The prompt. Writing one Siren can render
The prompt is the only natural-language input Siren takes from you. Everything else (template selection, copy direction, visual variation) is determined from there. A good prompt does not need to be long. It needs the angle.
Anatomy of a prompt that works
The Reasoner reads your prompt and your Brand DNA together. It needs three things from the prompt:
- The topic. What this is about. A product feature, an announcement, a take on an industry trend, a milestone.
- The angle. The specific frame you want. Not "Q2 update" but "Q2 update, contrarian angle: we shipped less and it performed better." The Reasoner builds the whole campaign around the angle. It needs something to build on.
- Optional: format hint. If you have a firm view of asset type or tone, include it. "Three-slide carousel, educational register." The Reasoner can pick for you. A hint helps when you already know the output.
Tip. One sentence is enough. The Reasoner is not a prompt engineer. It is a structured-output engine. Give it the angle. Brand DNA fills the rest.
Examples
Weak prompts and stronger versions:
- Weak: "Write about our new feature." Strong: "Announce the Reasoner, our two-stage generation engine that stress-tests every draft before it ships. Target: indie SaaS founders skeptical of AI content tools."
- Weak: "Post something for Q2." Strong: "Q2 wrap: 3 clients, 18 campaigns, 0 agency spend. Builder archetype, stat-forward."
- Weak: "Promote our pricing." Strong: "Why we don't run a free-forever plan. Honest economics. Every render costs real compute. Show the math rather than pretend it is free."
- Weak: "Video about the product." Strong: "30-second product demo loop: prompt in, campaign renders, approved post appears. No voiceover. Let the pipeline speak."
What the Reasoner does with your prompt
The Reasoner runs a two-stage process:
- Routing (Sonnet): reads your prompt and Brand DNA and picks a template from the ones your plan allows. The Reasoner cannot pick a template outside your tier.
- Designing (Sonnet): fills the SirenSeed, a structured JSON object with copy and visual direction. This is where your archetype, banned words, and required disclaimers are enforced. The Reasoner cannot produce a seed that violates your Brand DNA.
The SirenSeed is then handed to the renderer. The renderer has no access to your prompt. It only sees the structured seed. Natural-language drift stops at the Reasoner.
What the Reasoner cannot do
- Ask for video on Spark. The full motion library is available on Pro and above. Spark stills still render.
- Override your banned words. They are enforced at seed generation. Any draft that contains a banned word is regenerated internally before the seed is returned.
- Write your required disclaimers. Disclaimers are stamped by the renderer at a layer below the Reasoner. The Reasoner knows they exist. It does not author them.
- Generate an asset that requires photo uploads if you have not uploaded photos. App-showcase and portfolio-case-study templates require screenshots from your Prompt Assets library.
Prompt length
Shorter is often better. One strong sentence beats a paragraph of context. The Reasoner's context window can handle a detailed prompt, but more length does not guarantee better output. Extra length can dilute the angle. If the prompt is longer than three sentences, check whether you are adding context or hedging.