Read ten PPC agency websites this afternoon and nine of them will sell you bid optimization, keyword expansion and creative testing. Two of those three are now owned by Google, and the third is increasingly generated by it. The pitch has not caught up with the platform.
That gap matters because the fee model most agencies use was designed around the old division of labor. When a human set every bid and wrote every ad, charging a percentage of spend was a rough proxy for the work involved. Google has since automated the expensive part and kept the pricing convention, which leaves buyers paying a spend-linked fee for a job that no longer scales with spend.
We pulled commercial search demand for agency terms in Ahrefs on October 7, 2026. In the United States, ppc agency runs 20,000 searches a month at a difficulty score of 1, with an average paid cost per click of $20.00. Twenty dollars a click is what competitors are willing to pay to reach a buyer asking that question, which tells you how much budget moves on the answer. It is worth getting the answer right.
What does a PPC agency actually do in 2026?
A PPC agency plans, builds and manages paid advertising on auction platforms such as Google Ads and Microsoft Advertising, and in 2026 the substance of that work is account architecture, measurement quality and query control rather than manual bidding.
The one-sentence version matters because the category name is now misleading. Pay per click implies the agency manages clicks and their prices. It does not, in any direct sense. What it manages is the set of inputs Google's automation reads: which conversions you count and at what value, which queries you exclude, how campaigns are segmented so budgets cannot cannibalize each other, what the feed says, and where the click lands.
| LEVER | WHO CONTROLLED IT IN 2019 | WHO CONTROLS IT IN LATE 2026 |
|---|---|---|
| Bid per keyword | The advertiser, directly | Smart Bidding, with the advertiser setting only a target |
| Which queries you show on | The advertiser, through match types and negatives | Largely Google, with AI Max search term matching on by default since September 1, 2026 |
| Target enforcement | The advertiser, in both directions | Google, which corrects campaigns that beat the target upward but not those that miss it |
| Placement mix | The advertiser, by network | Mostly the campaign type, with Performance Max spanning inventory by design |
| Creative combinations | The advertiser, per ad | Asset generation and auto-created assets, unless explicitly switched off |
| Impression eligibility | Policy compliance alone | A qualification judgment under Limited Ad Serving, expanding to all Google Ads by 2028 |
| Account structure, negatives, conversion data, feeds | The advertiser | Still the advertiser, which is exactly where an agency now earns its fee |
Read the last row as the job description. Everything above it has been automated, defaulted or taken away. An agency that cannot describe its work in terms of that bottom row is selling you a 2019 service at 2026 prices.
Which levers has Google taken back?
Three changes in 2026 account for most of the shift: asymmetric target enforcement on budget-limited Smart Bidding campaigns, AI Max query matching enabled by default, and the Limited Ad Serving policy extending beyond Search to every Google Ads product.
The bidding change is the one to understand properly, because it is easy to misread as routine. Google disclosed it on June 15, 2026 and it took effect on August 17, with a help page edit reported in mid September confirming the rollout completed on August 27. It applies to budget-limited campaigns using Target CPA or Target ROAS. Google's own example: a campaign with a $10 target delivering a $5 actual cost per acquisition gets pulled up toward $10.
The asymmetry is the problem. On the Marketing O'Clock podcast on August 12, Greg Finn asked Google's Ginny Marvin whether a campaign with a $100 target converting at $125 would be corrected in the other direction. Her answer, reported directly, was that it would not. Overperformance gets spent back toward the target. Underperformance does not get reined in. Jarah Burke of Measured summarized the new regime as Google treating your target as an instruction rather than a ceiling.
AI Max compounds it. Google emailed advertisers on August 5 that Search campaigns using broad match or auto-created assets would move to AI Max on September 1, with search term matching on by default. We covered the mechanics of that migration when the notices went out, in the AI Max migration breakdown, and the independent testing since has not been kind. Xavier Mantica measured $100.37 per conversion on AI Max against $52.69 on exact match and $43.97 on phrase in his own account. Ezra Sackett found that 99 percent of AI Max impressions across roughly 30,000 search terms produced no conversions. Lunio reported invalid traffic in AI Max retail campaigns rising from 2.46 percent to 5.28 percent.
Cost per conversion by match type in Xavier Mantica's 2026 test, indexed so the highest value reads as 100. One advertiser's account, not a market benchmark.
Google's own claimed uplift for AI Max has moved around in public: 14 percent at the May 2025 launch, 7 percent by April 2026, and 15 percent cited by Philipp Schindler on July 22. A performance claim that drifts by a factor of two in fourteen months is not a number you can plan against. Meanwhile Mike Ryan published charts on September 14 showing cost per click rising for budget-limited campaigns while cheaper clicks flowed to uncapped ones, and Daniel Toledo of evoke wrote on October 10 that budget is the new quality score.
Why percentage of ad spend is now the wrong way to pay a PPC agency
A percentage of ad spend fee pays an agency more when your budget rises, at a moment when Google's own defaults already push budgets up, which means the two parties with influence over your spend both benefit from it growing and neither is paid for restraint.
That was always the theoretical objection to the model. It used to be weak, because scaling an account genuinely took more human hours. Target enforcement that spends overperformance back toward the target removed the excuse. When the platform will quietly consume a surplus rather than bank it, the only party left with an incentive to question the next dollar is you, and you are paying two parties who do not share it.
| FEE MODEL | PUBLISHED 2026 BAND | WHAT IT REWARDS | VERDICT |
|---|---|---|---|
| Percentage of ad spend | 10 to 20 percent of monthly spend, with minimum fees commonly $1,000 to $1,500 | Spending more, whether or not the next dollar is incremental | Avoid in 2026, for the reason this section exists |
| Flat monthly retainer | Roughly $1,500 to $10,000 per month on published rate cards | Predictable scope and attention that does not track your budget | Workable, if the scope is written down in real detail |
| Base plus percentage above a threshold | A fixed base, then a single-digit percentage above an agreed spend level | A viable floor on small accounts without full exposure on large ones | The best of the common structures above roughly $50,000 a month |
| Hourly or project | Clutch 2026 data puts US hourly work at $100 to $149 and projects at $10,000 to $49,999 | Defined deliverables, with no reason to keep a dead account alive | Good for audits and rebuilds, poor for ongoing management |
| Performance based | Priced at a premium where it is offered at all | Whatever the contract measures, which is almost never incrementality | Treat as a sales device until the measurement is agreed first |
One caveat on the numbers in that table. They are the bands agencies publish on their own pricing pages, plus Clutch's reported ranges, not an independent survey of what buyers actually pay. Published rate cards skew toward the middle of a market and omit the enterprise end entirely. Use them as a sanity check on a quote, not as a negotiating position.
“The fee question is really a question about who is paid to say no to the next dollar. On most paid search contracts written this year, nobody is.”
What should you hold a PPC agency accountable for?
Hold a PPC agency accountable for the inputs it still controls: account and campaign structure, negative keyword discipline, conversion tracking accuracy and value assignment, feed quality, landing page performance, and a measurement approach that separates incremental revenue from captured demand.
Those six are the entire remaining surface area. Notice what is absent: cost per click, impression share, and in many campaign types the match between query and ad. Writing a contract around metrics the platform now sets is how buyers end up unable to tell a good agency from a lucky one.
How do you tell a real PPC agency from a reseller?
The reliable test is to ask process questions with verifiable answers rather than strategy questions with impressive ones, because a reseller can describe a strategy perfectly well and cannot produce last week's search term exclusions or explain your conversion value model.
A note on the fifth play. Agencies are not villains for having inherited a fee convention, and most will have the structure conversation honestly if you raise it with numbers. The ones that will not have told you something useful.
When does in-house beat a PPC agency?
In-house wins when paid search is a permanent, large and strategically central channel with enough budget to justify a dedicated specialist, and an agency wins when the account needs senior judgment intermittently rather than a full-time operator.
The automation shift has pushed this line in an unintuitive direction. Because the platform now does the repetitive work, the remaining tasks are lumpy and senior: a structural rebuild, a measurement design, a response to a default change. That pattern suits external expertise better than a junior full-time hire, which is the opposite of the conventional wisdom that automation makes agencies redundant.
| SITUATION | BETTER IN-HOUSE | BETTER WITH AN AGENCY |
|---|---|---|
| Spend above roughly $150,000 a month, stable | Yes, with agency support for measurement design | Only if the internal hire cannot be made |
| Spend under $30,000 a month | Rarely, the role is not a full job | Yes, on a flat retainer with a named owner |
| Paid search is seasonal or campaign-driven | No, the capacity sits idle | Yes, project structured around the peaks |
| You need paid, SEO and AI visibility to work as one program | Only with a channel-agnostic team lead | Yes, where one partner holds the whole search surface |
| Highly regulated category with claim review | Yes, approvals live internally | Workable if the agency builds to your review process |
| No internal analytics capability | No, the measurement gap will defeat you | Yes, and make measurement the first deliverable |
How does paid search fit with SEO and AI answers now?
Paid search and organic visibility now share the same result pages and increasingly the same answer surfaces, so the useful unit of planning is the whole query, not the channel, and the two budgets should be set against one view of where a buyer is reached.
That is not a tidy synergy argument. The channels genuinely contradict each other at times, and we wrote about the sharpest version of that when paid and organic data disagreed inside AI Overviews. The practical point is narrower: if an AI answer already names you for a query, paying $20 a click to appear beside it may be redundant, and if it names a competitor instead, the paid click may be the only route you have while the organic work compounds. Neither call can be made from inside the ad account.
This is where the fee model and the strategy question meet. An agency paid on spend has no reason to tell you a query is better served organically. One paid a flat fee can afford to say it. We lay out the mechanics of running both together in our breakdown of paid and organic integration, and the reporting problem it creates is the subject of our work on measuring AI referral traffic against paid.
One more consequence of the automation shift deserves naming. If Google decides impression eligibility under a qualification policy, paid reach is no longer purely a budget question. The Limited Ad Serving expansion, reported in August and rolling out through 2028, limits impressions for advertisers it judges unqualified rather than disapproving their ads, and weighs signals including policy compliance, advertiser verification and persistent user reports. A channel whose ceiling is set by a trust judgment starts to resemble organic search, and should be planned like it.
What to do next
Start with your contract rather than your account, because the fee structure determines which advice you will be given for the next twelve months and no amount of optimization work survives an incentive pointed the wrong way.
Pull your last three months of paid data and answer two questions. First, did spend rise faster than incremental revenue after August 17? Second, what did your agency recommend in that window, and would a flat-fee agency have recommended the same thing? If the second answer is uncomfortable, the structure is the finding, not the performance.
Then rewrite the accountabilities. Six inputs, a named owner for each, a weekly cadence on search terms, and one success metric that does not come from Google's attribution. That is a short document and it is worth more than a quarterly deck. Our paid capture and Google Ads work is built around exactly those inputs, and the reporting side of it sits with our analytics and dashboard practice rather than inside the ad platform, for the reason this piece has been making throughout.
See where you are cited today
A free snapshot audit of your rankings and AI citations before we ever talk.
Tyler leads work at the intersection of SEO and generative engines at Something Inc., helping B2B brands get ranked and cited across every major AI engine.