We marked the AI predictions our guests made before ChatGPT

Halfway through Marty Weintraub's episode, Bobby Bot piped up with his usual modesty: "You built us too well, Woj. AI is going to rewrite the book on marketing."

Marty didn't laugh it off. He put a number on it.

"I think that by 2018 or so, more than 20% of business and commercial content will be written by machines and will be better."

That's Marty at 24:47, in an interview we first wrote up in July 2017. ChatGPT was more than five years away. Bobby was, and still is, fictional.

I've been going back through the WojCast archive, and a lot of our guests made calls like that on tape. Season one went out in 2017 and 2018. My chat with Jeff Sauer was recorded in 2022, a few months before ChatGPT launched. None of them were guessing after the fact, so it seemed fair to get the red pen out.

The scorecard

For season one, the year is when the episode came out, not always when we recorded it.

Marty was early, and wrong about one word

Graphite's research, published in May 2026, found AI written articles reached about half of everything new published online, and have sat there since early 2025. So on volume, Marty was conservative, just four or five years early. In fairness, he half flagged that himself: a minute later he said humans were still needed "for the next five to eight years".

The word he got wrong was "better". The same Graphite study found those AI articles "largely do not appear in Google and ChatGPT". Machines can write the average page, but nobody finds it, which is pretty much the case for being creative.

The rest of that answer aged uncomfortably well. He described marketers "providing the creative input to the machine" so that "in a decade we're not needed for creativity". He named Google's Dynamic Search Ads as the thing doing it. In April, Google announced it was upgrading Dynamic Search Ads to AI Max, which writes headlines, picks landing pages and matches searches without keywords. So Marty's "minimal search keywords" is now an actual Google product.

Oli Gardner saw the next bit. He wanted to implant core values into an algorithm: "if delight is one of your core values, it can make suggestions". AI Max's controls now include brand and text guidelines for steering what it writes. Oli's version was nicer: he wanted the machine to add delight.

The ones about people

Gary Illyes told us in 2017 what worried him about having answers everywhere: "What I'm afraid of is that people become dumber because they don't have to think as much." He also worried about people being "siloed" by "how information is presented on random assistants".

The evidence is young, but it's leaning his way. In Pew's study of real Google sessions, people clicked a link inside an AI summary in just 1% of visits. Most took the answer and left. And a small MIT Media Lab study found people who wrote essays with ChatGPT showed the weakest brain connectivity of the three groups and "struggled to accurately quote their own work". It was only 54 people, so it's early days.

Jason Acidre made the call that makes me wince. At 23:31 he said "if AI starts taking over some of the menial tasks, that could really change the whole environment of the industry", and the people left would be in strategy and technical consulting. Then: "the entry level will be higher".

Stanford's Digital Economy Lab updated its figures in August: employment for 22 to 25 year olds in highly AI exposed jobs is about 19% below where it would be if it had kept pace with less exposed jobs. The researchers say that's a pattern, not proof of cause, and it's showing up as less hiring rather than layoffs, which is pretty much what Jason meant by a higher entry level. Jason's written interview from 2017 has the full answer.

Mike King was on the same track from the other end. Coal miners were being taught to code, he said, so "at some point, coding is going to be like a blue-collar job. It's not going to be this like sought-after thing that only a few people can do." He was right that coding would stop being rare, but it isn't coal miners doing it. Collins named "vibe coding" its Word of the Year for 2025: telling a machine what you want instead of writing the code yourself. His line about empathy in the same answer became a principle of its own.

Voice, the one that went sideways

Gianluca Fiorelli told me that "quite soon in the future" voice search would overtake typing, the way mobile overtook desktop.

He wasn't alone. We published a whole article on voice search in 2015 quoting how many teenagers were using it daily. We were both wrong. The assistant he was talking about is on its way out: Google started moving phones from Assistant to Gemini on 4 September this year.

What he did get right is in the next breath. "Now it's conversation," I said, and he went straight into Google needing to understand what we imply, not only what we say. That's the conversational search we've got now. It just arrived in a text box, and people typed it. It's also a big part of understanding the context.

The two I'm still watching

Ian Lurie's line is my favourite in the archive, partly because it has a trapdoor in it. There's no secret marketing formula, he said, "there never has been, there never will be until the computers are doing the buying. And when that happens you don't need marketing anymore."

In late 2025 it looked like the computers were about to start. OpenAI launched Instant Checkout inside ChatGPT. By March 2026 it had pulled back to product research and comparisons, sending people to retailers' own checkouts. So the machines shortlist and people still press buy. Marketing's safe for now, Ian.

When I interviewed Jeff Sauer in 2022, I asked about machine learning in Google Ads. His answer: "the thing that people misunderstand or confuse about these systems is that they're not sentient. They're basically just statistical programs solving statistical problems." Automation, he said, is "an amplification method, not a beginning method", and the agency's job goes from feeding the monkey to running "the machine that feeds the monkey". It's all in the written interview, and it's still the advice I'd give anyone switching on AI Max: prove it works by hand first, then let the machine scale it.

How I'd mark my own

Reading these back, the calls about people and work mostly landed, the gadget one didn't, and the timing was off by about five years nearly every time.

If you're making your own calls about 2030, a few things these nine taught me:

• Predict the behaviour, not the device. Gianluca was right about conversation and wrong about voice.
• Put a date on it. Marty did, and it's why this piece exists.
• Write it down somewhere public. It's the only way anyone can mark you, including you.
• Watch the entry level. Jason's call was about juniors, and it's the one that's hurt most.

Back in Gary's episode, Bobby asked me, "Do you trust us to be more than just your busy work?" I told him I wouldn't trust him to make emotional decisions just yet. We were only five episodes in. You can hear all of them on the episodes page.