How to run a content audit that actually finishes.

You've started one before. A spreadsheet, a tab you meant to return to, a column called "notes" that filled up while the column called "decision" stayed empty. It stalled around row forty, and the library it was supposed to sort is still unsorted.

Audits don't die from lack of effort. They die from missing structure — usually three specific pieces of it: no defined scope, no scoring rubric, and no rule for turning scores into decisions. Without those, an audit is just inventory with nowhere to go, and inventory is boring enough that anything else on your calendar will beat it.

Here's the structure that survives, in the order that matters.

1. Scope, in one sentence

Before you open a spreadsheet, write down what you're auditing in a single sentence: a site section, a product area, a channel. If it doesn't fit in one sentence, it's too wide — and too wide is the first way audits die. You can run a second audit later. You can't finish an unbounded one now.

2. Inventory before judgment

One row per asset: title, URL, type, owner, last updated. Nothing else yet. The temptation is to evaluate as you go — resist it, because evaluation is slow and inventory is fast, and mixing them means doing both badly. Get everything in scope into the list first. This is also the step an AI agent does better than you do: reading a site section and producing rows is labor, not judgment.

3. Score against a rubric, not a mood

The reason two people disagree about the same page is that "good" isn't defined. Define it: four dimensions — Accuracy, Relevance, Findability, Quality — each scored 1 to 4 against written scale points. "Mostly accurate, minor details need updating" is a 3 whether it's Tuesday or the end of the quarter, whether it's your page or someone else's. The rubric is what makes the audit defensible later, when someone asks why their page got cut.

4. Let the score point to a decision

Every asset gets exactly one of four calls: Refresh, Repurpose, Retire, or Leave. The total score points to one — 12 to 16 leaves, below 8 is retire-or-repurpose territory, and any Accuracy of 1 retires the asset no matter what the total says, because a page that's wrong is doing damage regardless of how well it's written. Ranges are guides, not rules. But they turn a hundred open judgment calls into a hundred confirmations, and that's the difference between finishing and stalling.

5. Write the reason down

One sentence per decision — a justification, not a description. "Retiring: duplicates the onboarding guide and hasn't been updated since the tool it describes was replaced." Six months from now, someone will ask why. The reason sentence is the part of the audit that ages well, and it's the part almost everyone skips.

6. Route every decision somewhere

A decision that goes nowhere is a meeting that never ends. Refresh and Repurpose decisions should leave with a reference ID and become briefed work. Retire gets a retirement date. Leave gets a review date, because "fine for now" has an expiration. No row is done until it has a next step attached.

Where the AI fits

An agent can read every asset, suggest all four scores, and draft the reasons — the labor that kills audits. What it shouldn't do is set the decision. An audit where every call was made by a machine and signed off unread isn't an audit; it's a list. The useful boundary: the agent suggests, you decide, and the structure enforces that instead of trusting discipline to.

Kyle Simmons builds governed content systems at Fully Inhabit. Fifteen years across financial platforms, universities, and regulated tech; MFA, UNC Wilmington. The systems in these notes are the ones he sells — and runs.