How-to: Create a custom instruction set for consistent brand voice in outputs.

Another day, another "game-changing" feature announcement from an AI vendor. This time it's Le Chat and their custom instructions. The hype cycle suggests this will finally solve your brand voice woes, as if the problem was ever the tool's capability and not your own inconsistent inputs.

Let's be real. You can get a decently consistent tone out of almost any of these models if you stop treating them like a magic wand and start treating them like, I don't know, a tool that requires a manual. The issue is rarely that Marketo, HubSpot, or Pipedrive can't enforce a style guide—it's that the guide is a 50-page PDF no one reads, and the prompts you feed the AI are a chaotic mess of "make it sound professional but cool, you know?"

So, for Le Chat. The custom instruction field isn't a silver bullet. It's a parking spot for the fundamentals you should have been defining anyway. Forget the poetic brand manifestos. You need concrete, operational rules.

Think like you're briefing a very literal junior copywriter who has never heard of your company. Specify:
* Sentence length. Short? Medium? Avoid complex clauses?
* Vocabulary bans. List the forbidden marketing buzzwords. "Leverage," "synergy," "disruptive" – you know the ones.
* Active vs. passive voice. Mandate one.
* Key phrases to use. Your actual product names, your standard value props.
* Formatting rules for things like headings, lists.

For example, if your brand voice is "direct and unpretentious," your instruction might read: "Use short, declarative sentences. Avoid adverbs and jargon. Never use 'leverage' or 'solution.' Always refer to our product as 'the platform,' not 'the ecosystem.' Use bullet points only for lists of three or more items."

Test it. Feed the same messy draft brief with your custom instructions on and off. If the difference isn't stark, your instructions are too vague. This is integration work 101: garbage in, slightly more polished garbage out. The tool works fine. The question is whether you'll do the work to use it properly. 😏