By MD Fozla Rabbi, SEO Specialist

Here's a question I hear from Facebook sellers in Dhaka all the time. "My page has 18,000 likes. Why do my posts get 40 reactions?"

Short answer: page likes and post likes are two very different things. Page likes are a lifetime number. Post likes are what happens today, to this one post, with this one audience.

Most small sellers only watch the first number. It's big and it only goes up. Comfortable. But the second number is where sales actually start.

So I wrote a short Python script that reads a Facebook post export and ranks every post by real engagement. It tells you which posts people actually responded to, which ones flopped and what time of day works best for your page. Standard library only. Runs on any laptop.

Page likes vs post likes, in one minute

A page like means someone clicked "Like" or "Follow" on your page at some point. Maybe three years ago. Maybe they forgot you exist.

A post like (or any reaction) means someone saw a specific post and cared enough to tap. That's a live signal.

If your page has 18,000 likes but each post reaches 900 people and gets 40 reactions, your real audience is closer to those 900. The script below helps you measure that honestly.

Step 1: export your post data

In Meta Business Suite open Insights, then Content. Pick a date range like the last 60 days and export as CSV. The columns vary a bit but you'll usually see something like:

Publish time,Post type,Title,Reach,Reactions,Comments,Shares 11/02/2026 20:15,Photo,Eid collection drop,1840,96,14,6 11/03/2026 13:05,Reel,Behind the stitching,5210,188,22,31 11/04/2026 09:40,Link,Price list update,620,9,1,0 

Some exports call it "Reactions, comments and shares" in a single column. We'll handle both.

Step 2: read the file defensively

import csv from datetime import datetime from collections import defaultdict def pick(row, *names): for n in names: for key in row: if key and key.strip().lower() == n: return row[key] return "" def num(value): value = (value or "").replace(",", "").strip() try: return int(float(value)) except ValueError: return 0 def parse_time(raw): for fmt in ("%m/%d/%Y %H:%M", "%Y-%m-%d %H:%M:%S", "%Y-%m-%dT%H:%M:%S%z"): try: return datetime.strptime(raw.strip(), fmt) except ValueError: continue return None def load_posts(path): posts = [] with open(path, newline="", encoding="utf-8-sig") as f: for row in csv.DictReader(f): reach = num(pick(row, "reach", "post reach")) if reach == 0: continue reactions = num(pick(row, "reactions", "likes and reactions")) comments = num(pick(row, "comments")) shares = num(pick(row, "shares")) combined = num(pick(row, "reactions, comments and shares")) if combined and not (reactions or comments or shares): reactions = combined posts.append({ "when": parse_time(pick(row, "publish time", "posted")), "type": pick(row, "post type", "type") or "Unknown", "title": (pick(row, "title", "description") or "")[:50], "reach": reach, "reactions": reactions, "comments": comments, "shares": shares, }) return posts 

Posts with zero reach get skipped. They're usually drafts or deleted posts that still sneak into the export.

Step 3: score each post

Raw reaction counts lie. A Reel that reached 5,000 people will always beat a photo that reached 600. That doesn't mean the photo was bad.

So we score engagement per reach. And we weight actions by effort. A share takes more intent than a like. A comment takes more than a share, usually.

WEIGHTS = {"reactions": 1, "shares": 3, "comments": 4} def score(post): weighted = sum(post[k] * w for k, w in WEIGHTS.items()) return round(weighted / post["reach"] * 100, 2) 

A score of 10 means roughly ten weighted actions per hundred people reached. For small Bangladeshi pages I've worked with, anything above 8 is strong. Below 2 means the post basically got ignored.

These weights are a judgment call. Change them if your business runs on comments, like "price?" and "inbox please", which is very common here.

Step 4: build the report

def report(path, top=5): posts = load_posts(path) for p in posts: p["score"] = score(p) ranked = sorted(posts, key=lambda p: p["score"], reverse=True) print("Best posts by engagement per reach") for p in ranked[:top]: print(f" {p['score']:>6} {p['type']:<6} reach {p['reach']:>5} {p['title']}") print("\nWeakest posts") for p in ranked[-top:]: print(f" {p['score']:>6} {p['type']:<6} reach {p['reach']:>5} {p['title']}") by_type = defaultdict(list) for p in posts: by_type[p["type"]].append(p["score"]) print("\nAverage score by post type") for t, scores in sorted(by_type.items(), key=lambda x: -sum(x[1]) / len(x[1])): print(f" {t:<8} {sum(scores) / len(scores):.2f} ({len(scores)} posts)") by_hour = defaultdict(list) for p in posts: if p["when"]: by_hour[p["when"].hour].append(p["score"]) if by_hour: best = max(by_hour, key=lambda h: sum(by_hour[h]) / len(by_hour[h])) print(f"\nBest posting hour: {best:02d}:00 to {best + 1:02d}:00") if __name__ == "__main__": import sys report(sys.argv[1]) 

Run it:

python fb_engagement.py posts_export.csv 

Sample output from a clothing page in Mirpur:

Best posts by engagement per reach 11.36 Photo reach 1320 Customer review screenshot 9.24 Photo reach 1840 Eid collection drop 7.08 Reel reach 5210 Behind the stitching Weakest posts 2.10 Link reach 620 Price list update 1.20 Link reach 410 Website new arrivals Average score by post type Photo 7.95 (21 posts) Reel 6.80 (9 posts) Link 1.44 (6 posts) Best posting hour: 20:00 to 21:00 

What this told one seller

That clothing page had nearly 20,000 page likes. She thought her audience was huge. The report showed something else.

Link posts were dead. Posts that pushed people off Facebook to her website averaged a score under 1.5. Facebook doesn't love sending traffic away. Her audience didn't click either.

Real photos beat polished graphics. A blurry customer review screenshot outscored her designed banners. People trust other buyers more than ads.

After Maghrib and after dinner were the sweet spots. Her best hours were 8 PM to 10 PM. She'd been posting at 11 AM because that's when she had free time. A scheduled post fixed that in a day.

She also realised her page likes figure wasn't telling her much. A lot of those likes came from an old giveaway years back. That's why post reactions felt low compared to the page count. If you're weighing whether bought or campaign-driven likes help at all, this breakdown of page likes vs post likes for Bangladeshi Facebook pages explains the difference, real costs and the risks pretty clearly.

Small upgrades worth adding

  • Split Reels from photos in the ranking. Reels reach new people. Photos usually reach existing followers. Ranking them separately is fairer.
  • Track a rolling 30-day average score. If it slides for three weeks straight, your content needs a refresh.
  • Add Bangla titles safely. The utf-8-sig encoding already handles Bangla text. Just make sure your terminal font supports it.
  • Export weekly. Save each week's report into a folder so you can compare months later.

Wrapping up

This script won't grow a page by itself. But it replaces a vanity number with an honest one. And once a seller sees that her real audience is the 900 people who react, not the 18,000 on the counter, her decisions change fast.

I've used versions of this with a few local sellers. If you're building tools for the Bangladeshi social commerce scene, the team at SMM Booster BD publishes a fair amount about how Facebook and TikTok audiences behave here.

Try it on your own export. If a column name doesn't match, paste your CSV header in the comments and I'll add it to the pick() calls.