AI-current PPC agency: 5 questions to ask

Looking for a PPC agency, or auditing the one you already work with? The question is harder to answer than it looks. In the last 24 months, Google has shipped 14 major AI-based updates inside Google Ads. Meta has rebuilt its targeting algorithm twice. LinkedIn launched Predictive Audiences. Microsoft integrated ChatGPT directly into Bing Ads. The monthly reports your agency sends, however, look exactly the same as in 2022. So you have no way to tell whether your agency noticed.

A performance marketing agency current with AI changes and one running its 2022 playbook deliver similar results in the first month. The gap opens in month six, when the unadapted agency keeps holding a cost-per-acquisition 20-40% above market. You don't notice, because you don't see the alternative.

This article walks through what concretely changed inside Google Ads, Meta Ads and LinkedIn Ads between 2024 and 2026, how it shows up in your cost per acquisition, and five questions worth asking at your next agency meeting.

What changed in paid media between 2024 and 2026

The shift is mechanical, not communicational. Three levers moved from the media buyer's hands into the algorithm's, and each one has consequences on your bottom line.

1. Keywords are no longer units of control

Until 2023, the keyword was the atomic unit of a Search campaign. You decided what to bid on and what match type to use. The volume of phrase match your agency built was a direct proxy for the quality of their work.

Google introduced search themes inside AI Max in 2024. A year later, Performance Max got more control over asset groups. By 2026, broad match paired with Smart Bidding delivers lower CPCs than exact match in 60% of e-commerce accounts, for the first time.

An agency still optimising at keyword level in 2026 is not technically wrong. It is, however, blocking your access to 30-50% of the available volume at lower cost. That efficiency gain goes to whichever competitor made the transition.

2. Ads are written once; AI rewrites them daily

Asset Studio (Google), Advantage+ Creative (Meta) and Dynamic Asset Generation on LinkedIn have moved ad writing from a monthly delivery of 10-20 hand-written variants to continuous generation from a smaller pool of source assets. The algorithm produces hundreds of combinations every day.

An agency still delivering ten new ads per month is not wrong, but its delivery has become insufficient. The job now is writing 20-30 source assets per campaign, defining the generation rules, monitoring performance at combination level, and systematically removing the underperformers.

3. Attribution moved into AI's hands and became opaque

Data-driven attribution has been the default model since 2023. Two years on, the model uses AI signals to redistribute credit across touch-points in ways you can no longer audit at rule level. The last-click and first-click models you used to justify budget decisions are no longer viable options inside Google Ads for most accounts.

If your agency still sends last-click conversion reports in 2026, either they are not using modern Google data, or they are filtering it through old metrics because that is what you are used to. Both versions cost money: budget decisions taken on an outdated model allocate spend wrong.

The hidden cost of an agency that did not adapt

None of the shifts above feel sudden. They accumulate slowly. Month after month, your customer acquisition cost rises 2-3%, attributed to competition, seasonality, or the market. By month 12, your customer acquisition cost is 25-35% higher than it would be on an account configured for 2026 methods. Three concrete manifestations of that hidden cost:

Budget allocated where AI no longer works on the old model. A Search campaign structured on 200 exact-match keywords, when Performance Max could deliver the same volume at 35% lower cost, means money spent maintaining the old structure. The efficiency gain stays on the table.

Ad assets written rarely, therefore under-used by the algorithm. Meta and Google rotate between assets to find the optimal combination. Five assets per campaign produce a mediocre optimum. Twenty-five produce a better one and calibrate faster.

Scaling decisions taken on incomplete data. You want to raise budget 50%, but the agency without AI modelling cannot predict how your customer acquisition cost will respond. You raise the budget, the cost per customer explodes, you stop the increase, you blame the campaign. The reality is the agency simply did not have the data to anticipate the inflection.

Five questions for your next meeting with the agency

None of them require technical expertise. All of them require concrete answers.

1. How many of my campaigns are running with AI Max for Search enabled, and since when? You want a number followed by a date. If the agency says they avoid AI Max because it is too aggressive, ask for the data behind that decision for your specific account, not a general opinion.

2. How many source assets do we have active on the main campaign? The minimum threshold for Google Search starts at 15 headlines, 4 descriptions and 5 images per ad group. For Meta, 20 or more creative assets per campaign. Below those, the AI has nothing meaningful to compose with.

3. What attribution model do we use, and how does it affect cross-channel budget decisions? The answer should mention data-driven attribution and explain how it redistributes credit. If the agency replies that they use last-click for simplicity, ask which budget decision was taken last week based on that model.

4. How many hours per month does the agency spend specifically on my account, and on what activities? A modern agency spends less time on routine execution (Smart Bidding handles that) and more on strategy, asset writing and data auditing. If the ratio is inverted, you are paying for work the algorithm does better for free.

5. What was the last major platform update you implemented on my account, when, and what was its impact? A concrete answer names a specific update (search themes, Advantage+, brand controls), a date, and an impact number. A vague answer means the agency is not tracking the changes or not applying them to your account.

What an AI-current agency actually looks like

It is not a certificate, and not a blog full of AI-titled posts. It is an operational practice with six verifiable components.

Monthly internal updates on platform changes, with written agenda and notes. Not just reading the Google newsletter.

Active tests on at least 10% of each client's budget at any time. Without tests, the agency only learns from platform announcements, never from your actual customer behaviour.

Asset-writing capability at volume. The team writes 20-30 original headlines per major campaign, not five.

Monthly audit of the attribution model and of tracking discrepancies (GA4, server-side, platform reporting). Differences above 15% are investigated, not accepted.

Reporting that links the platform to profit. Return on ad spend shown alongside real per-sale margin, customer acquisition cost alongside projected lifetime value. Budget decisions are not taken on platform metrics alone.

Time transparency. You see what was worked on last week. The agency does not hide behind account management language.

How to verify in 30 minutes

A half-hour meeting. The five questions above. And one final request: the last three months of reports for a client similar to you, anonymised.

If the agency refuses on confidentiality grounds, ask for the report structure rather than the numbers. A modern agency uses a standardised format that links platform activity to commercial impact. An agency that ships monthly Google Ads screenshots is still on a 2018 model.

You have this conversation before signing a new contract, or at any renegotiation point of the existing one.

In paid media, the real cost of an agency that did not adapt does not appear on its invoice. It appears on your Google invoice.