Raw social API data vs normalized social data
Raw social API data is the payload a platform sends to its own developers. Normalized social media data is that history reshaped into one structure, with metrics calculated before your AI reads it. MySoMeData returns the normalized form.
What is hard about the raw payload?
- Field names and pagination differ for Instagram, TikTok, and Pinterest.
- A full raw response often does not fit an AI context window.
- Ratios such as a true engagement rate are left for you to compute.
- A trailing 28-day total and a calendar month can share a similar name and mean different windows.
What does normalization change?
You get one object shape, calculated engagement, period comparisons, format averages, and outliers. You do not get identical metrics. The normalizer keeps TikTok’s traffic source and leaves Pinterest reach null. That honesty is part of the structure. Your model should read the basis, not assume every network has reach.
Which should you put in a prompt?
Put the normalized snapshot in the prompt, or let MCP fetch it. Put a raw platform response in the prompt only when you are debugging that platform. For analysis, the calculated object is the one that leaves room for the question.