Every marketer hoards notes. Mine were spread across Notion, Google Drive, Slack and local folders: 80+ items that I could never find when I needed them.
So I rebuilt it as an LLM-readable wiki: every page has a type (client, campaign, concept, project, source), tags, and links. Claude reads it through MCP. It can search by client, pull a campaign's real numbers, or assemble a context pack for any task.
The payoff is compounding. When I write a proposal, the AI pulls the client's history. When I write content, it pulls real campaign numbers instead of inventing them. This website's case studies were written from that vault, every figure traced to a source page.
The hard part isn't the tooling, it's the discipline: one fact per page, a type for everything, and a rule that nothing gets cited unless it has a source. The same rule that makes a wiki useful makes an AI honest.