Social Media Algorithms Explained: What You Can Measure and What Nobody Can

Articles that explain a social feed read like they have seen the source code. They have

not. No platform publishes its ranking weights, so every precise-sounding claim about

them is reconstruction. We run a website with a 30-day distribution record we can show,

and this article uses it to separate what any algorithm observer can verify from what

nobody can.

What a ranking algorithm actually is

Every feed and search ranking system, social or otherwise, shares one structure. It

scores content against signals, orders the results, and measures what the user does with

them. The measurements feed the next round of scoring, which is the feedback loop behind

every "the algorithm killed my reach" story you have read.

The loop has three verifiable stages and one sealed box. Distribution, the impression

count, is observable from a platform's insights page or a search console. Engagement,

the click and interaction counts, is observable the same way. The next round of

distribution is observable over time. The scoring function inside the box is not

observable, and platforms describe it only in qualitative terms, named factors without

weights.

A 30-day distribution record you can check

Our own record comes from search rather than a social feed, and that is the point: it is

the same observation problem with numbers we can publish. Google Search Console reported

for the 30 days ending 2026-09-18 exactly 11 impressions and 1 click, the click from

Turkey. One click on eleven impressions is a 9.1 percent click-through rate.

That rate sounds respectable, and it means nothing, because eleven impressions is too

small a sample to estimate anything. Two more clicks would swing the rate to 27

percent. One lost impression would move it a full point. Any conclusion drawn from this

window would be a story, and the same caution applies to a week of post insights on a

small account.

What withheld distribution looks like

Before content can be measured, a system must decide to distribute it. Our record shows

what non-distribution looks like from outside. On 2026-09-24 an automated inspection of

20 URLs returned 12 crawled but currently not indexed, 6 unknown to Google, and 2 server

errors. Six of the twenty had never been crawled at all.

The crawled-not-indexed state is the exact analogue of a social platform fetching your

post, rendering it, and declining to show it. The system looked, scored, and withheld

distribution, and it explains nothing. An earlier report on 2026-09-08 found only the

homepage indexed, out of a sitemap holding 1495 URLs across 15 language clones, and an

RSS feed showing 28 submitted and 0 indexed.

Part of that was self-inflicted and verifiable: every non-English page ships a

noindex,follow directive by design, so thousands of pages were explicitly excluded from

distribution on purpose while the domain earned crawl budget. When your reach collapses,

rule out your own equivalent first, because ours was a directive we wrote ourselves, not

a hostile scoring decision.

Rules for reading tiny samples

1. Treat any rate computed from fewer than 100 events in the window as directional

only, and never as a trend.

2. Compare only rates built from the same source and the same window length. Mixing a

7-day platform insight with a 30-day console export produces fake movements.

3. Read the absolute counts next to every rate. Nine percent of eleven and nine percent

of eleven thousand are different facts.

4. Before blaming scoring, check exclusion. Robots directives, removed accounts, expired

tokens, and private profiles withhold distribution without any scoring involved.

5. Recheck a surprising number after one full cycle before acting on it, because early

reporting in both search and social platforms is routinely incomplete.

What honest algorithm reporting looks like

Platforms do publish some mechanics: which surfaces rank separately, which families of

signals matter, which behaviors violate policy. That layer is quotable. What no platform

supplies is the weight of one signal against another, so any article asserting exact

percentages of influence is inventing precision. The test to apply is simple: ask which

number in the claim came from a platform's own published data or the author's own

measurement. If the answer is neither, the claim is decoration.

We apply the same standard to ourselves here. We can tell you our 30-day record, our

index states, and our own noindex decision, because those are measured. We cannot tell

you why Google scored our pages as it did, and anyone who claims that insight about any

platform is selling something.

Build your own feedback record

1. Pick one distribution channel and one window length, and freeze both for at least

three months.

2. Log impressions, clicks, and the derived rate per period, from the same source every

time.

3. Record configuration changes the day you make them, directive flips, redesigns,

posting schedule changes, because a shift without a logged cause is unfixable.

4. Review only month against comparable month, and only after the platform's own

reporting lag has passed.

5. When a number moves, check exclusion before scoring, then check absolute counts

before rates.

Our traffic overview tool runs this log: monthly snapshots of sessions, users,

pageviews, bounce rate, average session duration, and source, with month-over-month

change and CSV export, stored only in your browser.

Checklist before you trust an algorithm claim

• The claim names its source: platform documentation or the author's own measurement.
• Rates you act on come from at least 100 events in the window.
• You checked your own exclusions before blaming the scoring box.
• You logged configuration changes at the time you made them.
• You know which of your numbers describe distribution and which describe engagement.

The algorithm is knowable at its edges, in impressions, clicks, and the states between

them. The sealed middle stays sealed, and your own record is the only part of it you

will ever own.

What does your own 30-day distribution record show? Log it in the

[traffic overview tracker](https://webrecast.com/en/traffic-overview) and compare three

comparable months before you draw a single conclusion.