A monthly content audit should end with decisions, not a screenshot folder. Its job is to compare what you intended to publish with what actually reached people, cost money or time, created risk, and taught you something useful. The ten questions below produce a one-page operating record and a small set of actions for the next month.
Platform numbers are not interchangeable. Definitions, attribution windows, interfaces, privacy thresholds, and view-count methods change. YouTube, for example, announced a public-view counting change in August 2026 while retaining the earlier concept as “Engaged views” in Advanced Mode. Freeze the metric definition and export date with every audit instead of treating a label such as “views” as timeless.
Prepare one trustworthy audit table
Set a fixed monthly cutoff in one time zone. Export available platform data after allowing for known reporting delays, and keep the untouched exports beside the audit. YouTube’s current Advanced Mode supports date controls, content groups, comparisons, filters, and exports; equivalent capabilities vary by platform and account.
Create one row per published item and include:
- platform and permanent content ID or URL;
- publication date and format;
- content pillar, audience task, and campaign;
- planned outcome and primary metric chosen before publication;
- comparable exposure, attention, action, and revenue fields where available;
- production hours, external cost, paid distribution, and gifted inputs;
- rights, disclosure, accessibility, correction, and takedown status;
- one controlled variable or “no experiment.”
Leave unavailable data blank rather than converting it to zero. Mark estimates, currency, tax treatment, attribution source, and any manual calculation. Do not export or circulate demographic or audience data beyond what the team is authorised to use.
1. What did we promise to produce, and what actually shipped?
Compare the calendar with the published record. Count completed, delayed, cancelled, and unplanned items by format and pillar. For each difference, record the cause: missing approval, underestimated edit, rights problem, unavailable subject, technical failure, urgent opportunity, or intentional scope change.
The decision is not automatically “publish more.” Repeated delay in one format may show that its workflow is unrealistic. An unplanned item may deserve a repeatable slot if it served a clear audience need without displacing higher-value work. Carry forward only work that still has a current purpose and owner.
2. Which audience job did each item complete?
Restate the viewer’s task in plain language: learn one process, compare options, solve an error, decide whether to buy, follow a story, or be entertained. Then inspect comments, search terms, saves, support questions, replies, and completion behaviour for evidence that the promised job was understood.
Do not call all positive reactions proof. A high-reach entertainment clip may fail as a tutorial; a lower-reach answer may prevent repeated support work. Classify feedback as observation, request, objection, confusion, or outcome report. Choose one unresolved audience question to investigate next month.
3. Are we comparing like with like?
Before ranking content, control for format, topic, distribution, age, publication time, paid support, audience size, and measurement window. YouTube’s guidance recommends comparing videos of the same format and supports groups plus equal-lifespan comparisons. The principle applies elsewhere even when the tools differ.
Compare first-seven-day performance with first-seven-day performance, not a new upload with an evergreen article that has accumulated traffic for a year. Separate paid and organic distribution where the source allows. If the cohort is too small or materially different, label the result inconclusive.
4. Where did attention stop matching the promise?
Inspect the path from impression or feed start through meaningful attention. Depending on format, this may include click-through, engaged views, early retention, average view duration, completion, scroll depth, or return behaviour. Use only metrics whose definitions you can document.
Review the content at the largest relevant drop or abandonment point. Ask whether the opening promised something the body delayed, whether context was missing, whether pacing obscured meaning, or whether the viewer’s question had already been answered. Retention is evidence of behaviour, not a verdict on artistic quality or a proof of cause.
Choose one edit-level hypothesis for a future matched test. Do not recut the whole system from a single curve.
5. Which actions mattered after attention?
List the downstream action appropriate to the content: qualified click, email signup, product-page visit, enquiry, download, application, sale, repeat visit, or another content item viewed. Check tracking health before interpreting performance. Broken links, missing tags, consent settings, and cross-device loss can make an action appear absent.
Separate platform-reported attribution from first-party records and final financial data. Use conservative language: “associated with,” “reported by,” or “occurred after” unless the design supports a causal claim. Record what cannot be reconciled instead of forcing every system to match.
6. What did each useful outcome cost?
Add production labour, contractors, software allocated to the project, licences, travel, props, shipping, paid media, revisions, and ongoing support. Record the estimation method for internal time. Compare both total cost and cost per relevant outcome, but do not let one ratio hide capacity or quality.
Include opportunity cost qualitatively: what was delayed because this item required three revision rounds? A profitable sponsorship may still damage the calendar; a reusable reference article may be valuable before direct revenue appears. The result should identify one cost to reduce without weakening rights, safety, accessibility, or review.
7. Which workflow step caused avoidable rework?
Trace one item from brief to archive. Count major revision loops and classify the cause: unclear promise, missing source, late stakeholder, wrong aspect ratio, unlicensed asset, caption repair, export mismatch, or disorganised handoff. Fix the earliest controllable cause rather than asking the editor to work faster.
Assign one workflow experiment with an owner and pass condition—for example, approving the claim sheet before filming or checking music scope before the rough cut. If the change cannot be evaluated next month, make the gate more concrete.
8. Did anything fail a trust, rights, safety, or accessibility gate?
Review every correction, complaint, muted track, copyright claim, expired permission, missing disclosure, unsafe demonstration, caption error, privacy issue, and inaccessible asset. Sample apparently successful posts too; low complaint volume is not proof of compliance.
FTC guidance says material relationships with brands should be disclosed clearly and in a way people are likely to notice, and unsupported product claims remain problematic even when an endorser repeats them. Check disclosure in the content itself where required, not only in metadata. Verify licences and releases against the actual platforms, territories, paid uses, clients, and publication term.
For video, review caption accuracy, meaningful non-speech audio, speaker identification, contrast, placement, flashing, and essential visual information. W3C guidance explains that captions carry speech and necessary non-speech information, while descriptive transcripts can include visual information needed for understanding. Open remediation tasks before planning new volume.
9. What did the controlled tests actually teach us?
List experiments registered before publication: hook wording, thumbnail, call to action, duration, structure, publishing time, or another single variable. Confirm the variants served comparable audiences under comparable conditions and that the success metric was chosen in advance.
Distinguish “winner,” “directional signal,” and “inconclusive.” Multiple simultaneous changes do not identify which change mattered. Small samples, seasonality, news events, recommendation shifts, and paid boosts can overwhelm the tested variable. Preserve losing results so the team does not repeat the same uncertain test later.
Turn each defensible learning into a bounded rule: “Use this opening for troubleshooting clips and retest after six examples,” not “all videos need this hook.” Set an expiry or review count for every new rule.
10. What will we continue, change, stop, and deliberately test next month?
End with four short lists. Continue only the formats or practices supported by audience value, strategic fit, manageable cost, and acceptable risk. Change one bottleneck with a named owner. Stop an item, series, tool, or meeting that lacks a current purpose or repeatedly fails a gate. Test one uncertain assumption with a comparable design.
Every action needs a due date and observable completion condition. “Improve retention” is not an action; “produce two matched tutorials with the same body and different evidence-first openings, then compare equal seven-day engaged-view and completion windows” is testable. Avoid committing to an outcome the team does not control.
The one-page monthly decision record
Summarise the audit under these headings:
| Field | Required answer |
|---|---|
| Period and cutoff | Exact dates, time zone, export timestamp |
| Data scope | Platforms, accounts, exclusions, missing fields |
| Metric definitions | Primary measures and source definitions |
| Audience evidence | Jobs completed, questions, confusion |
| Comparable cohorts | Format, lifespan, distribution controls |
| Economics | Labour, cash cost, revenue basis, uncertainty |
| Risk and quality | Rights, disclosure, safety, accessibility, corrections |
| Learning | Confirmed, directional, inconclusive |
| Decisions | Continue, change, stop, test |
| Owners | Person, due date, pass condition |
Archive the record with its raw exports and calculations. Limit access to audience and revenue data, apply the appropriate retention policy, and note deleted content because platform aggregates may still behave differently from item-level records. Never paste sensitive exports into an unapproved AI or analytics tool.
The audit is finished when the next month becomes simpler to run: fewer unsupported assumptions, one repaired bottleneck, one bounded experiment, and a documented reason for every series you continue. Ten answered questions are enough. More dashboards are not.
