AI for PPC Managers: What Actually Works
The platforms automated the levers PPC managers were hired to pull. What the job is now: inputs, audits and the AI work beyond auto-bidding.
PPC was the first marketing job the platforms automated from the inside. Bids went first - Google’s own documentation notes that more than 80% of its advertisers were already using automated bidding by 2021 - then targeting, then ad assembly. So you might expect the people running campaigns to describe an easier job. The opposite: in the State of PPC survey of 1,151 practitioners, 52% said managing campaigns is harder than two years ago, against 13% who found it easier, with lost control and black-box automation among the named reasons.
I don’t read that 52% as nostalgia. I read it as an accurate report that the job changed shape without changing title. The levers a PPC manager was hired to pull - bids, match types, placements - are increasingly pulled by the platform, and the work that remains is less visible and harder to demonstrate: deciding what the machines optimize toward, feeding them inputs worth optimizing, and auditing results the seller of the ads has graded for itself. This guide is my map of that job, written for the person who owns the ad account, whether the business card says PPC manager, performance marketer or media buyer.
The job now: inputs and audits, not levers
My one-sentence version of the role in 2026: you no longer operate the machine, you govern it. Concretely, the platform now owns auction-time bidding, query matching and creative assembly. What it cannot own, because they require knowing your business:
- What counts as a conversion, and what it is worth. The bidder optimizes exactly what you feed it. Conversion definitions, values and margins are the highest-leverage settings in the account, and they are business decisions wearing a settings menu.
- The constraints. Brand exclusions, negative keywords, placement rules, the claims allowed into ad copy. Automation without constraints is the platform spending your budget on its own judgment - the reason I set Performance Max guardrails before launch, not after the first weird placement report. The quarterly version of the same duty is checking who the delivery actually reached, because engagement-optimized delivery skews by demographics on its own.
- The verdict. Whether any of it worked, measured somewhere the platform doesn’t control. This is the part of the job that grew the most, and I’ll come back to it, because it is also the part that makes you hardest to replace.
The task split I’d run
The same hand-over, share, hold split I use in every role guide, applied to the ad account.
Hand over: bidding. Fighting auction-time bidding manually is a solved argument for most accounts. The real work moved one level up - choosing which strategy fits your conversion volume and value data, which is a decision I’ve mapped in full, and then protecting the learning period from your own itchy fingers.
Hand over: variant production. Fifteen distinct headlines per responsive search ad, sized and reworded across ad groups, is exactly the work generative models do well. My ad copy prompts carry the character limits and a compliance fence. What does not get handed over is the claim list the variants draw from - every promise in the pool is one you have to be able to substantiate. The demand side of variant production is creative fatigue: the detection signals there set the pace this pipeline has to keep.
Share: query and search-term mining. A model is a fast first-pass reader of search terms reports: clustering queries by intent, flagging the ones that smell like a negative. I treat its output as candidate lists, never as applied changes, because the model does not know that a weird-looking query is your best customer’s vocabulary.
Share: reporting. Numbers in, narrative out. A model drafts the client-facing story from figures you supply, and the standing rule from my marketing manager guide applies unchanged: the moment a model is allowed to fill a data gap with a plausible number, the report stops being a report.
Hold: measurement independence. The platform reports its own conversions, models its own attribution and recommends its own products. None of that is scandal, all of it is structure: the grader and the graded are the same party. Your counterweights are an attribution lens you understand - I’ve written about what data-driven attribution can and can’t tell you - and, for decisions that matter, a holdout or incrementality test the platform has no hand in.
Hold: the budget argument. How much to spend, where, against which margin math. Every automated recommendation in the interface has a seller’s interest attached, and the recommendations tab is not a strategy. The quarterly homework behind the argument is the wasted spend audit - it is hard to lose a budget conversation holding the list of leaks you just closed.
Reading platform claims like a professional
Here is a working example of the audit mindset, using a current one. When Google launched AI Max for Search campaigns, its announcement claimed advertisers typically see 14% more conversions or conversion value at similar CPA or ROAS, rising to 27% for campaigns still mostly on exact and phrase match.
How I read a claim like that, in three moves. First, the source: the seller, measuring its own product, on internal data - which does not make it false, it makes it unverified in your account. Second, the mechanism: expansion features find conversions partly by loosening your targeting, so “more conversions at similar CPA” is plausible while the kind of query you win changes underneath - the platform-by-platform version of that loosening is in the lookalike audiences guide. Third, the test: if I want to know, I run it as an experiment with a holdout, judged on conversion values I trust, over a window agreed before launch. Adopting platform AI features is fine. Adopting the vendor’s measurement of them is the actual mistake.
That three-move habit, applied to every shiny announcement, is worth more than any individual tool opinion I could give you, because the announcements will keep coming and the opinions will keep expiring.
The skills that appreciate
If the survey’s 52% are right that the job got harder, the compensation is that the hard parts are the durable parts. What I would deliberately get better at this year:
- Writing constraints. Negative keyword architecture, brand safety rules, claim libraries for ad copy. The account is increasingly a policy you author and the platform executes.
- Test design. Experiments, holdouts, pre-agreed evaluation windows. The practitioner who can say “here is what the automation actually added, measured independently” owns the room - and the budget conversation.
- Value plumbing. Getting real margins and lead scores into conversion values. Value-based bidding is only as smart as the values, and most accounts still feed it revenue or placeholders.
- Explanation. Automated accounts fail quietly and win opaquely. Translating “the black box did something” into “here is what changed, why, and what we’re doing” is the client-facing skill the tools conspicuously lack. The black-box frustration in that survey is also your differentiation.
What stays human
The pattern across all these role guides holds here. Judgment about what the business needs stays human: the machine can find people likely to convert, and has no opinion on whether those conversions build the business you want. Accountability stays human: when spend is wasted, “the algorithm did it” is an explanation, not a defense. And skepticism stays human, professionally so - the platforms are counterparties as well as partners, and in my view the PPC manager is now the person on the marketing team whose job description includes remembering that.
FAQ
Will AI replace PPC managers? It already replaced the version of the job that was bid changes and match-type housekeeping. What it has not touched: deciding what to optimize for, constraining the automation, and proving results independently - the survey’s own respondents say the job got harder, not emptier. The at-risk profile is the manager whose whole value was interface operations. The appreciating profile is the one who can design a test and argue with a platform’s homework.
Which AI should I add first? Nothing new - first, govern what’s already running. Audit conversion definitions and values, since every automated system downstream inherits them. Then a generative model for ad variant production with a fixed claim list. Then reporting drafts. Third-party AI optimization layers come last, if ever, and only with their ROI counted honestly.
How do I show my value when the platform “does everything”? Measure the gap between governed and ungoverned automation, because that gap is you. Guardrails that prevented junk placements, negatives that cut wasted spend, an experiment that caught an underperforming recommendation, values that redirected the bidder toward margin. Keep receipts - a before-and-after log of interventions is the most persuasive self-review a PPC manager can bring.
Is manual bidding worth learning at all now? As a daily practice, no. As literacy, yes - understanding what the auction is doing is what lets you diagnose the automation instead of just believing it. It is the same reason I still teach the attribution models even though a machine picks the weights: you cannot audit what you cannot explain.
This post is one of the role-by-role playbooks - the seat one level up, running paid as one experiment loop among several, is the growth marketer’s. The tactical layer lives in the paid media guides - Smart Bidding, responsive search ads and Performance Max guardrails - and the wider map is the complete guide to AI marketing.