Join or Die, Six Years On: What Comes After You Surrender to the Algorithm | Patrick Gilbert

Two books, six years apart. More than 500 clients. Over $1.5 billion in managed ad spend. And a thesis that did not reverse so much as move on.

In 2020 Patrick Gilbert published Join or Die: Digital Advertising in the Age of Automation. The argument was in the title: stop defending manual bidding, learn how the machine works, get on board. In May 2026 he published Never Always, Never Never: Strategic Marketing in an AI World, which argues that the shortcuts that once produced easy returns have been commoditized, and that what is left is judgment.

Side by side, those covers look like a reversal.

In the latest episode of Velocity: Performance Marketing Podcast by Hawky, host Surender (Co-Founder, Hawky.ai) asked Patrick which version of him was wrong. Patrick declined the premise, and the answer opened out into why the metric most performance teams run on is a trap, why he calls his own industry the reddest ocean he has seen, and why he is sitting out the ad platform everyone else is rushing into.

About Patrick Gilbert: From the Google Ads UI to Google's Own Certification

Patrick is CEO of AdVenture Media, founded by Isaac Rudansky in 2012. It did not begin as an agency but as an educational venture teaching people to run AdWords campaigns. The agency grew out of the teaching, because enough students came back asking whether the team would just run the campaigns for them.

Both halves still exist. Isaac leads education. Patrick joined in 2015, ran operations, and has been CEO since January 2024.

His own path runs straight through the subject. He began writing about the inevitability of AI in 2017. Google contacted him about it in 2018, which led to a speaking tour and to his recorded content being used in Google's official Google Ads certification, in the section on AI-driven campaigns.

Why He Says the Two Books Do Not Disagree

Asked which Patrick was wrong, he rejected the framing. He does not read the two books as conflicting opinions so much as a record of how the market moved underneath them.

The first was aimed at a specific moment. Nobody in the PPC community wanted to hear about AI in 2020, and he took real criticism for it. People lined up on podcasts and YouTube channels to debate him, because the profession had already decided automation was the enemy.

Its origin was not strategic insight. It was fear. In 2015 his whole job was manual work inside campaigns: open the Google Ads multi-client center in the morning, close the tab ten hours later. When automated bidding arrived he was, by his own account, deeply insecure about it, and nobody in the industry was admitting to that out loud.

The industry had a ritual instead. Teams split-tested manual bidding against Target CPA, manual won, and the office celebrated. "We would pop champagne and say, we beat the machines again."

Then came the thought that produced the book: "What if the reason these Target CPA campaigns keep failing is because I actually don't know the first thing about how Target CPA works?"

So the second book is not a retraction. The first taught people how to use automation in a world where nobody knew how. The second asks the question that only makes sense once everyone has the technology in hand: what should strategy actually look like now?

From Blackjack to Poker: The Method Inside the New Title

The new title comes from poker, and it carries the argument.

Get dealt pocket aces in Texas Hold'em and you are holding the best starting hand in the game. Plenty of people still lose money with it. Plenty of others have made millions on 7-2 offsuit, the worst hand there is. The quality of the decision and the quality of the outcome are two different things.

Blackjack is the opposite, because the math has already been done and the correct play is knowable. That, he says, is where he spent his career: "Most of my career in marketing has been spent sitting at a blackjack table."

That era has closed. The table is now poker, where there is no single right answer and everything carries nuance and context.

The practical rule is not to abandon best practice but to hunt for the exception: "Assume that the best practices probably fit for you, and then look for the nuance."

He has been held to his own book title over exactly this. In 2022 his team moved a large brand to manual bidding and the CMO pushed back, pointing out that the automation guy had supposedly written that you should never use manual bidding. That was never the point of the first book. This particular account had no reliable conversion tracking, no meaningful competition, and a cost per click that had stopped correlating with conversion rate. Under those three conditions the bidding strategy that theory disqualifies was clearly the better one.

Why ROAS Is a Lagging Indicator, and the Only Lever It Leaves You

He opens with a client he has run for ten years. Their bottom line profitability tripled over twelve months. Over the same period their reported ROAS was the lowest it has ever been. The client is delighted.

The reason is arithmetic. ROAS is revenue over spend, and those are the only two terms in it. Sitting at the keyboard today, a marketer cannot move revenue directly. The one input actually within reach is spend.

So a team told to raise ROAS has exactly one mechanism, and it is not growth. It is cutting the budget.

Which gives the line the episode is worth an hour for: "If you're optimizing for return on ad spend and you're holding your marketing agency accountable to return on ad spend, and the only lever they can pull is by reducing spend, you're never gonna grow your business."

He is not banning the metric. As a guardrail on an individual campaign it earns its place. It just cannot be the only number an account is judged on, and a ROAS benchmark read in isolation says almost nothing about whether a business is growing.

His team works instead from the four numbers ROAS is made of: total website traffic, the cost to acquire it, conversion rate, and average order value. A ROAS drop starts a diagnosis rather than ending one. And the language he wants marketers to learn is finance. His instruction is to go to CFO school, understand how a P&L behaves, and recognise that the person holding the budget cares about top line revenue and bottom line profit, not a ratio.

The alternative, in his words, is the old lazy way of doing it, and a race to the bottom.

The Reddest Ocean: If Your Job Title Explains Your Job

Asked whether channel specialists should move closer to the business, he answers with Blue Ocean Strategy. Red oceans are saturated markets, red because the sharks have eaten all the fish.

"I look at the PPC space, performance marketing, as the reddest ocean I've ever seen."

Every characteristic is present at once: heavy competition, services commoditized to the point where one agency can be swapped for another with no discernible difference, and client expectations about cost and scope that are fixed before the first conversation.

Then the test, aimed at the individual: "If you can tell me what your job title is and I can already explain back to you what you probably do all day long, that means you're probably in a red ocean."

He does not call it unsurvivable, only expensive. Sustainable profit and growth in that water requires being in the top 5% or better. If you genuinely are top 1%, stay and be the best at it. If you suspect you are merely above average, he thinks you should be planning what comes next.

His blue ocean example is deliberately unflattering and entirely fair: the fractional Chief AI Officer. He cannot tell you their qualifications, what they do all day, or what they charge. Not being able to answer any of those questions is precisely what makes it blue ocean.

He is applying this to his own company, which is moving out of traditional paid media on the belief that that ocean will keep shrinking.

Why He Will Not Put a Dollar Into ChatGPT Ads Yet

His position is blunt. There is enormous noise about ChatGPT ads, and the ads are not good yet.

The example is from his own team. A colleague researching answer engine optimization for a client in the auto repair business ran a query about auto repair and was served a ChatGPT ad for roof repair. One word matched, and it was the wrong one.

"Guys, this is the stuff that people hated on Google about for the last 20 years. And it's query matching."

His point is about accumulated investment rather than OpenAI's ambition. "You know what's still the best bidding platform on the planet? Google Ads." Fifteen to twenty years of engineering has gone into that auction, and that is not a gap a new entrant closes quickly. It is also why the Google versus Meta question still has a defensible answer while the ChatGPT one does not.

He will not concede the first-mover case either. Across his entire portfolio, large brands and small, he cannot name one whose next incremental dollar is best spent on ChatGPT ads. Put it into YouTube for the next six months instead. And on the risk of being late: "I will sit here on the sidelines and I will wait for you to report back on it."

Build or Buy: The Line Is Builders Versus Practitioners

The build-versus-buy question is not really about tools. "The real line in the sand is to what extent will your team be builders as opposed to practitioners."

AdVenture drew that line two years ago. Everyone on the team has to be proficient with Claude Code and capable of building their own solutions, and he is direct that the agency may not be the right place for people who will not.

Once that is settled the answer follows. A team of practitioners keeps buying SaaS licences, because that is what practitioners need.

For builders, he starts with subtraction. A $20 subscription here, a $40 one there, a $100 one somewhere else, and then the realisation that a weekend in Claude Code replaces the tool that automates getting access to client accounts. Project management, proposal software and analytics tools have all gone the same way.

HubSpot has not, and the reasoning is the useful half. It is the system by which prospects become paying clients, and the risk of breaking it outweighs the saving. Building with AI is genuinely accessible now, but maintaining software remains a real cost, and for a revenue-critical system the licence is "essentially an insurance policy."

Key Takeaways from the Episode

• The two books are a sequence, not a contradiction. Join or Die taught marketers to use automation when nobody knew how. Never Always, Never Never asks what strategy looks like now everyone has it.
• ROAS gives you one lever, and it is spend. Holding an agency to ROAS asks it to shrink your account. Diagnose with traffic, cost per session, conversion rate and average order value instead.
• Reported ROAS and profit can move in opposite directions. His ten-year client tripled bottom line profitability while reporting its lowest ROAS ever.
• The attention arbitrage is closed. The 2011 gap between where consumers spent time and advertisers spent money was a ten-year run. Distribution, pricing and creative are what is left.
• If your job title describes your job, you are a commodity. Surviving the red ocean means the top 1 to 5%. Everyone else should be planning a way out.
• ChatGPT ads are not ready, and query matching is why. A query about auto repair returned a roofer. The same dollar performs better on YouTube over the next six months.
• Build or buy is decided by whether your team are builders. Cut the small subscriptions first, keep paying for anything whose failure would stop revenue.