Which Social Media Metrics Actually Matter.

Most social media reports are full of numbers and short on decisions.

The metrics that matter are the ones that change what you do next, and the ones a platform cannot quietly rewrite overnight. That is a short list: retention and completion on video, saves and shares, comments and messages with substance in them, the quality of the audience you are attracting, the traffic and enquiries that arrive from social, and the off-platform signals that show the brand is getting easier to find.

Follower count is not on that list. In May 2026, Instagram made the argument better than any agency could.

In May 2026, Instagram proved the point

In the first week of May 2026, Instagram follower counts fell across the platform. Large accounts lost millions. Small accounts lost a few hundred.

Meta’s explanation was short. “As part of our routine process to remove inactive accounts, some Instagram accounts may have noticed updates to their follower counts,” a company representative said, adding that “active followers remain unaffected” and that any restored suspended account would be included in the count again after verification.

Routine. That is the word worth sitting with.

Inc reported Instagram’s own account down 15.2 million followers to 686 million, with the largest celebrity and brand accounts on the platform losing somewhere between four and ten million each.

Treat those figures as reported rather than confirmed. Meta published no numbers of its own, and the per-account totals quoted across outlets that week do not agree with each other.

Inactive accounts, bots and a recalculation are three different things

Coverage that week folded three separate events into one story.

• An inactive account is a real person who signed up and stopped opening the app. Removing them changes your number without changing your audience, because they were never going to see a post.
• A bot or spam account was never a person at all. Removing them corrects a number that was inflated from the day it was counted.
• A recalculation removes nothing. The platform changes how it counts and the figure moves on its own.

Meta’s own wording described the first. Most headlines described the second. The line about restored suspended accounts returning after verification points at a third thing again, because some of the drop was reversible.

Three different causes, one identical symptom on your dashboard. If you cannot tell which one moved your number, that number was never measuring your work.

A number the platform can revise overnight is not a target

Nothing about the real audience changed in May. The people who were going to watch, save and share still were. Only the number moved.

That is the test worth applying to every metric on your scorecard. If someone else’s product team can redefine it, delete part of it or rename it, it is a reading, not an asset.

Follower count also fails the second test, which is what happens when a number becomes the goal.

Chase reach and you make broad content that builds no trust. Chase engagement and you write for reaction rather than relevance. Chase followers and you can grow quickly while attracting exactly the wrong people.

The answer is not to ignore the number. It is to demote it, and to be honest about what a realistic goal actually looks like before the reporting period starts rather than after it.

Start with the job of the post

Before you pick a metric, decide what the content was for. Different posts do different jobs and should not share a scorecard.

Common jobs:

• Building awareness with people who have never heard of you
• Earning attention in a feed that offers no second chances
• Building trust with people already considering you
• Explaining the offer
• Driving a click, an enquiry or a booking
• Testing a topic, a hook or a format

A hook test judged on clicks will look like a failure. A proof post judged on reach will look like one too.

Measurement without a content job is just scorekeeping.

The metrics that survive

These are the signals worth putting in front of a leadership team, and what each one is evidence of.

• Retention and completion: the content held attention after the hook, not just at it
• Saves: someone expects to need this again
• Shares and sends: someone was willing to put their own name behind it
• Comments with a full sentence in them: relevance rather than reflex
• Direct messages and enquiries: intent
• Returning viewers: you are building an audience rather than renting one
• Profile visits and link taps: curiosity turning into a step
• Follows attributable to a specific post: content that earned an audience

None of these means much on its own, and none has a universal good number. Rates vary so widely by category, format and account size that a borrowed benchmark is worse than no benchmark.

Your baseline is your own last 90 days.

Retention is the honest measure of a video

Views count starts. Retention counts whether anyone stayed, which is the only part of the transaction that reflects the work.

The shape of the drop-off tells you what to change.

• A fall in the first three seconds usually means the opening frame or first line did not match what the feed promised
• A slide in the middle means the pacing sagged or the payoff sits too far back
• Strong completion with low views is a packaging problem, not a content problem

Rewatches are the strongest version of the same signal. Someone chose to see it twice.

This holds wherever short-form video lives. Reels, TikTok, Shorts and Spotlight all reward the same thing, which is a video that earns its own length.

A share is worth more than a like

A like costs nothing. Most are a scroll reflex and mean roughly that much.

A share is a small reputational act. Someone decided your content was worth attaching to their own name in front of people they know.

A save is a private admission that the content will be useful later, which is the clearest signal you get that educational content is landing.

So never report engagement as one number. A post with fewer likes and many saves may be doing more commercial work than a post with the opposite profile, and a single engagement rate hides that completely.

Audience quality is not follower count

Growth is about the right people, not more people. A brand can add followers quickly and become less commercially relevant in the process.

The questions worth answering monthly:

• Where are these people, and does that match where you sell?
• What do they do, and are decision-makers among them?
• Do they come back, or is every month a fresh set of strangers?
• Are the comments coming from people who could realistically buy?

For a Dubai or wider MENA brand, location is often the whole question. For B2B, seniority and sector matter more than volume. For a founder-led account, who is in the replies is the metric.

Working out which of your current numbers are load-bearing is the first hour of any serious audit, and the social media audit checklist walks the same route.

Judge each platform against its job, not one scorecard

Instagram, TikTok, LinkedIn and YouTube do different work, so they need different scorecards.

• Instagram: reach by format, Reel retention, saves, shares, profile visits and follows per post
• TikTok: average watch time, completion, rewatches, shares, and what you learned about hooks
• LinkedIn: who commented rather than how many, profile and company page views, inbound messages, engagement from target sectors
• YouTube Shorts: watched versus swiped away, average view duration, subscribers gained, and traffic through to longer videos

One warning about totals. Adding a Facebook view to a LinkedIn impression produces a number with no unit and no meaning. Report platforms side by side, never summed.

The metric names changed too, so this year is not last year

Meta introduced Views on 14 November 2024 as a single distribution metric across Facebook and Instagram. For photos, text and Stories, impressions became views, and repeats now count. Meta’s own example is a photo viewed three times in one day counting as three views rather than one impression. For Reels, plays became views with no change to the calculation. Watch time became watch minutes.

On the Instagram API the change carries a date. Impressions and plays were deprecated from 21 January 2025 for v22.0, and for all versions from 21 April 2025, with views introduced in their place.

Reach survived both passes and still means the number of unique accounts that saw the content at least once.

YouTube moved the same way. From 31 March 2025 a Shorts view counts every time a Short starts or replays, with no minimum watch time, and the older, stricter metric is now called Engaged views.

The practical consequence is simple. A chart that runs impressions into views, or Shorts views across March 2025, has a step in it that has nothing to do with your content. Label the change on the chart, or start a new series.

The measurement no analytics screen gives you

The most useful evidence that social is working sits outside every platform dashboard.

• Branded search volume over time, meaning people typing your name rather than your category
• Direct traffic and returning visitors to the site
• The “how did you hear about us” field on your enquiry form, asked as an open question
• What prospects mention unprompted on sales calls
• What candidates mention in interviews

These are slow, imprecise and the most honest signals you have. A branded search line rising across two quarters alongside consistent publishing is not proof, but it is a pattern, and it is worth more than one strong month of reach.

Ask the question properly and it costs nothing. Most brands never ask it at all.

What belongs in a monthly report to leadership

A leadership report is not an archive. It is an argument about what to do next.

Include:

• What you set out to do, and whether it happened
• Three to five numbers, each attached to a decision
• The best and worst piece of the month, with a reason for both
• What you learned, and what changes next month
• One system problem: the approval, the brief or the capture that cost you something
• The off-platform read, once a quarter

Leave out:

• Follower count as a headline figure
• Screenshots of every platform dashboard
• Percentages with no base number behind them
• Any metric nobody is going to act on
• Combined totals that add unlike numbers together

How to change what you report without looking like you are hiding a fall

This is the real problem for a marketing lead, and almost nobody writes about it. You have reported follower growth for three years. You now know it is close to worthless, and the month you stop reporting it is the month it looks like you are covering something up.

The way through is sequence, not spin.

1. Change at a boundary. A quarter or a financial year, never mid-cycle.
2. Run both scorecards in parallel for one cycle, old and new side by side, so nothing disappears without being seen.
3. Rebuild the last two or three periods on the new metrics before you present them, so the first report shows a trend rather than a fresh start.
4. Write one line per new metric saying what it is evidence of. Nobody defends a number they cannot explain.
5. Name the platform changes that forced part of the shift. Impressions being retired is not your opinion, it is a dated fact you can point at.
6. Keep the old number visible but demoted. One line at the back beats a deletion, because a deletion invites the question.
7. Agree the new scorecard before the first report that uses it, not inside it.

Do it in that order and the conversation is about measurement. Do it in one jump and the conversation is about you.

What good social media measurement looks like

Good measurement makes social calmer, not busier.

The team knows what each post is for before it goes out. The report is short, and every number in it is attached to a decision. Stakeholders understand why something worked, not just that it did.

The content calendar visibly changes because of what last month taught you.

And when a platform renames a metric or removes a few million accounts overnight, nobody panics, because the scorecard was never resting on a number the platform controls.

How NBK thinks about social media measurement

NBK treats measurement as part of the operating system, not the paperwork at the end of it.

Reporting that never feeds back into content pillars, platform roles, workflow and production is admin. Reporting that does is the fastest way to make next month better than this one, and it belongs inside a social media system that actually runs rather than beside it.

NBK’s team came up through publishers including UNILAD, LADbible Group and SPORF, and now works at a scale of 600M+ views a month. At that volume, the gap between a number that predicts what happens next and a number that only describes what already happened becomes very hard to miss.

For a lot of brands the problem is not that the numbers are bad. It is that nobody has decided what any of them are for.

Next step

Start by cutting the scorecard down. Decide what each platform is for, decide what each content type is meant to do, then keep only the metrics that show whether the job was done.

Then make every report answer one question. What should change next?

If your team is tracking social performance and still feels unclear on what is working, NBK can help build a measurement rhythm that connects reporting to better decisions.