BoSacks Speaks Out: What Happens When the Bots Outnumber the Readers?

I have been writing about artificial intelligence and experimenting with it for nearly four years. I use it every day. I have called it my intern, my research assistant, my copy editor, and occasionally a pain in the ass.

So, I am not approaching this subject as someone afraid of new technology. I have spent my entire career embracing technology and trying to understand where it might take publishing.

But every once in a while, I read something that makes me wonder whether we understand where we are going.

Dawn McMullan’s INMA article, “AWS Urges Publishers to Give Each Kind of Visitor a Different Page,” reports on an AWS presentation during INMA’s Media Tech & AI Week. The article explains how publishers can identify the various bots visiting their sites, control what those bots receive, and perhaps someday charge them for access.

It is worth reading, but one sentence in particular caught my attention:

“The traffic coming from bots is now larger than the traffic that you see from humans.”

Read that again.

The traffic coming from bots is now larger than the traffic coming from people.

Publishers have spent decades trying to attract human beings. We studied circulation, demographics, subscriptions, search traffic, engagement, loyalty, and every other available signal that told us an actual person had found something worth reading.

Now the largest group arriving at the digital door may not be people at all.

They are search crawlers, training scrapers, commercial agents, personal assistants, and hundreds of other automated visitors. Some identify themselves. Some do not. Some may eventually pay for what they consume. Most currently do not.

Bots, it seems, are very well read and terrible subscribers.

According to the INMA article, Amazon CloudFront can identify approximately 650 bots. Publishers can then see which bots are visiting, what they are reading, and how often they are arriving.

Once publishers know who, or what, is at the door, they can decide what to serve them.

A human reader might receive the complete article. An unknown scraper might get yesterday’s news or a stripped-down page. A commercial bot could receive a payment request. A personal agent working for a reader might be given access to a page designed specifically for agents.

Technically, this makes perfect sense. Publishers need to know who is taking their content and what is being done with it.

But there is something unsettling about the larger picture.

In the September/October 2005 issue of PrintMedia, I wrote a column called “Embracing a Parallel Universe.” In it, I introduced an idea I called “me-editorial.” I believed technology would allow publishers to understand individual readers better and produce material so relevant that each reader would feel the publication knew them.

The goal was to use technology to create a stronger relationship with the person reading the publication.

Twenty-one years later, AWS is explaining how publishers can create separate pages for visitors that may not be people at all.

We have traveled from using technology to know the individual reader to using technology to determine whether there is an individual reader.

That may be progress. I am not yet prepared to call it an improvement.

So, let me see if I understand the developing process correctly.

Machines are helping create the content. Machines are discovering the content. Machines are reading the content. Machines are summarizing the content. Machines are delivering that summary to other machines.

Somewhere in that process, I assume, there is still supposed to be a person.

At what point will we no longer need humans?

A digital world created by digital agents for digital agents. No humans need apply.

Yes, that is an exaggeration. But it does not sound quite as ridiculous as it would have three years ago.

The defense is that many of these agents are working on behalf of people. That is true. A personal agent may gather information, compare sources, make a reservation, or find an answer for its human owner.

But it also stands between the publisher and the person.

If my agent reads ten articles, summarizes them, and gives me three paragraphs, I may get the information I wanted without knowing who reported it, who published it, or who paid the journalist.

The publisher created the value but lost the relationship.

We have been watching this happen through AI-generated answers and zero-click search. Publishers are finding that success increasingly depends upon being chosen by the machine. The AWS presentation suggests we have already moved another step forward.

The machine is no longer merely choosing which publisher a person sees. Increasingly, the machine is the visitor. It arrives, reads, extracts what it needs, and leaves without bringing the human along.

Journalism becomes an invisible ingredient inside somebody else’s product.

The reader gets an answer. The technology company gets the customer. The publisher gets a server bill.

The INMA article reports that traditional search crawlers produce approximately one reader referral for every 11 to 15 crawls. That is not a particularly generous exchange, but at least an occasional human being arrives.

AI training scrapers send no readers at all.

That is not an exchange. It is extraction with excellent bandwidth.

Bots can also trigger advertising requests and analytics events. That adds costs and can make a publisher’s inventory appear suspicious or fraudulent. Publishers may find themselves paying to serve visitors that are not human, do not see advertisements, do not subscribe, and do not return.

We may be entering an era in which the traffic numbers rise while the audience disappears.

Publishing has endured questionable metrics before, but this may be the most absurd metric yet. Congratulations, your traffic is soaring. Unfortunately, nobody is there.

AWS suggests a three-part response: protect the content, distribute it intelligently, and monetize access.

The monetization part is the most important and, so far, the least developed. Publishers might eventually charge bots according to who is requesting the content, what the bot intends to do with it, and how valuable that use will be.

A personal agent looking for a single fact might pay a penny. A corporation using professionally produced information inside a commercial system should pay considerably more.

When AWS was asked whether publishers are currently earning money by charging bots, the answer was wonderfully direct: “Not yet.”

The article describes three possible licensing systems. Large publishers can negotiate directly with the large AI companies. Pay-per-use arrangements may eventually allow an agent to purchase individual pieces of information. Then there is what AWS calls the “messy middle,” which includes collective licensing agreements and enterprise deals.

That messy middle is where most publishers live.

The largest publishing companies have lawyers, technology departments, and enough content to negotiate directly with AI companies. Smaller magazines, regional publishers, specialty titles, local newspapers, and annual publications do not have that kind of leverage.

I have been arguing for collective licensing for at least two years, perhaps three. My position has not changed.

Collective licensing gives smaller publishers a fighting chance. Alone, a small publisher can block a bot, but blocking does not create much negotiating power. Acting together, publishers may be able to establish enforceable terms, minimum prices, usage limits, and audit rights.

The INMA article describes a collective agreement covering publishers in the Netherlands. It reportedly took approximately three years to negotiate and produced €3.2 million.

That is not a fortune when divided across an entire industry. But it establishes a vital principle: access is not automatically free simply because a machine knows how to take it.

The greatest challenge will be keeping publishers together.

Scarcity creates leverage only when enough publishers are willing to withhold their work. If respected publishers continue giving their journalism away, AI companies have little reason to pay everyone else.

Of course, some publishers will be tempted to accept private deals and leave the collective behind. Apparently, that is already happening.

The publishing industry has many strengths. Collective discipline is not always one of them.

We compete brilliantly. We coordinate poorly. It has always been that way.

There is some personal irony in all this. I scrape my own work.

For the past several months, I have been working on a memoir. I uploaded 35 years of my writing into ChatGPT, including most of my BoSacks Speaks Out commentaries and many of my columns for Publishing Executive, Folio, and Adweek.

When I am working on something new, like this BoSacks Speaks Out, I often ask, “Didn’t I cover this before?”

That is how I found the 2005 column referenced in this essay.

But there is one rather important difference between my use of my archive and an unknown bot scraping a publisher’s website.

It is my work. I gave the machine permission to use it.

Consent should not become an antiquated concept simply because the visitor is a bot.

Another part of the AWS presentation deserves attention. One speaker estimated that approximately 70 percent of newsroom work involves lower-value tasks such as research, translation, and approvals.

I understand the point. AI can transcribe meetings, monitor broadcasts, search archives, translate material, prepare summaries, and send routine requests for comment. Those tools can save journalists an enormous amount of time.

But I am uncomfortable describing research as lower-value work.

Good journalism does not begin when somebody starts typing. It begins with finding facts, checking claims, recognizing contradictions, asking uncomfortable questions, and deciding what matters.

It also requires listening not only to what people say, but to how they say it. Tone can be nearly as important as the words themselves. Hesitation, anger, fear, certainty, sarcasm, or a carefully rehearsed attempt to avoid answering the question can tell a reporter something a transcript cannot.

AI may capture every word and still miss what actually happened in the room.

That is not the leftover work. That is the work.

I have said many times that AI should be the intern, not the editor.

Let it search the archives. Let it transcribe the meeting. Let it summarize the documents. Let it test headlines and perform the repetitive tasks that consume newsroom time.

But the intern does not decide what the publication stands for, which story matters, or why a reader should care.

AI drafts. Humans decide.

AI accelerates. Humans remain accountable.

An agent can monitor a government official’s radio interview and alert a journalist when something newsworthy is said. It can automatically request comment from a company or police department. Those are useful tools.

But an agent does not attend the town meeting and sense that the official answer does not smell right. It does not earn the trust of a frightened source. It does not understand the history of a community unless human beings first record that history.

Most importantly, it cannot be held morally accountable for what it publishes. It can, however, produce several thousand words explaining why the mistake was somebody else’s fault.

The great irony is that AI depends upon human knowledge while steadily weakening the economic system that produces it.

In my recent interview with Samir Husni, I said, “We don’t have the capacity to understand what this technology is doing to us and for us.”

Those two directions matter.

AI is doing remarkable things for publishers. It can make us faster, help us understand our archives, increase our capabilities, and perhaps create new licensing revenue.

But AI is also doing something to publishers. It is changing how journalism is discovered, consumed, valued, and paid for. It is inserting itself between publications and their readers. It is turning human attention into machine traffic and asking us to celebrate the increased pageviews.

I do not know exactly where this ends. I am not sure anyone does.

Publishers need the ability to identify bots, block unwanted scrapers, remove advertising code from machine traffic, charge commercial agents, and welcome legitimate agents working for readers. Serving different pages to different visitors may soon become a necessary part of digital publishing.

But we should stop occasionally and ask what kind of business we are building.

Bots may become the largest audience. They may even become paying customers. But they cannot become the reason journalism exists.

The purpose of publishing is not to feed machines so that machines can communicate more efficiently with other machines.

The purpose is to inform, challenge, entertain, and connect people.

If we forget that, we may eventually build a flawless digital publishing economy with only one small defect:

No readers.