01 · THE PROBLEM

AI search was telling the wrong story.

AI-generated search summaries were sometimes citing forum threads and affiliate sites over a brand’s own content, even when the brand’s information was better. In one case, a summary implied a product was only sold in physical stores, ignoring the brand’s own online shop.

That’s as much a content and structure problem as a search one, and it’s hard to grasp from slides.

02 · THE IDEA

Be the system.

After a colleague’s presentation on AI search, the fastest way to make it click was to let teams play the AI themselves.

1. Each team gets a fixed token budget, like an AI’s limited context window.
2. They choose from source cards of mixed quality: brand pages, forum posts, comparison sites, marketing copy and product specs.
3. They write a search summary in which every key claim needs evidence.
4. Midway, new evidence arrives, and they have to revise.

03 · WHAT TEAMS LEARN

Clear facts get quoted. Vague copy doesn’t.

• Write simple, specific facts that can be quoted safely.
• Make every way to buy explicit and easy to find.
• Add structured sections such as FAQs, “where to buy” and constraints.
• Include trade-offs and edge cases, so AI doesn’t fill gaps with generic language.
• Build pages that still make sense if they’re the only thing retrieved.