You have read the LinkedIn post that says “78% of marketers report productivity gains from AI.” You have shared it. Six months later, your CMO asks where the figure comes from in a vendor review and the trail goes cold — the post cited an aggregator that cited a vendor blog that cited a survey that no longer resolves.
This is the failure mode the Playbook’s evidence standard is built to prevent. The full standard is on the public site; the version-in-three-rules below is the operator’s-desk summary.
Rule 1 — every figure has a named primary source
Not “studies show”. Not “research suggests”. Name the organisation, name the year, link the publication. McKinsey’s State of AI survey, Stanford HAI’s AI Index Report, the Salesforce State of Marketing — these have names and dates. “Industry experts agree” does not.
The test: can you click through to the source from the citation, and does the figure match? If yes, ship it. If not, find a primary source or cut the claim.
Rule 2 — methodology matters more than the figure
A 78% figure from a 200-respondent self-selected vendor survey is not the same evidence as a 78% figure from a 20,000-respondent peer-reviewed study. Cite the methodology when it differs from the reader’s default assumption.
In practice: “Adobe’s 2024 study of 348 US marketers found…” beats “Adobe’s 2024 study found…”. Six extra words; the reader can weigh the figure correctly.
Rule 3 — the date check kills half of stale citations
AI-tooling statistics rot fast. A 2022 figure on ChatGPT adoption is from a different era. A 2023 figure on multimodal performance is already a half-truth.
The Playbook’s rule: any figure older than twenty-four months has its year named explicitly in prose. Any figure older than thirty-six months gets replaced with a fresher source where one exists. Any figure predating November 2022 (ChatGPT public launch) is framed as belonging to a different era.
What this earns you
A piece of marketing copy that survives a vendor review, a procurement question, or a journalist’s fact-check. Specifically:
- The CMO who asked “where is this figure from” closes the question in five minutes, not five hours.
- The legal review that lands on AI-generated copy has a sourceable claim to anchor against.
- Your team’s content does not look interchangeable with the LinkedIn-thread economy that the Playbook is trying to step out of.
The full evidence standard — three tiers, the verified case-study register, the rules on vendor-published cases — is at /standards/evidence. The case study standards page shows the bar a register entry has to clear, with two worked examples; the full register reads inside the subscriber library.