Engagement rate formula: which denominator to use and why two tools disagree

How do you calculate engagement rate?

Add up the interactions on a post, divide by the audience number you picked, and multiply by 100. Our example post has 420 likes, 35 comments, 20 shares and 25 saves, which is 500 engagements. Only the audience number changes between the three rows below.

1. Add the interactions you can see: likes + comments + shares + saves. Leave a metric out if the platform hides it, and write down that you did.
2. Pick one audience number: followers, reach (unique people who saw the post), views (every play or display) or impressions.
3. Divide the engagements by that number and multiply by 100.
4. Write the result with its denominator, for example "5.00% by followers".

Our arithmetic, run by script (page-numbers.py in the test files): 500 / 10,000 = 5.00%, 500 / 8,000 = 6.25% and 500 / 4,000 = 12.50%. Instagram defines Viewers as unique accounts and Views as every play or display, so views are at least the number of viewers and the rate by views is never higher than the rate by reach for the same post. The audience values are an example we chose, not data from a real account.

Free tool:Engagement rate calculator (free, in your browser)

Which audience number should you divide by?

Choose by the question you are asking. This table is our suggestion, based on what each denominator measures.

Pick one denominator for a report and keep it for every post and every month. Changing it midway breaks the comparison, because the same post moves from 5.00% to 12.50% when the denominator moves from followers to reach.

What counts as an engagement on each platform?

The platforms define their own counts, and none of the help pages we could read gives an engagement-rate formula. This is what we found on October 28, 2026.

Instagram's help page marks Viewers and Accounts engaged as "estimated and in-development". The Socialinsider formulas show the problem in one place: its Instagram formula counts likes and comments only, while its LinkedIn formula adds shares and clicks. The two official lists differ too: Instagram counts likes, comments, saves and shares, while YouTube names views, likes, dislikes and subscriptions. Two tools can both be correct by their own definitions and still print different numbers for the same post.

What does one post look like under each platform's own counting?

Here is one invented post per platform, counted the way each source describes. Every count below is made up to show the arithmetic; none is real data. Where the source gives no formula (YouTube and TikTok), the choice of what to count is ours and labelled.

Source: our arithmetic, run by script (worked-examples.py; the output is in worked-examples-results.txt, and its control, 50 engagements over 1,000, printed 5.00%). The Instagram numerator is the interactions list from Instagram Help (likes, comments, saves and shares) and the audience numbers are the ones Instagram Help defines (Viewers, Views); the Socialinsider rows use its formulas as quoted above. YouTube Help lists views, likes, dislikes and subscriptions as the metrics it counts but gives no formula, and Socialinsider's TikTok formula says "Post Engagement" without listing the actions, so the numerators in those two rows are our choice.

The same invented Reel is 17.39%, 10.39% or 9.95% depending on the audience number and on whether saves and shares are counted. Do not compare a rate in one row with a rate in another: each answers a different question, and the guide above says to write the denominator next to every rate.

Where do you find these numbers in the apps?

Both platforms describe the path on their own help pages, read October 28, 2026.

• Instagram: tap View insights under a post or reel. Instagram Help lists Views, Viewers, Interactions and Accounts engaged for that content, for public accounts, with a timeframe you can pick within the past 90 days for the overview.
• YouTube: in YouTube Studio open Content, choose the video, select Analytics, then Engagement in the top menu. YouTube Help names four reports there: Audience retention, Likes (vs. dislikes), End screen element click rate and Top remixed.
• TikTok: we could not read TikTok's analytics help pages, so we cannot give the path.
• Our reading of the Instagram definitions: Interactions is the numerator, Viewers (unique accounts) is the closest audience number to reach, and Views is the audience number for the rate by views.

Why does the same post get a different rate in two tools?

Because the numerator or the denominator differs. We can show the numerator effect with our own post: by followers, likes alone give 4.20%, likes plus comments give 4.55%, and all four interactions give 5.00%. That is a 0.80 point spread from the choice of what to count, before any change of denominator.

Instagram adds a second source of difference. It reports Interactions (actions) and Accounts engaged (unique accounts) as separate numbers. In a made-up example, 120 interactions from 80 accounts on 2,000 viewers is 6.00% if you divide interactions by viewers and 4.00% if you divide accounts engaged by viewers. Both use Instagram's own metrics.

• Ask which interactions are counted: likes only, likes and comments, or all four.
• Ask which audience number is the divisor: followers, reach, views or impressions.
• Ask whether the tool counts actions or unique accounts.
• If the answer is not written down, you cannot compare the two numbers.

How do you average engagement rate across many posts?

Add all the engagements, add all the audience numbers, and divide once. Do not average the per-post percentages. In our example, post A has 100 engagements on 1,000 views (10.00%) and post B has 900 engagements on 100,000 views (0.90%). The mean of the two rates is 5.45%. The pooled rate is 1,000 engagements divided by 101,000 views, which is 0.99%.

The pooled rate is the better summary of what happened across the audience, because post B reached 100 times more viewers. Some tools average per-post rates on purpose, to give small posts the same weight as large ones. If you use one of those, note it next to the number. This is our suggestion; neither method is an official standard.

Does the Koldflux calculator give the right numbers?

Yes on 9 of the 10 cases we ran, and the tenth is off by 0.01 percentage points. We drove the live engagement rate calculator with Playwright on October 28, 2026 (Chromium, the public page) and compared every result with exact fraction arithmetic rounded half up to two decimals. The first two cases are controls whose answers we knew before running them: 500 engagements on 10,000 followers must read 5.00% and 0 engagements must read 0.00%. Both passed.

Three things to know from this test. First, the tool shows a dash instead of a rate when the audience number is 0, which is the right behavior for a division by zero. Second, a rate can pass 100%: 900 engagements on 800 views is 112.50%, because one viewer can like, comment and share. Third, the one difference is a rounding case: 201 / 20,000 x 100 is exactly 1.005, which rounds half up to 1.01%. The tool printed 1.00%. In JavaScript, 201 / 20000 * 100 gives 1.005 but its toFixed(2) prints 1.00 (the log has the line), which is the likely cause; we did not read the tool's source. The effect was at most one hundredth of a percentage point and only on exact ties. We fixed it on October 28, 2026: the calculator now rounds whole numbers half up in integer arithmetic, a regression test checks more than 60,000 number pairs against exact fractions, and the same 10 cases run again on the fixed build matched 10 of 10 (test-results-after-fix.txt).

We ran the same 10 cases through a second free calculator that takes typed-in numbers, the Elfsight engagement rate calculator, in its followers mode and its reach mode (20 runs, Chromium, October 28, 2026). It matched exact arithmetic on 16 runs. On the other 4 (zero engagements, and a zero audience number in each mode) it showed a message such as "Enter the number of followers." instead of a rate. On the 1.005% case it showed 1.01%, the exact half-up result, where the Koldflux calculator showed 1.00% before our fix. Both tools count likes, comments, shares and saves, so on ordinary numbers they agree.

Scope: 10 cases, one browser, one day, two calculators. We did not test others, and the sample is too small to rank them. The tests and their output are in the files linked under the sources.

What does engagement rate look like on real accounts?

We read public counts from three organization accounts on Mastodon on October 28, 2026, because Mastodon shows favourites, boosts and replies on every post and the follower count of the account, with no login. It shows no reach, views or saves counts to outsiders, so the only public audience number is followers. These are Mastodon counts, not Instagram, TikTok or YouTube counts, and they show how the arithmetic behaves rather than what a good rate is.

For each account we took its 20 newest original posts (no replies, no boosts) that were at least 7 days old and divided the total by the follower count at the time we read it.

Three things show in the log (real-data-results.txt). First, what you count matters: favourites, boosts and replies together came to 1.63, 1.56 and 1.45 times the favourites alone, which is the same numerator effect as the 4.20% to 5.00% in our made-up post. Second, in this sample the larger account had the lower rate: 0.05% at 879,119 followers, against 0.44% and 0.29% at 34,347 and 25,516. Three accounts show a pattern in these accounts, not a rule for all accounts. Third, the posts of one account differ a lot: the 20 Wikipedia posts ran from 0.08% to 1.23% each.

The mean of the 20 per-post rates equalled the pooled rate for every account, because every post has the same denominator, the follower count. The averaging trap above appears only when the audience differs by post, as it does with views or reach.

Limits of this table: the follower count is the one on October 28, 2026, while the posts are up to 16 months older (the Wikipedia posts start on 2025-06-14), so the rates are approximate; counts change after you read them; and three accounts are not a benchmark.

What is a good engagement rate?

There is no official benchmark. Instagram and YouTube Help do not publish one, and we found none from TikTok. Third-party benchmark reports exist, but each defines engagement and audience its own way (see the table above), so a number from one report does not fit another tool's result. We do not quote one.

What the arithmetic does show is that the same engagements make a smaller rate on a bigger audience: 500 engagements are 50.00% of 1,000 followers, 5.00% of 10,000 and 0.50% of 100,000. Whether real accounts behave like that depends on the platform and the account, and our three Mastodon accounts are too few to say. Compare an account with accounts of a similar size on the same platform, over the same period, with the same kinds of posts, counted the same way.

A worked baseline with made-up counts: ten posts, nine of which reached 1,500 to 2,600 people and drew 79 to 140 engagements, and one (post 5) that went viral, reaching 21,000 people with 520 engagements. Post 5 has 35.1% of all engagements but 53.3% of all reach, so it pulls the pooled rate down to 3.76%. The mean of the ten per-post rates is 4.94% and the median is 5.21%. Without post 5 the pooled rate is 5.22% and the median is 5.24%, so the median barely moved while the pooled rate changed by 1.46 points.

Build your own baseline in four steps. Take your last 10 to 30 posts. Add their engagements and their audience numbers (reach or views). Divide once to get the pooled rate, which answers "how engaged was my whole audience". If you want to know what a typical post does, take the rate of each post and use the median, which one viral post cannot move much. Name which of the two you report. Recompute it every month with the same definition and watch the trend. A post that beats your own baseline by a wide margin is worth studying; a number from another account is not.

What are the limits of this page?

Tested by us: the Koldflux engagement rate calculator on 10 cases and the Elfsight calculator on the same cases (20 runs), against exact arithmetic. Computed by script: every percentage in the examples. Read, not tested: the Instagram and YouTube Help pages and the Socialinsider formulas, all read on October 28, 2026. The audience numbers in the examples, including the per-platform and baseline examples, are made up to show the arithmetic (worked-examples.py). The real-data table uses public Mastodon counts only; we did not use data from a real Instagram, TikTok or YouTube account.

Not covered: TikTok's own metric definitions (we could not read them), engagement on X, LinkedIn and Facebook beyond the Socialinsider formulas quoted, paid-post engagement, and benchmarks. Instagram marks its Viewers and Accounts engaged metrics as estimated and in-development, so those figures can change.