AI Social Media Growth
0/15 complete

Module 02 · Trend Analysis

Spotting Trends Early Using AI Summarization Tools

Trend spotting finds patterns repeating now, before you commit production time. It is not AI prediction. Evidence comes from native platform surfaces and dated observations; AI compresses it.

By the end, you will have a filled five-day trend log and one trend selected for a small content test. The log will show what you saw, where and when you saw it, whether it is a reusable format or a short-lived moment, and why you adopted or rejected it.

// concept

Run the 15-Minute Native-Surface Scan

Scan one primary platform and use one second surface only to validate a signal.

  1. Minutes 0–2 — set the lane. Record date, time, platform, niche, language, and one question: “Which openings repeat in Pakistani home-baking Reels?”
  2. Minutes 2–7 — inspect a native surface. TikTok: Creative Center > Trends, filter the available industry/time frame, then open See Analytics. Instagram: inspect niche Reels and open the audio page when a sound recurs; its official Trending audio dashboard is limited by account, app, and country, so check current availability. YouTube: Studio > Analytics > Trends, if shown; some insights vary by country, language, and device. LinkedIn: search a niche term, choose Posts, then Date posted and Content type filters.
  3. Minutes 7–10 — capture observations. Save three to five links or post IDs. Note visible post date, first frame/line, format, audio or visual cue, angle, language, and differences. Do not copy full captions or scripts.
  4. Minutes 10–13 — cluster with AI. Paste observations, not “this is trending.” Require source IDs, evidence/guess separation, and a small-sample warning.
  5. Minutes 13–15 — decide. Mark each candidate format, moment, or unclear, then adopt, watch, or reject, with a reason and review-by time.

Confirm current paths in the official pages below. A “trending” label is an input, not proof of audience fit.

// concept

Summarize Without Letting AI Invent a Trend

Use compact notes. Unless browsing is explicitly available, a model cannot inspect a linked post.

// prompt — copy me19 lines
You are organizing a dated social-media trend scan. Use only the observations
inside <observations>. Do not claim that anything is viral, growing, or likely
to perform. Do not infer views, locations, demographics, or motives.

Task:
1. Cluster cues that appear in at least two source IDs.
2. Label each cluster FORMAT, MOMENT, or UNCLEAR.
3. For each cluster, quote only my short cue, list its source IDs, state what
   repeats, state what changes, and name the missing evidence.
4. Score niche fit and effort Low/Medium/High with one reason. These are
   planning judgments, not performance predictions.
5. Recommend ADOPT, WATCH, or REJECT. ADOPT requires an original niche angle
   and a review-by date/time.

Return a markdown table followed by no more than three verification questions.

<observations>
[Paste 3–5 dated observations. Keep labels S1, S2, S3...]
</observations>

Reject a cluster that drops source IDs, joins unrelated cues, or turns “seen twice” into “audiences want this.” Summarization does not strengthen evidence.

// worked_example

Worked Example

This is a hypothetical sample for a Karachi home baker. Its labels and observations are invented for practice, not real-account claims.

// prompt — copy me8 lines
Scan: 18 July 2026, 10:10 PKT | Instagram Reels | home-baking tutorials
S1, today: overhead hands; collapsed sponge first; “Why this cake sank”;
three cause cards; corrected slice.
S2, yesterday: face-to-camera question; story; recipe at end.
S3, today: overhead hands; failed buttercream first; “3 reasons it split”;
each cause gets a close-up fix.
S4, two days ago: overhead hands; burnt cookie first; “Stop doing this”;
one cause, corrected tray, temperature reminder.

The prompt produced this realistic sample excerpt:

FORMAT — failure-first diagnostic tutorial. Evidence: S1, S3, S4. Repeats: visible failed result before explanation, overhead demonstration, cause-to-fix progression. Changes: cake, buttercream, cookies; number of causes. Missing evidence: whether the format is repeating beyond this small feed sample. Planning judgment: niche fit High, effort Medium. Decision: WATCH pending a second-day scan.

Draft one wrongly included S2 merely because it was a tutorial. The fix was: Do not group by topic alone. Require two matching structural cues. Re-evaluate S2. The revision excluded S2 and kept the evidence warning.

Here is the resulting sample five-day trend log:

Day/dateNative surface and evidenceCandidateTypeFreshness checkDecision
1 — 14 JulInstagram Reels, S1–S4Failure-first tutorialFormatThree matches; small sampleWatch
2 — 15 JulTikTok Trends, T1–T3“Stop X; do Y” openingFormatPoor baking fitReject
3 — 16 JulLinkedIn recent Posts, L1–L3Annotated before/afterFormatWrong asset for ReelsReject here
4 — 17 JulInstagram Reels, R1–R3Rain-day delivery jokeMomentCurrent weather; no useful lessonReject
5 — 18 JulInstagram Reels, V1–V3; Day 1 recheckFailure-first tutorialFormatRepeated; original topicAdopt: “Why buttercream looks grainy”

The adopted plan keeps the structure, not anyone's wording: show the failed texture, explain two causes, demonstrate a correction, and state its limits. This is a production choice, not a performance prediction.

// failure_cases

Failure Cases to Diagnose

6 cases to diagnose

  • AI declares a trend from one post.

    The cluster has only one source ID. Change the status to unclear and collect another independent observation.

  • Topic and format are confused.

    Three posts mention Eid cakes, but their structures differ. Log Eid as a moment/topic; do not claim a repeatable format.

  • The summary erases dates.

    Without observed and posted dates, you cannot judge decay. Put timestamps back into the input and output.

  • A global signal is labelled Pakistani.

    The selected country or audience is missing. Record the exact region available and validate in your own Pakistan-facing feed.

  • The adaptation copies expression.

    Your plan retains a creator's sentences, joke, shot sequence, and ending. Keep only the broad public format and rebuild the substance.

  • The scan never decides.

    Require an adopt/watch/reject status, reason, and review-by time.

// pakistan_angle

Pakistan Angle

Never label another country's chart Pakistani. TikTok and Instagram features differ by region/account, and YouTube insights may vary by country/language. Record the exact surface and region, then validate a global format in a Pakistan-facing feed. Test English, Urdu, and Roman Urdu separately: “three mistakes” and “yeh teen ghaltiyan” may carry different tones.

PSL, Ramzan, Eid, monsoon, or load-shedding moments decay quickly and can carry religious, political, or safety sensitivities. Verify the news and contribute only when relevant; a Karachi boutique need not force every match or weather event into a sale.

On limited data, save links and notes rather than videos; draft offline during outages, then reopen shortlisted sources. Never paste WhatsApp screenshots, phone numbers, addresses, CNIC details, or private DMs into AI. Use an authorized paraphrase.

// hands_on

Hands-On Exercise

5 steps

Build your own five-day trend log with one adopted trend.

  1. Create columns for date/time, platform, surface/region, source IDs, cue, type, missing evidence, freshness, niche fit, effort, decision, and review-by time.

  2. Run the 15-minute scan for five days; capture three to five dated observations daily.

  3. Use the summarization prompt. Check every cluster against its source IDs.

  4. Mark each candidate adopt, watch, or reject; keep one rejection and its reason.

  5. For one adoption, write a three-line plan that keeps the format but replaces all substance. You are done when the five rows are dated, every decision traces to observations, and one adopted trend has an original production plan plus a review-by time.

// completion_rubric

Completion Rubric

5 checks — tick as you verify

0/5

// sources

Sources

// check_yourself

Check yourself

4 questions · answers and options are taken word-for-word from this course

0/4
  1. 1 / 4 · diagnose

    Your work shows this failure mode: “AI declares a trend from one post.” The lesson describes it like this: “The cluster has only one source ID.” What does the lesson tell you to do about it?