WANTED: One Thing the Robot Cannot Summarize

TL;DR: Dan Oshinsky argues that an email address is no longer a distribution channel, because something else now reads the mail first. His three answers are brand, community, and building past the newsletter itself. The ad copy underneath all three says the same thing: sell what does not survive being condensed.

WANTED: One (1) reason to open the actual email.

Must survive summarization. Must not be reproducible from a subject line. Must not be the information itself, as the information will be extracted, condensed, and served to the reader in a bulleted list by something that does not care whose byline was on it.

Compensation: the only audience anybody keeps after this.

I have written that ad a thousand times for other people. Sixty cents a word teaches you fast which ones are load-bearing, and the word nobody wants to pay for is reliably the one doing the work.

So here is the trade in one line. Anything that can be summarized will be. And anything that can be summarized was never worth sixty cents a word to begin with.

Serious inquiries only.

What exactly is arriving in the inbox?

Oshinsky’s piece lists the features going into mail clients right now, and none of them are speculative. Better search. Automatic summarization of long emails. Filtering that decides which messages are important and which ones contain something you have to act on. Drafted replies written in your voice. Daily briefings, with Gemini’s version pulling a task summary out of Gmail and Calendar and handing it over as a single digest.

Read that list as a sender instead of as a user. Every one of those features sits between your writing and the person you wrote it for. The summarizer is a reader who arrives first, has no loyalty, and produces a shorter version that the human may accept instead of yours.

Oshinsky’s line is the whole ad: “Simply getting an email address isn’t going to be enough anymore.”

Which is a hell of a thing to say to an industry that spent ten years telling people the email address was the asset. It was. Then somebody built a middleman into the delivery vehicle.

Sale item one: a voice the summarizer flattens

The first answer is brand, and the useful version of that word is narrower than the usual one. Not a logo. A recognizable way of talking that a reader can identify with the sender’s name stripped off.

Oshinsky’s examples are the ones I would have picked. Axios sells smart brevity, which is a format promise, so the format is the product. Sioux Falls Simplified sells an explanatory structure. The Onion keeps its voice intact even in transactional email, which is the tell of an outfit that understands what it actually owns.

Here is the mechanism, and it is not sentimental. A summarizer is very good at extracting propositions and very bad at carrying tone. Feed it a newsletter that is five facts in a neutral wrapper and it will return five facts and lose nothing, because there was nothing else in there. Feed it something where the way the thing is said is the value, and the summary comes back as a description of a joke rather than the joke.

If your newsletter can be losslessly compressed, it will be. That is not an insult, that is a description of what compression does to redundancy.

Sale item two: a room the software cannot enter

The second answer is community, and Oshinsky’s examples all share a shape: they put something on the far side of the email that email cannot deliver.

BackScoop runs regional events, introduces founders to each other, and sells an Inner Circle membership whose product is curated introductions. The New Fatherhood built a Discord, funded therapy, and sells coaching. Bulletpitch looked at an audience full of founders and investors and started an investment syndicate. Midcoast Villager opened a cafe.

A daily briefing feature can tell you a newsletter arrived. It cannot introduce you to anyone. It cannot put you in a room. The membership sale here is not “more content,” it is proximity, and proximity has never once been summarizable.

I will flag the cost honestly, since this is a classifieds page and not a brochure. Every item on that list is a second business. A cafe is a cafe. A Discord that nobody tends is a ghost town with your name on the door, and it will read as evidence that the operation is smaller than it claims. Do not buy this one unless you intend to staff it.

Sale item three: stop selling the newsletter

The third answer is the one with the number attached, and the number is the reason to take it seriously. Every.to built software (Spiral, Monologue), ran courses, and sold consulting, and Oshinsky reports the consulting alone brought in $215,000 in a single month, March.

That is not a newsletter business with side income. That is a business whose newsletter is the top of the funnel, which is a different thing wearing the same coat.

The uncomfortable read for a small operator: if the letter is your only product, the AI inbox is a direct threat to your only product. If the letter is how people find the thing you actually sell, an inbox assistant that surfaces and summarizes your mail is arguably doing you a favor, because a reader who gets the gist and then buys the guide has still bought the guide.

The threat model depends entirely on where your money is. Work out which one you are before you decide how worried to be.

The line I would put on the sign

Oshinsky closes on something plainer than any of the three strategies: “Readers will make time for the newsletters from the people they like and trust.”

That is either a platitude or an operating instruction depending on whether you can name the specific thing you do that earns it. “Be trusted” is not a plan. “Answer every reply personally for the first thousand subscribers” is a plan. “Publish the number that makes us look bad” is a plan. “Put a real person’s name on it and let them have opinions the sponsor might not enjoy” is a plan, and it is the one most operations flinch from.

Liking and trusting are the two things a summary cannot transfer. You cannot compress a relationship into three bullets, and nobody has built the feature that tries, because there is no feature there to build.

FOR SALE: everything else.

Frequently asked questions

Does an AI-powered inbox actually hurt open rates?

Nobody can tell you cleanly yet, and the reason is worth knowing: open-rate measurement was already compromised. Apple’s Mail Privacy Protection has been firing tracking pixels regardless of whether a human looked at the message for years now, which means the metric most senders would use to detect this problem is the metric least able to detect it. Watch replies, clicks to things that require a decision, and paid conversions instead. Those still mean something.

Is this an argument against ever writing an informational newsletter?

It is an argument against writing one whose entire value is the information, when that information is also available elsewhere. A roundup of links about a topic is close to the ideal input for a summarizer. Reporting somebody else does not have, a method you actually ran, or a judgment call with your name on it survives condensation, because the summary of it still has to say who concluded that and why anyone should care.

What is the cheapest one of the three to start on this week?

Voice, and it costs nothing but nerve. Pick the two or three things about how you write that a stranger could identify without the byline, and stop sanding them off. Most newsletters are less distinctive than their writers, because the writer keeps editing toward a neutral register they think sounds professional. That neutral register is exactly the thing a machine can now produce infinitely, for free, forever.