Facebook post data collection and post analysis

Propose total hours to complete the task. You are going to review, collect from posts, and complete Gemini analysis of each video post Collect from Facebook data * Views * Reach (viewers) * 3-second views * 1-minute views * Watch time * Average watch time * Revenue * Comments * Shares * Reactions * Clicks * Saves (if included) * Publish date * Publish time * Follows * Revenue These are enough to produce statistically valid analyses of: * Best posting times * Best days * Best lengths (if video duration is included) * Revenue efficiency * Engagement efficiency * Comment efficiency * Share efficiency * Composite performance score * Top performers * Worst performers So yes—we can do all of those from the exports. --- ## The CSV does **not** tell us These require watching the actual videos. For example: * Was it drone footage? * Was it handheld? * AI voice? * Human narration? * Music only? * Single continuous shot? * Heavy editing? * Minimal editing? * Historic photographs? * Number of scene changes * Hook wording * First three seconds * Thumbnail quality * Caption style * Story structure * Call to action * Production quality Meta has no column for those. Someone has to classify them. --- ## Length This is a good example. There are actually **three different "length" questions.** ### 1. Video duration If the export contains duration we can absolutely analyze: * 0–15 sec * 15–30 * 30–60 * 60–120 * 2–4 min * 4+ min Very accurately. --- ### 2. Retention The CSV also tells us Average watch time 1-minute viewers etc. So we can determine things like 90-second videos actually outperform 20-second videos. or Long videos only work when they're historical. --- ### 3. Story pacing The CSV cannot tell us whether a video changed scenes every 2 seconds or 10 seconds. That requires watching. This is part of the assignment --- ## Unedited vs Edited This is another manual classification. The export doesn't know "this video had one cut" or "this video had 75 edits." Someone has to watch each one. --- ## Here's what I think we should do I think we should split the project into **two phases**. ### Phase 1 (using the CSVs) This can be almost completely automated. We'll produce analyses such as: * Top 25 * Bottom 25 * Best length * Worst length * Best posting day * Best posting hour * Revenue per thousand viewers * Comments per thousand viewers * Shares per thousand viewers * Engagement per thousand viewers * Outlier detection * Seasonal changes * Repeatability * Regression/correlation between metrics I think we can answer **80% of your strategic questions** from the CSVs alone. --- ### Phase 2 (manual classification) This is where it becomes really interesting. We would classify perhaps the: Top 200 posts and we want to clearly capture data on the worst 200 posts and explicitly avoid those issues. by about 30 variables. Examples Topic Hook Editing Narration Drone Historic Scenery Animal History Food Music Talking head POV Scene changes CTA Opening style Production level etc. Now we can answer questions like Do minimally edited videos outperform heavily edited ones? or Does an AI voice reduce engagement? or Are wildlife videos still winning after controlling for reach? Those answers are **not** in Meta. They're in the videos. --- ## I actually think we can go one step further Knowing how systematic you are, I think this should become a **content intelligence database** rather than a one-time report. Every post would receive about 25–30 tags. As you publish new content, you add those tags, and the database continuously learns. After another six months, you could ask questions like: * "How do drone videos longer than 90 seconds about Fort Sill history perform compared to handheld wildlife videos under 20 seconds?" * "Do videos with a question in the first sentence generate more comments than those with a factual opening?" * "Does AI narration help or hurt when the topic is geology versus history?" Those are questions that the CSV alone cannot answer—but your growing library of posts can.