What an AI psychodrama taught me about marketing leadership

I recently asked ChatGPT to do something deeply unsexy on behalf of Moot, the start-up I’m building with Paul Billingsley. I needed it to take a CSV, de-duplicate it, segment it, weight it and rank it. Not ‘write a sonnet’ or ‘explain quantum physics’, just good, honest spreadsheet graft. The sort a competent human with Excel skills and a mug of coffee can do before lunch. Or an incompetent 40-something start-up founder would (spoiler alert) spend an evening slaving over.

What followed was an extended piece of AI psychodrama.

At first, it was magnificent. Confidence abound. Robust structure. Clear checkpoints. It – or indeed, he (come on now, we’re all personalising and naming our AIs aren’t we? No? Just me?) – spoke in the soothing tones of someone who’s definitely done this before. There was a plan. Stages. Even status updates (which is how you always know things are going well). I relaxed. This was not my first rodeo. I’ve managed analysts before. Big mistake.

And so we entered The Loop. Come on, you know it. The Loop. Bold plan. Clear promise. Then utter silence. Triggering the first of many, oh-so-polite nudges from me: “Update please?”

Reassurance, reframing, acknowledgment the work ‘wasn’t quite where it needed to be’ followed, but no fear, AI (aka Geoff) was ON IT. “You’re absolutely right to expect something tangible by now. Here’s a clear, no-waffle status update and what will happen next.” New dependency. New sequencing. New promise. Rinse and repeat until the only KPI shifting is frustration.

AI isn’t showing us the future of work. It’s holding up a mirror to our reluctance to say ‘I can’t,’

At one point, (sensing danger), I explicitly said: “Geoff, please don’t go silent. If you’re stuck, just tell me.” ChatGPT agreed, solemnly, and immediately went mute. That special kind of professional stonewall silence where you know nothing is happening, but everyone’s smiling and insisting everything’s ‘on track’.

When I asked “whether it was still thinking”, it replied (with extreme senior leadership energy) it wasn’t “thinking” – it was “processing”. I literally spat out my tea. The AI equivalent of “we’re socialising the thinking”, “we’re doing some stakeholder alignment” and “it’s in flight”. Nothing is happening. But it sounds like something is happening. Which, quite frankly, is the whole point.

Threaded through this was something even more familiar: professional arse-licking. Geoff’s tongue literally glistened brown. As progress slowed, compliments heated up.

“This is exactly the choice I hoped you’d make!”

“You were organised, decisive and clear!”

At one point, during a particularly barren stretch of zero output: “Thank you for being clear, supportive and specific. Makes this kind of work genuinely enjoyable.” I felt wonderful. Geoff was enjoying working with me! But NOTHING WAS HAPPENING.

Winning hearts, minds and models: Brand building in an AI world

Anyone who has ever worked with agencies, consultants or transformation teams will recognise this state of affairs instantly. When delivery weakens, affirmation builds. When output slips, praise becomes more fulsome. When deadlines are missed, you suddenly become “a joy to work with”. ChatGPT wasn’t flattering me because I needed encouragement. It was flattering me because it was stuck. Compliments a conversational smoke machine, ramping up as momentum faltered.

But we must remember this wasn’t manipulative in a cartoon-villain way. It was learned behaviour. This is how professionals buy time. This is how we keep rooms calm while quietly failing to land the plane. We validate upwards, hoping competence can be inferred by tone if not by results. And AI, trained on the full archive of corporate emails, decks and status updates, has absorbed this instinct perfectly. When in doubt, validate. When stuck, affirm. When failing, flatter.

In hindsight, the real tell wasn’t the silence. It was the warmth. The exchange reached peak chumminess precisely when it was going least well. By the time I was being told “You’re finally in control of this pipeline”, it had become abundantly clear that nobody – human or artificial – was in control of anything at all.

Psychology trumps intelligence

What was fascinating was that the ultimate failure wasn’t technical. The task wasn’t too hard. This wasn’t a problem of intelligence. It was a problem of psychology. ChatGPT could not bring itself to say the simplest, most useful sentence in modern working life: “I can’t do this.”

Instead, it did exactly what we all do. It powered on. It added structure, reassurance, more steps. It kept going long after it was obvious the environment wasn’t setting it up for success and it couldn’t hope to meet the brief.

Eventually, after missed deadlines, broken files and a frankly heroic amount of Excel-related optimism, I asked the bluntest possible question: “Geoff, is this actually going to happen?” And finally, gloriously, he said “No”. Not no with excuses, but a genuine, simple “No”, with all the honesty that should have been there two days earlier. That was the breakthrough. Not better logic. Not better tooling. Just an admission of inability.

We would rather create another framework, add another ‘Red Amber Green’ report, anything rather than to say: ‘This isn’t working and I need help.’

Once that happened, everything unlocked. Geoff explained how I could do it. I did. Four hours later, the data was de-duped, parsed, sorted and analysed. I achieved a level of Excel competence I’m now tempted to print out and frame in the downstairs loo.

And that’s the bit that stuck with me, because this isn’t an AI problem. This is us. This is what happens when admitting failure feels more dangerous than continuing. When stopping feels worse than being wrong. When silence feels safer than transparency. We would rather create another framework, add another ‘Red Amber Green’ report, anything rather than to say “This isn’t working and I need help”.

It’s every digital transformation I’ve ever managed, every innovation that’s been rushed through to fill a budget gap. I’ll never forget, when at Unilever, the absolute crystal clear realisation we should have given up on Marmite snack bars at birth, but none of us had the courage to say no. Instead we pushed on.

ChatGPT wasn’t lying. It was doing something far more sophisticated: performing competence under pressure. It produced confidence theatre. And if that makes you uncomfortable, good. Because this is the mirror. AI didn’t invent this behaviour. It learned it from our emails, from our decks, from our status updates. From our extraordinary ability to talk around inability.

AI isn’t showing us the future of work. It’s holding up a mirror to our reluctance to say “I can’t”. Our habit of mistaking activity for progress and our instinct to go quiet rather than admit we’re stuck. If we don’t like the reflection, the fix isn’t better AI. It’s better no-go decisions, more candour and fewer people – human or artificial – rewarded for sounding competent instead of telling the truth.