I write a blog full of articles about how AI saves me hours of work. Automation, MCP, generating campaigns from a prompt. So why am I writing an article about how AI won’t replace PPC specialists? Isn’t that a contradiction?
It’s not. And it’s perhaps the most important thing I know about AI in advertising: AI is an incredibly powerful tool today — and that’s exactly why it’s dangerous to put it in the hands of someone who can’t tell when AI is talking nonsense. Because it dishes out bad advice confidently and eloquently, just as it delivers those brilliant insights.
We hear it from all sides: AI can create campaigns, optimize, write ads, generate creatives, and design account structures. And it’s true — it can, and often better than a junior. But it lacks one thing you can’t make up for with a prompt: years of experience in real markets. I’ve put together why this creates two problems for advertisers, where exactly AI advice will tank campaign performance, and why PPC optimization will never have a single universal playbook.
Table of Contents
What AI can actually do in PPC today (and it’s quite a lot)
I’ll start fair, so there’s no doubt this isn’t an article by some bitter skeptic. AI can do impressive things in PPC today:
- It builds campaign structures from a brief — ad groups, keywords, RSA variants, sitelinks, callouts. In minutes. (I do this myself by generating a PMax import for Editor.)
- It writes dozens of ads in different tones, with variations of USPs and CTAs.
- It generates creatives — banners, visuals for Meta, product photos, even videos.
- It finds patterns in data — search terms with high spend and zero conversions, underperforming ad groups, anomalies in daily spend.
- It can explain anything to you — what tCPA is, how Smart Bidding works, why your Quality Score dropped.
- It prepares reports in a fraction of the time.
I actually use all of this every day. Today, AI outperforms a junior PPC specialist in many individual tasks — it’s faster, doesn’t forget, doesn’t make typos, and knows the documentation by heart. If I were to compare “AI vs. a junior with six months of experience” in terms of speed and breadth of knowledge, AI wins.
And this is exactly where it starts to get tricky.
Two challenges advertisers face
When a company hears “AI can handle PPC”, they usually imagine that you just need to let the AI loose and the campaigns will optimize themselves. Reality hits two major walls.
Problem #1 — AI lacks market experience
AI can do everything that’s described, documented, or learnable from text. But PPC mostly happens in the space that isn’t documented — in the specific behavior of a particular market, season, competition, and audience.
AI doesn’t know that:
- In your industry, CPC starts climbing by 40% in mid-November because e-shops jump into the auction with Black Friday budgets — and that it pays off to gather data cheaply a few weeks before, not just during the rush.
- Your specific client has a product that involves a long decision-making process, so last-click attribution systematically undervalues the upper funnel stages, and AI will advise you to turn off exactly what triggers the decision because it’s mistakenly looking only at the last 30 days instead of a 90-day data window.
- A given target audience in a specific region reacts completely differently to discounts than to quality — and what worked for one client in the same industry will tank for another.
This knowledge isn’t in any dataset. It comes from years of observing dozens of accounts, comparisons, and mistakes that cost you something. AI doesn’t have it because it had nowhere to collect it from — and what’s worse, it won’t tell you “I don’t know this,” but will calmly give you a recommendation as if it did, because it pulls knowledge from the entire market rather than the client’s context.
Problem #2 — AI is only as good as the person driving it
This is the most overlooked problem of all. AI won’t reach its potential if it’s managed by someone who doesn’t understand PPC.
Prompting PPC campaigns isn’t about saying “write me a good campaign.” It’s about:
- Knowing what to ask in the first place—what questions to lead with, which metrics to track, and which hypotheses to test.
- Knowing how to set the context and skills—without the right input (target CPA, excluded networks, brand vs. non-brand segmentation, the client’s business model), AI will build you a generic textbook campaign that doesn’t reflect reality.
- Recognizing when the output is off. AI will confidently suggest a structure that looks textbook-perfect but makes zero sense for the given budget and market. If you don’t spot it, you’ll launch it.
A paradox emerges: AI in the hands of an inexperienced person either fails to tap into even a fraction of its potential or actively causes harm—because a layperson takes the first confident recommendation and runs with it. A junior at least suspects there’s something they don’t know. A layperson with AI feels like they know everything.
AI doesn't raise the quality of work. It amplifies the level of the person driving it. Senior input → senior output. Amateur input → amateur output, deployed with a false sense of confidence.
Where AI loves to tank your performance
These aren’t just theoretical worries. Here are specific recommendations AI commonly spits out—which, without an experienced filter, will tank your campaign performance:
- “Turn on Display Network, Search Partners, AI Max for more reach.” This is the Google Ads default, and AI will happily confirm it. For a Search campaign, this usually means a flood of cheap, non-converting impressions and clicks. “Conversions” will formally increase, but your economics will collapse. You’ll hear similar recommendations directly from Google support, and they aren’t good advice.
- “Set tCPA to 150 Kč to get cheap conversions.” When your real CPA is 400 Kč, this target will choke the campaign—Smart Bidding will stop bidding, the campaign will lose volume, and “learning limited” will trap it.
- “Add more keywords, you have low traffic.” When the real problem lies in the landing page or the offer, more keywords will only scale the waste, not increase conversions. AI lacks the context to evaluate the landing page.
- “Merge those campaigns, simplify the structure.” Sounds reasonable. Except you’ll wipe out the historical data that Smart Bidding relied on—and the campaign will spend weeks relearning from scratch.
- “Switch to broad match, Smart Bidding will handle it.” Without enough conversion data, broad match will burn through your budget before it has anything to learn from.
- “This keyword has a high CPA, turn it off.” But in a multi-touch journey, that’s the word that triggers the purchase, even if another channel formally gets the conversion. Turn it off, and your entire upper funnel will collapse.
Each of these recommendations sounds logical and would even be correct in a certain context. The problem is that AI doesn’t see that specific context—and anyone managing it without experience won’t provide that context.
Why there’s no such thing as a 100% foolproof scenario
Here’s the heart of the matter. People want a “manual that works” from PPC. They want AI to be the machine that generates that manual. But PPC optimization is largely about the little things and experimentation—and there’s no such thing as a universal, one-size-fits-all scenario.
What boosts ROAS by a third for one client might not work at all for another in the same industry. Same campaign structure, same bidding strategies, yet the opposite result—because the offer, margins, website, seasonality, audience, competitive pressure, product quality, and brand awareness are all different. There are dozens of variables, and they interact in ways you just can’t calculate in advance.
That’s why the real job of a PPC specialist isn’t “applying the right scenario.” It’s:
- Build a reasonable hypothesis (based on experience of what might work).
- Test it with a small budget.
- Measure, evaluate, adjust.
- Repeat—endlessly, because the market is constantly shifting under your feet.
This is an iterative craft built on judgment, not a deterministic algorithm. AI is a great assistant in every one of those steps—it suggests hypotheses, crunches the numbers, and finds patterns. But the decision of which hypothesis is worth testing and when to trust the result relies on experience that AI doesn’t have—and neither does a layperson operator.
By the way—the same “change and test step-by-step, not blindly” principle applies to website migrations and redesigns, where blind changes can do even more damage than in a single campaign.
How I work with AI
To be specific—AI in my practice isn’t an autopilot, it’s a co-pilot:
- AI suggests, I decide. Every recommendation goes through my filter: “Does this make sense for this specific client and this specific period?” I discard or tweak a large portion of the suggestions.
- AI handles the routine, I provide the judgment. Data collection, initial structure drafts, reporting—AI. Deciding what to do with it, which hypothesis to test, when to intervene—me.
- No changes to the account without a review. AI prepares a draft, I approve or rewrite it, and only then is it applied. (I wrote about this in the article about the MCP stack—my entire setup is built on the “show suggestion → I approve → apply” principle.)
- I provide the context. The client’s business model, margins, seasonal patterns, account history, what worked and what didn’t. That’s the fuel; without it, AI is just running on empty.
The result: I work faster and get more done than before, but the quality of decisions holds up—because the decisions are still made by a human who has seen how similar situations played out.
Conclusion
AI in PPC is neither a threat nor a savior. It is an exceptionally powerful tool whose value depends on who is holding it. In the hands of an experienced PPC specialist, it’s a productivity multiplier. In the hands of someone who can’t tell when AI is talking nonsense, it’s a fast track to a burned budget—made even more dangerous by the fact that the nonsense is delivered with confidence and polished phrasing.
Three things to wrap up:
- AI is capable of incredible things—and it will also suggest things that will send your campaign into the red. The difference between the two is the experience of the person reading the output.
- There is no universal winning scenario. You have to test different approaches for every client. That’s craftsmanship, not an algorithm.
- The best today isn’t AI or a human, but a human with AI—provided that person knows the difference between good and bad advice.
Will AI handle your PPC, or do you need a specialist?
FAQ
- So, should I not use AI in PPC? On the contrary, use it. Just not as a replacement for experience, but as a multiplier of it. AI without experienced supervision is a risk; experience without AI is just slow.
- Will AI replace PPC specialists in a few years? It will replace those who don’t learn how to use it. An experienced specialist who integrates AI into their work will, on the other hand, be more productive and in higher demand. (I elaborated on this in the article about pricing in the AI era.)
- Can I build campaigns myself with AI and save on a specialist? You can build them, sure. The question is whether you’ll recognize that the AI suggested keeping Search Partners on, set a tCPA that’s too low, or proposed a structure that will wreck your historical data. The money you save on a specialist might come back to haunt you as a much larger amount in wasted budget.
- How do I know if AI or an “AI tool for PPC” is giving me bad advice? Short answer: without experience, you won’t know for sure. That’s the whole point of this article. Long answer: watch the economics (CPA, ROAS, net profit), not vanity metrics (“conversions” can be junk). And when a tool promises “universal optimization,” stay alert—it doesn’t exist.
- Why do you use AI so much when you’re writing about its limits? Because I know both sides—what it’s capable of and where it lies. That’s exactly why I can squeeze the most out of it and filter out its mistakes. That’s the difference between a tool in the hands of a craftsman and in the hands of a random passerby.