Why endorsements are worth almost nothing, and what replaces them
Any signal that costs nothing to give ends up meaning nothing. Endorsements were free, so this was always where they were heading.
There is a particular kind of professional comedy in being endorsed ninety-nine times for a skill you have used twice. It is not fraud. It is what happens to any signal that is free to give: it inflates until it carries no information, and everyone quietly stops reading it.
The economics of a worthless signal
A signal is worth roughly what it costs to produce. Endorsing someone costs one click and carries no accountability, so supply is effectively infinite. Worse, a large share of endorsements are reciprocal, which means the underlying data is not measuring ability at all. It is measuring social reach and politeness.
- The endorser often has no direct knowledge of the skill they are confirming.
- Nothing distinguishes a manager who watched you work from a stranger clearing a notification.
- Volume is trivially manufactured, and the display rewards volume.
What a real evidence model does instead
The fix is not to ban human testimony. It is genuinely useful, it is simply weighted wrongly. A credible model does three things a click count never can.
- It caps by kind. Testimony is inherently weaker than demonstration, so no quantity of it reaches the top tier. Ten friendly vouches structurally cannot outrank one live demonstration.
- It discounts by independence. Five endorsements from the same company domain are mostly one endorsement repeated, and should count roughly like one.
- It verifies the source. An endorsement from a proven identity at a corporate domain carries more weight than an anonymous one, and the model should say so explicitly rather than pretend they are equal.
Why this matters more than it sounds
Once evidence is weighted properly, gaming stops being profitable. There is no cheap route to the top of a system whose cheap routes are capped by design. That is what makes a score worth reading, and it is the whole difference between a number that describes you and a number that describes your network.
The uncomfortable implication is that a credible system will rank some well-connected people lower than they expect. That is not a flaw in the model. That is the model working.
Frequently asked
Do recruiters actually look at endorsements?
Rarely as evidence. Most treat them as decoration, because the cost of giving one is a single click and there is no relationship between the endorser's knowledge and the endorsement's weight. A signal anyone can give about anything converges on noise.
What makes a skill signal credible?
Three things: it should be expensive to fake, weighted by the independence of its sources, and checkable by someone who was not there. Live demonstration meets all three. A click meets none of them.