Make Best Practices Stick to Real Work
Generic advice is easy to write and impossible to apply
You ask the assistant to review a client onboarding guide against UX best practices. It produces a list: 'use clear headings, keep paragraphs short, include a call to action, ensure mobile responsiveness.' Every item is correct. None of them tell you what to change about your specific guide.
The problem is that generic standards have no hierarchy and no . 'Use clear headings' and 'ensure WCAG AA color contrast' are both best practices, but one is a suggestion and the other is a legal compliance requirement. Without hierarchy, the assistant treats them as equally important. Without , it cannot tell which standards apply to your guide and which do not.
This chapter teaches you to build a best-practice checklist that the assistant can apply to any : a guide, a website, a report, a process, a communication, or a project plan. The checklist has three tiers (non-negotiable, -sensitive, and nice-to-have), includes the reason behind each standard, and produces pass/partial/fail ratings the assistant can use to review its own work.
Three tiers give every standard a priority level
The essential move is separating standards into three tiers. Without tiers, the assistant treats every best practice as equally important, which means it either flags everything (overwhelming you with minor issues) or flags nothing (deferring to your judgment on items it should have caught).
Non-negotiable standards block delivery. If the guide lacks contact information, it is not ready for the client, regardless of how polished the prose is. -sensitive standards improve quality but have legitimate exceptions. Nice-to-have standards are suggestions the assistant surfaces when the is already passing the required tiers.
Each standard needs a reason, or the assistant applies it mechanically
A standard without a reason is a rule the assistant follows blindly. 'Keep paragraphs under five sentences' is a standard. 'Keep paragraphs under five sentences because long blocks of text cause readers to skip content' is a standard with a reason.
The reason matters because it helps the assistant handle edge cases. A paragraph with six sentences that walks through a connected example might be fine, because the reason behind the standard (preventing readers from skipping) does not apply when the content is engaging and sequential. Without the reason, the assistant would flag it as a violation.
Pass, partial, and fail ratings replace subjective feedback
When the assistant reviews an , it rates each standard as pass (fully met), partial (close but needs adjustment), or fail (not met). Each rating includes a specific citation: the passage or element that passes, the element that partially meets the standard and why, or the element that fails and what would fix it.
This replaces the vague feedback loop ('this could be better') with a specific review ('Standard 3 fails: the FAQ section is missing. The standard requires a FAQ for guides over three pages. This guide is eight pages. Add a FAQ covering the five most common client questions.').
The suggested fix is what makes the review actionable. A fail rating without a fix is a problem report. A fail rating with a fix is a task. The assistant should produce tasks, which is the same principle behind the correction loop from earlier in the book.
When two standards conflict, the tier and reason resolve the tension
Standards sometimes pull in opposite directions. 'Keep the guide under ten pages' conflicts with 'include a FAQ section' when the guide is already at nine pages. 'Use simple language' conflicts with 'use precise technical terms' when the audience is mixed.
The three-tier hierarchy resolves most conflicts: non-negotiable standards win over -sensitive ones, which win over nice-to-haves. When two standards at the same tier conflict, the reason behind each standard helps the assistant decide which matters more for this specific .
When the assistant cannot resolve a conflict, it should flag it: 'Standards 4 and 7 conflict for this . Standard 4 (keep under ten pages, -sensitive) and Standard 7 (include FAQ, context-sensitive) are both context-sensitive. The FAQ would push the guide to eleven pages. Recommendation: include a condensed FAQ (half page) that covers the top three questions instead of five.' You approve or override.