with a simple text version.

Examples:

• "We write concretely and briefly."
• "We do not use hype words like revolutionary, groundbreaking, or magical."
• "We always say the next step: book, reply, visit, download, or ask."

3. Set boundaries for the offer

AI is good at making things sound more persuasive than they should. Write what the material must not promise.

For a consultant, that could be: "Do not promise results in 30 days." For a school: "Do not imply that a place is guaranteed." For ecommerce: "Do not say free returns unless it applies to every product."

4. Write the approval step

Decide who reviews the finished design. Make it a person, not "the team." If nobody owns the final check, the asset becomes a Slack message that everyone assumes someone else read.

Useful review questions:

• Is the offer true?
• Is the next step clear?
• Does the copy sound like us?
• Is anything sensitive, wrong, or too specific?
• Should this design become a template for next time?

5. Save the result as a reusable card

Put the checklist where the work actually happens: Notion, Google Docs, Canva, a project card, or your CRM. The point is that the next person should not start from zero.

This is Skill Forge in miniature: not a large AI platform, but a work habit that makes the AI result easier to trust.

Copy the prompt

Use this prompt in ChatGPT, Perplexity, Gemini, Claude, or directly in Canva. Replace the bracketed parts.

You are helping me create a first draft for a Canva design, but start by checking brand and purpose.

Material: [for example Instagram post, A4 flyer, presentation, school information]
Audience: [who will read it]
Purpose: [what the person should do next]
Offer or message: [what must be true]
Brand voice: [3-5 sentences describing the tone]
Visual rules: [colors, fonts, image style, format, Brand Kit if available]
Things we must not promise: [boundaries]
Approver: [person/role]

Do this:
1. Ask at most five control questions if anything is missing.
2. Suggest three design directions with a short reason for each.
3. Choose the most practical direction and create a first draft.
4. Write a review checklist for the approver.
5. Mark which parts must be checked manually before publishing.

Write concretely. Avoid advertising fluff. If something is uncertain, say what needs to be checked instead of guessing.

The point is not that the prompt is perfect. The point is that it forces purpose, boundaries, and review into the workflow before the design becomes publishable.

Three small workflows to try

Restaurant or shop: this week's campaign without copy chaos

Add the menu, opening hours, price boundaries, and image style. Ask AI to make three versions: one for social media, one for print, and one for a newsletter. Publish nothing until someone has checked price, date, and terms.

School or trainer: turn long information into a clear visual

Paste the approved information for an open house, course, or theme day. Ask AI to create a clear Canva layout with date, place, who it is for, and the next step. Let one person check the tone and confirm that no student or personal data slipped in.

Consultant or service business: from customer insight to one-page sales asset

Take a summary from a customer meeting or a common question. Ask AI to create a one-page pitch: problem, offer, example delivery, next step. Pay extra attention to promises, price hints, and wording that sounds more finished than the service really is.

Integrate for real, but keep the first version narrow

When Canva connects to ChatGPT, Perplexity, or other AI tools, the integration becomes useful when permissions are thought through.

Start with read access where that is enough. Use Brand Kits instead of pasting brand material into every prompt. If tools connect to files, campaigns, or publishing, use separate accounts, scoped API keys, a secret manager, or environment variables where that is technically relevant. Put an approval gate before anything goes live and keep a simple log: who asked the AI to create the asset, which source was used, who approved it, and where it was published.

That may sound dry. It is often the difference between a fun AI test and a workflow you can repeat without losing control.

The Hammer angle

For Hammer Automation's audience, this is a good first AI routine because it is visible. Everyone can see whether an image, presentation, or campaign feels wrong. That makes it easier to discuss than a hidden automation that sorts data in the background.

In Mindset Forge, we start with purpose, tone, and decisions: what should AI help with, and what should it never invent? In Tool Forge, we connect the right tools, such as Canva, ChatGPT, Perplexity, Google Workspace, or a CRM. In Skill Forge, we make the routine repeatable: prompt, checklist, template, approval, and log.

Start with one design that already has to be made this week. Write the checklist first. Let AI help after that.