Running a Social Media Content Audit That Avoids Causal Claims

A social media content audit delivers three dated artefacts: a content inventory, a mutually exclusive taxonomy, and an outcome view grouped by comparable cohorts. It describes what content existed and how it performed within a specific distribution context. It does not explain why something worked. If you need causal claims, you need an experiment with a control group. That is a different project.

Build The Inventory First

Export every post from Meta Business Suite Insights, TikTok Analytics, and X Analytics at the finest granularity each service allows. Post-level export is standard across all three. Impression-level export requires API access on most services, and the UI will not give it to you.

Apply The Taxonomy Second

Segment every post by format, service, and paid/organic/earned distribution before comparing performance. A 15-second TikTok video and a 30-minute YouTube upload are not the same thing, and a boosted Instagram reel is not an organic one.

Build The Outcome View Third

Group posts into cohorts that share format, service, and distribution channel. Then compare interaction rate by impressions, not by reach. Reach-based rates are always higher and are the default on native services. Impression-based rates are the only cross-format comparability you can defend.

A social content inventory is a dated export of every post in scope. The date matters because service metric definitions change without notice, and a retroactive recalculation can silently alter numbers you already reported. Meta changed its video-view threshold, X changed its interaction definitions, and TikTok revised its reach methodology. Your inventory is a snapshot, not a database.

Start with Meta Business Suite Insights for Facebook and Instagram. Export at post-level granularity for the exact date range you are auditing. Do not rely on the dashboard numbers alone. API data completeness differs from UI data completeness, with a typical discrepancy of 5 to 15 percent on post-level metrics. The service's own analytics are the source of truth, but the export is what you annotate and share.

TikTok Analytics exposes watch-time metrics that other services obscure. Use those, not just views. A 3-second view on Meta and an immediate view on TikTok are not comparable, and neither is a YouTube view, which has a different threshold. Video view threshold is a definition you must check against current service documentation before you merge any numbers.

Data freshness varies. Dashboards labelled real-time have 5 to 30 minutes of latency for standard metrics, and demographic breakdowns can lag by up to 24 hours. Do not run a post-mortem on yesterday's posts and expect the numbers to be final.

A content taxonomy framework forces you to label every post before you analyse it. This is the step most teams skip because it is tedious. It is also the step that makes the rest of the audit defensible.

Your taxonomy has three axes. Format is the first: text, image, video, story, live. Service is the second: Facebook, Instagram, TikTok, X, LinkedIn, YouTube. Distribution channel is the third: paid, organic, or earned. A boosted post is paid. A post shared by an employee to their own feed is earned. Your own account posting without spend is organic.

Make the categories mutually exclusive. A video post that you boosted is one row with one format, one service, one distribution channel. If it was also shared by an influencer, that is a separate earned row. The goal is that every post falls into exactly one cell of the taxonomy.

Content-type coverage is a metric you must state before comparing across tools or services. Not every tool ingests every format. Stories are ingested as metadata only with no content-level analysis. If your audit includes Stories, say what you could and could not measure.

Social media post analysis is where the failure mode lives. You will see a 12-second TikTok with an interaction rate of 9 percent and a 40-second video with 2 percent. The instinct is to say short videos win. That is a causal claim you cannot make from this data.

Ask what else differed. Was the 12-second post boosted with paid media spend? Was it published at a time when your audience composition skewed toward heavy scrollers? Was the 40-second post uploaded during an algorithm change? If you cannot rule those out, you cannot claim format caused the difference.

The correct statement is narrower. In this cohort, on this service, during this date range, posts of this format with this distribution channel recorded this interaction rate by impressions. That is a description of a pattern, not an explanation of it.

Do not treat high-interaction posts as a replicable formula. The dashboard shows correlation, not causation. The spike in interactions and the spike in sales on the same screen are two unrelated time series until you run an experiment. Service attribution models credit social for conversions where social was any touch in the path. Randomised controlled experiments show social's incremental contribution is 30 to 60 percent of the service-claimed figure.

A content performance cohort is a group of posts that share the same format, service, and distribution channel. You compare interaction rate only within a cohort, never across them. A LinkedIn text post and a TikTok video are not comparable, even if both are organic.

Reach definition is the reason. Meta counts reach after a threshold of pixels loaded. X counts any impression served. If the same person sees a post three times, that is one reach and three impressions on every service, but the services do not agree on what counts as an impression or a view. Cross-service reach is incomparable. Do not sum it.

Interaction rate by reach is the default on native services and is always higher. Interaction rate by impressions is what third-party tools use for cross-service comparability. Use the impression-based rate inside a cohort. Use the reach-based rate only when you are reporting within a single service's native dashboard.

Sampling threshold is another reason to stay within a cohort. Above a certain data volume, services switch from census to sample, and the threshold is dynamic and undocumented. A 50-post cohort may be complete. A large cohort may be sampled. You cannot tell from the dashboard.

UTM parameters are the only cross-service attribution mechanism not owned by any single vendor. The five standard parameters are source, medium, campaign, term, and content. Use them on every outbound link from every social post, or your social-referred traffic will appear as direct traffic in your web analytics.

UTM parameter decay is a real failure. Campaign URLs get shared without parameters, or privacy tools strip them. When that happens, the click disappears from your social attribution. The fix is to append parameters at the point of posting, not after, and to test one link before you schedule fifty posts.

GA4 attributes a conversion to the last non-direct click. Service analytics attributes it to any exposure within the attribution window. These will never match. The gap is the measure of multi-touch influence, not an error to reconcile.

The separation of organic and paid social data pipelines is architectural. Organic post analytics and paid ad analytics run on different service infrastructure with different data definitions. Do not merge them into one spreadsheet and call it a unified view.

Sampling threshold is the data volume above which a service switches from census to sample. The threshold is dynamic and undocumented. The sampling methodology is not disclosed. When you run cohort analysis on a large dataset, you are almost certainly working with a sample, and you cannot know how it was drawn.

This matters most when a dominant theme appears in one refresh and disappears in the next. Sampling-induced insight reversal is caused by non-representative sampling in time or population. A fast-moving topic on X is especially vulnerable. Refresh the data before you write the report. If the theme does not survive the refresh, it was noise.

Cohort analysis is the antidote to false confidence. Compare the median interaction rate by impressions within a cohort, not the mean. A single viral post will drag the mean upward and make every other post look like an underperformer. The median tells you what a typical post did.

Data freshness compounds the problem. Real-time dashboards display incomplete data that is silently backfilled hours later. A cohort built at 9am will not match the same cohort built at 5pm. Build your inventory once, from an export, and analyse that snapshot.

Content-type coverage is the number of content formats a tool or service can ingest and analyse. Meta Business Suite Insights covers text, image, video, stories, and live for Facebook and Instagram. TikTok Analytics covers video and live. X Analytics covers text, image, and video. Ephemeral formats like Stories are ingested as metadata only, with no content-level analysis.

Export granularity is the finest row-level unit available for download. Post-level is standard. Impression-level requires API access on most services and is rate-limited. Do not plan an audit that needs impression-level data unless you have confirmed API access and budgeted for the rate limits.

API rate-limiting is a structural constraint. Every service imposes limits on data access volume and velocity. The documented limit and the enforced limit diverge under load, and some endpoints silently throttle before the stated ceiling. Pulling a year of post-level data can take hours, not minutes.

When numbers conflict between a third-party tool and a service's native dashboard, the service's numbers are the source of truth for billing and reporting. This hierarchy is structural. Your audit should state that you used native exports wherever possible.

The single most common failure in a social media content audit is treating a high-interaction post as a formula. You see a 7-second video with a 12 percent interaction rate by impressions. You declare that 7-second videos are the answer. You make five more. They all underperform.

What you missed: the original post had paid media spend behind it. Or it was posted during a window when your audience composition shifted toward a demographic that interacts more. Or the algorithm gave it a temporary distribution boost. None of these are visible in a post-level export with no cohort controls.

A social media content audit describes what content existed and how it performed within a specific distribution context. It does not explain why. If you need causal claims, you need an experiment with a control group, not an audit. That is a different project.

The honest caveat: you are measuring traces left on proprietary services whose data definitions change without notice. Your inventory is a snapshot. Your taxonomy is a human judgement. Your cohorts are an approximation. Treat them as such, and your audit will survive contact with a sceptical stakeholder.