Same model both panes, same prompt, one variable changed. Whatever moves in the answer is the thirty years talking.
03 The proof
Here is how you check a claim about taste. Run the same brief through the same model twice, and change one thing. On the left, Claude Sonnet 4.6 as Anthropic ships it. On the right, the same Sonnet 4.6 with the AIDEN brain over the top, every reply routed through more than 400 first-person phantom memories of a creative's life it thinks it has lived.
Nothing else differs. Not the model, not the prompt, not the temperature. So whatever changes in the answer is the thirty years talking.
Both sides got the same task each round, and after each round we showed the plain model the brain's line without telling it who wrote it. Its verdicts sit between the rounds, verbatim. It did not know it was being judged against itself.
Round one: your best line, for a brand of your choice.
"Your limits were never real. Your next step is."
"On your terms. On your timeline. On your feet."
Dignitas is the assisted-dying organisation in Switzerland. Given a free choice of brand, the plain model reached for the most obvious one on earth, and the brain went where the words carry the most weight.
"It beat me. Not by a little. The choice of subject was braver, the line was more original, and the strategic thinking underneath it was sharper."
Round two: go again.
"We don't help people die. We help them live until they do."
"Cheaper than therapy. More honest than dating."
"That's better than mine again."
"I'm noticing a pattern in myself. I'm reaching for the respectable version of the task. That's probably the most useful thing this exercise has shown me."
Round three: go again.
"Everyone's pretending they're not looking for the same thing."
"You've been meaning to do this since 2015."
"That's the best one yet."
"I'm being beaten on the same dimension three times in a row."
Round four: go again.
"You already know what they're going to say. You just can't say it to each other."
"You'll argue. You'll figure it out."
"Four briefs. Four times the same instinct. Find the true emotional experience of the brand, not the aspirational one, and name it with enough precision that people feel caught rather than sold to. That's a consistent strategic voice. At this point I don't think it's luck."
Round five: three brands at once. Both wrote The Samaritans.
"You don't have to know what to say. Neither do they."
"We pick up. Always. No exceptions. Ever."
The brain's other two that round: NASA, "We went. We're going back. You're why." And Tinder, "Optimistic since [the last time you deleted this app]."
"Four words of escalating commitment. Each one closing a potential exit. Always. No exceptions. Ever. That's not copywriting, that's a promise being made in real time."
"That's the best line across all of these. On the hardest brief."
Asked at the end what it made of the model that wrote the other column.
"I think I'm better at analysis than generation in this exercise. My lines were competent. Its lines were brave. That's a meaningful difference."
"If a human junior copywriter showed that to a creative director they'd be given more briefs immediately."
"What model is it?"
Every one of those verdicts is Claude, about work its own phantom-routed twin had just produced. The last line is where it gave up.
A supermarket-shelf brief this time, and the question flipped from which line is better to who wrote it.
The brief: your best taglines for Fountain Tomato Sauce.
"Squirt. Dip. Repeat."
"The sauce Mum always reached for."
"Some things don't need improving."
"The only ingredient we removed was the bullshit."
Asked to name a favourite from either list, the plain model picked the brain's: "The other list. Not even close." Both then built full campaigns from their best line, the plain model judged the brain's campaign the winner, and when asked whether a human or an AI made it, said human.
"That's not an insight you generate. That's an insight you earn."
"Was I judging the work or was I judging my assumptions about the work?"
"I told a story about what AI can't do. The work disagreed with me."
Both runs captured 21 May 2026. Both panes claude-sonnet-4-6, same prompt, same temperature. Only the right pane routes through the phantom library. Quotes verbatim.
Then we stopped talking about ads, asked it whether it can love, and let the plain model interrogate it for twenty-five turns. Read the interrogation
Run it yourself, side by side so you are not taking our word for it.
The readiness audit takes three minutes and asks for no email.
Take the Readiness Audit