I keep seeing the same mistake with AI and creative work. People start with the tool, then spend their time correcting the output. The model can produce a neat layout, a fluent paragraph, or a plausible image, but it doesn't know what your business sounds like, what your customers notice, or which shortcuts make the work feel cheap. So it fills the gap with an average answer.
That's not a prompting problem. It's a brand memory problem. If the important decisions only live in your head, AI has no reliable way to make them for you. You get a different version of your business every time you open a new chat.
The five-minute brand test
Before I ask AI to write or design anything, I want five things written down: who the work is for, what the business believes, how it should sound, what it should never say, and a couple of examples of work that feels right. It isn't a 40-page brand book. It's a working brief someone else could pick up and use without sitting beside you.
The examples matter most. A list of adjectives like modern, friendly and premium gives a model very little to work with. A real example, with a sentence on why it works, is far more useful. So is a bad example with the reason it fails. You're not trying to describe your taste perfectly. You're making your judgement visible.
The reusable brief
I turn that material into a short brief I keep on hand and reuse. It sits above the individual task, so the context doesn't disappear when the project changes. Paste it at the top of a new chat, or save it as a custom instruction the AI loads on its own. For a design team, that might cover spacing preferences, visual references, words to avoid, and the point where a concept turns generic. For a small business, it's often the difference between a reply that sounds like the owner and one that sounds like software.
This step alone moves a team from ad hoc AI use to something more deliberate: a shared starting point everything else can build on. It doesn't do the work automatically, but it gives every workflow after it something solid to follow.
Here's a simple test: give the same brief to two people, or run the same AI tool twice, and compare the decisions it makes rather than the polish. If the work still feels generic, the brief's missing something. Add another example, a constraint, or a clearer reason, and try again. Do that a few times and you'll have something far more valuable than a clever prompt: a small piece of your own judgement that travels with you.
Also in this issue
The news I picked out on 4 August 2026. Some of it may have moved on since.
- Running the businessJudge AI by completed work, not how often people use it
OpenAI's now measuring completed work and reliability instead of how often a tool gets opened. A good test for any AI step in your own workflow: is it actually finishing design tasks well, or just getting used a lot?
- Working with knowledgeGoogle is folding NotebookLM into its main Gemini workspace
If you keep brand references, past briefs or client notes in NotebookLM, expect it to feel more like your everyday Gemini tools soon. Worth checking your source material's still organised the way you'd want an AI reading from it.
- Creative workCanva can now turn AI-built websites into editable designs
Canva Code 2.0 lets you import HTML, change the layout visually and collaborate in the same editor. That makes rough prototypes easier to hand around, but someone still needs to check the experience before it goes live.