The Algorithm Started Paying a Premium for Humans

Hey Everyone - Hope you had a great weekend. Four major platforms made the same move in eleven days at the end of July, and almost nobody framed it correctly. The coverage treated it as more AI labeling news. It is not. Something considerably more interesting happened, and if you are building an audience it changes the math on what you should be making. Let me walk you through it.

This week:

Let's dive in.

Four platforms made the same move in eleven days at the end of July.

Snapchat announced on July 31 that only videos created by real people are eligible for Spotlight recommendations. AI tools remain fine for enhancement and editing, but a fully synthetic video stops getting distributed. LinkedIn added a "seems like AI slop" report button on July 30. Substack shipped a Pangram-powered AI detector on July 22 that lets any reader scan any post published after July 21. On July 20, YouTube clarified that "inauthentic content" cannot be monetized, explicitly including generic, repetitive, template-based work and AI personas discussing health and finance. Meta pulled an Instagram AI photo-editing feature on July 10 after backlash, and Pinterest is now labeling AI images and giving users a control to see fewer of them.

For three years the platform response to AI content came in the form of labels. Disclosure toggles, tags, honor-system checkboxes. A label leaves the decision with you, the same way a grocery store prints the sugar content on the box and still puts it at eye level in the middle of the aisle.

July was a different kind of move. Snapchat and YouTube are adjusting ranking and payment, which is the grocery store deciding on your behalf. Bottom shelf, back corner, or pulled from the store entirely. Nobody at YouTube asked whether viewers dislike synthetic content. They ran the numbers, reached a conclusion, and changed the payout.

The lesson most creators will take from this is that human-made work now gets rewarded, so make more of it and collect the premium. That reading holds up for this quarter. I think it is a trap.

The rules moved four times in eleven days. They will move again, and the next revision may cut against you. Every one of these companies has reversed itself before. The detection tools are imperfect and already flagging writers with clean, structured prose, and the appeals processes are immature. Any system with the power to put your work in front of a million people has the same power to bury it, and it will use that power on its own schedule according to its own economics, with no obligation to warn you first.

So here is where I land. Winning the ranking is a temporary position. The durable work is reducing how much of your livelihood depends on any ranking at all.

That means owning the connection directly. An email list where your message arrives without an intermediary deciding whether it deserves to. A community that exists because people chose it, not because a feed served it to them. Local relationships built in actual rooms. Customers who know your name and could find you if every platform you use vanished overnight. In each of those, nothing sits between you and the people who care about your work.

I have been building toward this for a year, and the July news finally made me articulate why. Hours spent optimizing for a feed improve an asset that somebody else owns and can reprice whenever they want. Hours spent moving one person from a platform onto your own list build something nobody can take.

I want to be clear that the July moves are good news. The platforms are responding to something real, which is that audiences got tired of synthetic content faster than almost anyone predicted. Treat it the way you would treat a favorable decision from a landlord. Useful this year, pleasant to receive, and no reason to stop looking for a house.

So take the reach while the ranking favors you. Then spend it on the one thing the platform will never hand you, which is a direct line to the people who actually want to hear from you.

Most of my last few weeks went into something with no glamour attached to it whatsoever. I built a CRM for the newsletter and the agency.

That sounds like the least interesting sentence in this issue, but has been one of the highest-leverage things I have done all quarter. Knowing who my customers are, when I last talked to them, what we discussed, and where every conversation actually stands turns out to be enormous value. Before this, all of that lived in my head and in a scattering of threads. Every dropped follow-up was revenue I never saw, and I could not tell you which ones I had dropped.

The broader shift is that I have stopped thinking about growth as content and started thinking about it as infrastructure. Systems, pipelines, the plumbing. The unsexy layer determines whether anything above it works.

Here is the part I want to draw out, because it runs against the usual framing of what AI is for. Almost all the AI conversation is about efficiency, doing the same work faster. The way I am using it here is different. I am using AI to augment my relationships. It surfaces who I have not talked to in six weeks. It reminds me what somebody mentioned about their business three calls ago. It drafts the check-in I would have meant to send and never sent.

None of that makes me faster. It makes me more present with more people than I could hold in my head alone. That is a different category of value than time saved, and I think it is the more durable one. The efficiency gains get competed away as everyone gets the same tools. Being the person who actually remembered and actually followed up does not.

The scarce skill in 2026 is choosing the humans you work with. AI collapsed the cost of execution, so the remaining bottlenecks are judgment and trust, and both of those arrive attached to specific people you either picked correctly or did not.

The traditional signals are getting weaker fast. Gartner predicts that by 2028, one in four candidate profiles worldwide could be fake. In its survey of 3,000 candidates, 6 percent admitted to interview fraud outright, either posing as someone else or having a stand-in sit the interview for them, and four in ten said they use AI during the application process. Bloomberg reported in July on a class of overlay tools that feed candidates answers in real time during live video and coding rounds, invisible even when the candidate is sharing their screen.

Gartner's recommended control is the in-person interview, and candidates seem to want it too. 62 percent said they were more likely to apply to a job that required one.

Here is what works instead, for hires and for partners alike.

Test with real work. Give a small paid project with a real deadline and a real cost if it slips. Two weeks of that teaches you more than six hours of interviews. Watch how they operate: whether they close loops, whether they flag problems early, whether they lead with the bad news. Their ability to do the task is almost beside the point.

Align the incentives before you need them to hold. Same lesson as last week's Signal, applied at human scale. Design the arrangement so their winning move and your winning move are the same move. If you cannot construct that, you are not ready to work together yet.

Then watch the small things over time. People are remarkably consistent across surfaces, and the small commitments read more clearly than the big ones because nobody performs on them. Did they reply when they said they would. Did they remember what you mentioned last time.

The upside compounds harder here than almost anywhere else. One right person changes what you are capable of attempting. One wrong one costs you a year.