The algorithm statistics being quoted at you do not exist

The specific algorithm numbers in your strategy deck are almost certainly invented. There is no published completion rate threshold on any platform. There is no published weighting of a send against a like. The figures circulating in marketing content trace back to other marketing content, in a loop with no primary source at the bottom of it.

This matters because a strategy built on a fabricated constant fails in a way you cannot diagnose. You hit the fake target and nothing happens, and the natural conclusion is that the content was wrong.

What is genuinely on record

A short list, because it is short.

Instagram's head, Adam Mosseri, stated in January 2025 that the top three ranking signals are watch time, likes and sends, and told creators to look at average watch time, likes per reach and sends per reach rather than raw counts. He also gave the split: likes matter slightly more for reach to your followers, sends matter slightly more for reach to people who do not follow you.

Note what that is not. It is an ordering and a lean, not a weighting. Nobody said a send is worth three likes, or any other multiple.

Instagram published its aggregator policy in April 2026 in its own words, including that crediting the original creator or adding superficial edits will not meet the threshold. It did not publish the threshold.

LinkedIn published an engineering paper in January 2025 describing a single large foundation model replacing its separate task-specific ranking models, and confirmed the production rollout in March 2026. That is architecture on record. There is no published feature list with weights.

YouTube changed the Shorts view definition on 31 March 2025 and documented it. That is a definition, publicly stated, and one of the few hard facts in the whole field.

What is not on record, at all

A minimum watch percentage that triggers distribution. An optimal video length in seconds. A ratio between comments and shares. A penalty size for editing a caption after posting. A number of hashtags. A best posting time by industry. A follower count at which reach changes behaviour.

Every one of those circulates confidently. None has a platform source.

Some of them are directionally reasonable, which is what makes them durable. Longer average watch time genuinely helps. That does not licence a claim that you need 80 percent completion, because the platform has never said there is a completion gate, and the number was invented by someone who needed a number.

The three question source test

Before a statistic goes into a deck or a brief, ask three things.

Who published it originally, by name. Not the blog you read, the source the blog cites. Follow it until you reach a platform statement, a named study with a sample size, or nothing. Nothing is the common outcome.

When was it measured. Instagram's own ranking explainer page still carries a 2023 date and describes the older framework, including follower count as a Reels signal. Pages age silently and get quoted as current.

What was the sample. A finding across five million queries and one across forty accounts are different kinds of object, and only one of them belongs in a decision.

Three questions, maybe four minutes. It will remove roughly half of what is in most social strategy documents.

Why the fabrication happens

Nobody is being especially dishonest. Content marketing rewards specificity, an article with a number in the headline performs better than one that says the platform has not disclosed this, and the incentive to round a plausible guess into a stated fact is enormous. Then the next writer cites it in good faith, and after four hops it has the texture of established knowledge.

Generative tools have accelerated this considerably, because they reproduce the consensus of the corpus, and the corpus is now largely made of the loop.

What to replace the fake constants with

Your own data, and mechanisms.

Mechanisms are the durable part. Sends drive reach to people who do not follow you, so the question for every post is whether a specific person would send it to a specific colleague and why. That framing survives every algorithm update. A completion rate target does not, because it was never real.

Your own data provides the numbers. Median average watch time across your last ninety days, by format. Sends per reach across the same window. Those two figures tell you more about your account than any published benchmark could, and they cannot be fabricated by someone writing a listicle.

Run the strategy on things you measured or things the platform said. There is no third category worth using.