AI and Paid Search: Who Is in Control?
AI and paid search have become inseparable. Google’s automated bidding, broad match expansion, Performance Max campaigns, and AI-generated ad copy have fundamentally changed who makes decisions inside a paid search account. The question is not whether AI is useful. It clearly is. The question is whether marketers understand what they have given up in exchange for that utility.
The short answer: more than most realise, and less than Google would like you to believe.
Key Takeaways
- Google’s AI optimises for its own auction efficiency first. Your business objectives come second unless you configure campaigns with that tension in mind.
- Broad match and Performance Max expand reach aggressively. Without strong negative keyword management and audience signals, that reach is often wasted spend.
- Smart bidding works best when it has enough conversion data to learn from. Below roughly 30-50 conversions per month, the model is guessing more than it is optimising.
- AI removes tactical execution from the marketer’s hands but does not replace strategic thinking. The marketers who understand campaign architecture, audience logic, and business context will outperform those who just press “go”.
- The paid search channel still works. The skill required to run it well has shifted, not disappeared.
In This Article
I ran my first paid search campaign in the early 2000s when the mechanics were almost laughably simple by today’s standards. You chose your keywords, wrote your ads, set your bids manually, and watched what happened. When I was at lastminute.com, we launched a campaign for a music festival and generated six figures of revenue within roughly a day. The campaign was not sophisticated. It was well-targeted, well-timed, and pointed at people who were already looking to buy. That simplicity made it easy to understand what was working and why. That kind of clarity is much harder to find now.
What Has Actually Changed in Paid Search
Paid search has been evolving for a long time. The changes Google has made to AdWords over the years have consistently moved the platform toward automation and away from granular manual control. Each update was sold as a performance improvement. Many of them were. But the cumulative effect is a platform where the advertiser’s role has fundamentally shifted.
The clearest example is keyword matching. Exact match used to mean exact match. Now it means “close variants,” which in practice can include searches that are conceptually related but commercially quite different. Broad match has expanded further still, and Google’s AI decides what searches your ads appear against based on signals the advertiser cannot fully see or override.
Smart bidding is the other major shift. Target CPA, Target ROAS, Maximise Conversions, and Maximise Conversion Value all hand bid decisions to Google’s machine learning models. The models are genuinely impressive when they have enough data to work with. When they do not, they make expensive mistakes and the feedback loop is slow enough that you may not notice for weeks.
Performance Max is the logical endpoint of this trajectory. It is a single campaign type that runs across all of Google’s inventory, uses AI to determine ad formats, placements, and bids, and gives the advertiser a set of asset inputs and audience signals rather than direct control. The reporting is limited by design. You cannot see which placements are driving results with the same granularity you had in traditional search campaigns.
If you want a broader perspective on how paid advertising channels are evolving, the Paid Advertising hub at The Marketing Juice covers the strategic questions that sit above any single platform or campaign type.
Where AI Genuinely Helps
It would be dishonest to frame AI in paid search as purely a problem. There are areas where it creates real value, and marketers who dismiss it wholesale tend to be the same people who are still manually adjusting bids at midnight and wondering why their CPAs are inconsistent.
Bid optimisation at scale is the clearest win. A human cannot process the volume of real-time signals that Google’s auction system handles: device, location, time of day, audience membership, search history, and dozens of other factors that influence the probability of conversion. A well-configured smart bidding strategy, with a clear conversion action and enough historical data, will outperform manual bidding in most mature accounts. That is not a controversial claim. It is what the data consistently shows when the conditions are right.
Responsive Search Ads are another area where AI adds genuine value. Testing ad copy combinations manually is slow and statistically unreliable unless you have very high impression volumes. RSAs run combinations automatically and weight toward the ones that perform. The marketer’s job becomes writing strong, varied assets rather than managing a testing matrix. That is a reasonable trade.
Audience expansion and in-market segments have also improved. Google’s understanding of user intent signals has become more sophisticated, and for campaigns targeting broad commercial categories, letting the AI find users who are likely to convert can work well. The conversion rate dynamics between paid and organic search reflect how intent-driven this channel is, and AI is better than humans at identifying that intent at scale.
Where AI Creates Problems Marketers Are Not Talking About Enough
The problems are structural, not incidental. They are baked into how Google has designed these systems, and understanding them is not optional if you are managing meaningful ad spend.
The first problem is the data threshold issue. Smart bidding models need conversion volume to optimise effectively. Below a certain threshold, roughly 30 to 50 conversions per month at the campaign level, the model does not have enough signal to make reliable decisions. It will still spend your budget. It will just do so less intelligently than you might assume. I have seen accounts where advertisers switched to Target CPA on low-volume campaigns and watched CPAs double within a month because the model was essentially operating on noise.
The second problem is the misalignment between Google’s objectives and yours. Google’s AI is optimising for conversions within its auction. Your business objective might be profitable revenue, not just conversion volume. If your conversion tracking is measuring form fills rather than qualified leads, or purchases rather than profitable purchases, the AI will optimise for the metric you gave it, not the business outcome you actually want. This is not a flaw in the AI. It is a flaw in how most accounts are set up.
The third problem is reduced transparency. When I was growing iProspect from 20 to 100 people and building out our paid search capability, one of the things we could do was show clients exactly where their money was going and why. Keyword-level data, placement-level data, device-level bid adjustments. That transparency built trust and allowed for intelligent optimisation conversations. Performance Max, in particular, makes that kind of conversation much harder. You are managing inputs and hoping the outputs make sense, with limited ability to interrogate the middle.
The fourth problem is what I would call automation dependency. When platforms do the tactical work, the in-house skill to do it manually atrophies. If Google changes its algorithm, deprecates a campaign type, or your account loses its data history, you need people who understand the fundamentals well enough to adapt. I have seen teams that could not explain why their campaigns were structured the way they were because everything had been set up by a platform’s automated recommendations. That is a fragile position.
The Specific Risks of Performance Max
Performance Max deserves its own section because it is where the tension between AI capability and advertiser control is most acute.
PMax is not a search campaign with extra features. It is a fundamentally different product. It runs across Search, Shopping, Display, YouTube, Gmail, and Maps. It uses your asset groups and audience signals to determine where and how to show ads. The algorithm decides everything else.
The reporting problem is significant. You cannot see a breakdown of spend by channel within PMax. You cannot see which search terms triggered your ads with the same granularity as a standard search campaign. Google has added some search term insights over time, but they are aggregated and incomplete. For advertisers who need to understand what is driving results, this is a genuine limitation.
There is also a cannibalisation risk that is underappreciated. PMax campaigns can compete with your existing branded search campaigns and Shopping campaigns. Google’s guidance is that PMax takes priority in most auction scenarios, which means it can absorb credit for conversions that would have happened through your more controlled campaigns anyway. Attribution becomes murky, and incremental value becomes hard to measure.
None of this means PMax should not be used. For e-commerce advertisers with large product catalogues and strong conversion data, it can perform well. The point is that it requires a different kind of strategic thinking. You are managing signals, not settings.
What Marketers Should Actually Be Doing
The marketers who get the most out of AI-driven paid search are not the ones who trust the platform most. They are the ones who understand the system well enough to configure it intelligently and challenge it when it is underperforming.
Conversion tracking is the foundation. If your conversion actions are not accurately reflecting business value, everything built on top of them is compromised. This means tracking the right events, assigning realistic values where possible, and being honest about what the data actually represents. The strategic philosophy behind paid search has always depended on understanding what you are actually measuring.
Audience signals matter more than they used to. In a world where keyword control has loosened, the audience signals you provide to Google’s AI become one of the few levers you can pull. First-party data, customer match lists, and well-defined remarketing audiences help the algorithm find the right people rather than just finding people who convert on paper.
Negative keywords are more important than ever, not less. The instinct when adopting automation is to step back and let the system run. The opposite is true for negatives. As broad match and PMax expand the potential search space, tightly managed negative keyword lists are one of the primary ways to prevent budget waste. This is not glamorous work. It is essential work.
Campaign structure still matters. The temptation to consolidate everything into a single PMax campaign because Google recommends it should be resisted if your business has meaningfully different product lines, margin profiles, or audience types. The AI cannot differentiate between a high-margin product and a low-margin one unless you build that logic into your campaign structure and conversion values.
Finally, keep some controlled campaigns running alongside automation. Standard search campaigns with exact and phrase match keywords give you a reference point. They tell you what is actually happening in the auction for your core terms, and they give you data that is not filtered through an opaque algorithm. Google has been evolving its AdWords infrastructure for years, and the advertisers who have maintained strong fundamentals through each change have consistently fared better than those who chased each new automated feature.
The Broader Question About AI and Channel Expertise
There is a version of the AI conversation that I find frustrating because it conflates two different things: AI as a tool for improving execution, and AI as a replacement for strategic thinking. They are not the same, and treating them as equivalent leads to bad decisions.
When clients ask for innovation in paid search, which they do regularly, the question I always ask is: what problem are you trying to solve? If the answer is “we want to use AI because it sounds modern,” that is not a business problem. If the answer is “our CPAs have been rising for 18 months and we need to find more efficient inventory,” that is a business problem, and AI tools might genuinely help with it.
The same logic applies to how agencies and in-house teams approach automation. Adopting smart bidding because Google recommends it is not a strategy. Adopting smart bidding because your account has the conversion volume to support it, your conversion tracking is accurate, and you have a clear view of what good performance looks like: that is a strategy.
I judged the Effie Awards for several years. The campaigns that won were not the ones with the most sophisticated technology. They were the ones where every decision, including channel and format decisions, was traceable back to a clear business objective. That standard does not change because the platform has become more automated.
Paid search has always been one of the most commercially direct channels available. Spend in the channel has grown consistently because it captures intent at the moment it exists. AI has made the execution of that capture more sophisticated and in many ways more efficient. It has not changed the fundamental logic of why the channel works.
For more on how paid channels fit into a broader acquisition strategy, the Paid Advertising section at The Marketing Juice covers the decisions that sit above individual platform mechanics, including how to think about budget allocation, channel mix, and what paid media consistently gets wrong.
About the Author
Keith Lacy is a marketing strategist and former agency CEO with 20+ years of experience across agency leadership, performance marketing, and commercial strategy. He writes The Marketing Juice to cut through the noise and share what works.
