// Blog post — educational, describes real features (AI Generator,
// Performance-Aware Prompting, Generation History, waim_ai_review_before_send)
// grounded in features.jsx / CLAUDE.md, not invented.

const POST_SOURCE = `# How AI Ad Copy Actually Works Here (and Where Human Review Still Matters)

"AI writes your ads" undersells what's actually happening, and oversells what AI is good at. Here's the real pipeline, and the one setting we think every store should leave switched on.

## It starts with your product data, not a blank prompt

Generating copy from a vague brief typed into a chat window produces vague copy. The AI Generator instead starts from your actual product record — name, description, price — plus a campaign goal, a tone, and the target channel (Meta, Google, or email). Claude (Anthropic's model) writes from that context, not from a guess at what your store probably sells.

## Multiple variants, not one best guess

A single AI output is just a faster first draft — it's still one person's (or one model's) opinion of the best angle. Every generation returns multiple headline and copy variants in one call, so the choice is between options you can actually A/B test, not a single draft you either accept or rewrite from scratch.

## Performance-aware prompting

Copy that ignores what already worked is copy that relearns the same lessons every time. On the Growth plan and up, new generations are automatically enriched with your store's historical creative performance data — so a prompt isn't starting from zero, it's building on what's already been proven to work for your specific audience.

## Every generation is logged

Generation History records the prompt, the output, the channel, and the product behind every single AI generation — browsable, and re-publishable to Meta or Google later. Nothing about the pipeline is a black box you have to trust blind; you can go look at exactly what was generated and why.

## Human review is a setting, not an afterthought

This is the part we'd push back on if a tool skipped it: "Require human approval before any send" is a real setting (not a hidden default), and we'd recommend turning it on for any store where a wrong send has real cost — which is most stores. AI drafts fast. It doesn't know your brand's specific sensitivities, a live PR situation, or that a particular phrase reads wrong to your specific customer base. That judgment call stays yours.

## What AI still isn't good at

To be direct about the limits: AI-generated copy is a strong first draft, not a finished campaign strategy. It won't tell you *which* segment to target for a re-engagement push as well as someone who actually knows your customers (though Audience Advisor gets closer than a blank spreadsheet). It won't catch a factual error about your own product unless the underlying data was correct to begin with. And it has no opinion about brand risk — that's exactly why review-before-send exists.

The honest framing: AI collapses the mechanical part of the job — assembling a brief, writing a first draft, generating variants — from hours to seconds. The judgment part was never the bottleneck an agency retainer was solving anyway.
`;

const InfernoBlogAiCopy = () => (
  <ContentPage
    title="How AI Ad Copy Actually Works Here"
    kicker="Blog · How it works"
    active="Blog"
    source={POST_SOURCE}
    sibling={['← All posts', 'blog.html']}
    cta={{ text: 'See it generate copy from your own catalog', href: 'index.html', label: 'Start 14-day trial →' }}
  />
);

Object.assign(window, { InfernoBlogAiCopy });