On-brand AI

​

Ground AI in real brand data

import Anthropic from "@anthropic-ai/sdk";

const anthropic = new Anthropic(); // reads ANTHROPIC_API_KEY

async function answerAboutBrand(domain: string, question: string) {
  // 1. Fetch the brand context; Markdown drops straight into a prompt
  const context = await fetch(`https://api.brandfetch.io/v2/context/${domain}`, {
    headers: {
      Authorization: `Bearer ${process.env.BRANDFETCH_API_KEY}`,
      Accept: "text/markdown",
    },
  }).then((res) => res.text());

  // 2. Ground the model: the brand profile becomes the system prompt
  const message = await anthropic.messages.create({
    model: "claude-opus-4-8",
    max_tokens: 1024,
    system:
      `You answer questions about the company behind ${domain}. ` +
      `Ground every answer in the brand profile below; if the profile doesn't cover the question, say so.\n\n${context}`,
    messages: [{ role: "user", content: question }],
  });

  return message.content
    .filter((block) => block.type === "text")
    .map((block) => block.text)
    .join("");
}

console.log(await answerAboutBrand("patagonia.com", "What does Patagonia sell, and to whom?"));

​

Why not just ask the model?

• Long-tail coverage. Models are weak on small businesses, regional companies, and brands that launched or rebranded after their training cutoff. Brand Context resolves domains live when they’re not yet in the dataset, so coverage extends to brands no model’s memory holds.
• Consistent structure at scale. Run enrichment over 10,000 companies and you need the same fields back every time. Request Accept: application/json and every response follows the same schema: no reformatting between calls, no invented fields, no phrasing drift. At scale, that consistency is the product.
• Cost and latency. A cached, deterministic API call is faster and cheaper than an LLM inference per lookup, and the difference compounds when you’re personalizing 50,000 landing pages by visitor domain.
• Freshness. Taglines, audiences, and messaging get updated by marketing teams all the time. A live-resolved profile reflects this month’s positioning; a model reflects whatever was true when it was trained.
• Auditability. Every response is tied to a fetchable source rather than a model’s recollection, and for compliance-sensitive tools in sales or finance, “the model guessed” isn’t good enough.

​

Implementation

curl "https://api.brandfetch.io/v2/context/stripe.com" \
  --header "Authorization: Bearer YOUR_API_KEY"

​

Brandfetch MCP

​

Pricing

​

Measuring impact

​

Going further