AI in Paid Media Optimization: What Your Team Still Controls

Paid media optimization used to mean a specialist in the account every day, raising and lowering bids, adding keywords, splitting audiences and pausing ads that lagged. In Google Ads and Meta, much of that happens inside the platform. Their machine learning sets a bid for each auction, decides who sees an ad, chooses placements and assembles creative from the pieces you supply.

The optimization work has not gone away. It has moved to the inputs the systems learn from: what you count as a conversion and what it is worth, the targets and budgets you set, the creative they can choose from, the limits they must respect and how long you let them run before you judge. This guide explains what the platforms optimize on their own, the five levers that remain with your team or agency, a rhythm for checking them, and the questions to ask whoever runs your accounts.

DisclosureinBeat, which is mentioned in this article, is part of Fieldtrip, a group of marketing agencies in which Jake Jorgovan is a partner.

What the platforms optimize on their own

Each platform's documentation divides the work between its system and the advertiser. The table summarizes the main automated features in Google Ads and Meta.

Image

Zoom

The approach described in this article, shown as a visual guide.

Smart Bidding (opens in a new tab) is Google's name for bid strategies "that use Google AI to optimize for conversions or conversion value in every auction". At auction time it weighs signals such as device, location, time of day, the actual search query and the remarketing lists a person belongs to, including signals and combinations that Google says are available only to Smart Bidding. It needs conversion tracking switched on, and Google makes the advertiser responsible for using it within its personalized advertising policies.

Performance Max (opens in a new tab) takes this further. Google's own summary is a useful division of labor: you define the conversion goals and values, supply the assets, share audience signals and set budgets and brand safety settings; Google bids, generates images and video from your assets and "funnels your budget toward the highest performing inventory and placements."

On Meta, a campaign shows as an Advantage+ sales, app or leads campaign (opens in a new tab) when three automated settings are on together: Advantage+ audience, Advantage+ campaign budget and Advantage+ placements. Meta describes the sales and app versions as updated, streamlined versions of Advantage+ shopping and app campaigns.

One distinction is worth making before the levers. Automated rules (opens in a new tab) in Meta Ads Manager, such as pausing an ad when it hits a metric you choose, are your own if-then instructions. They save time, but they do not learn. The systems above do, which is why what you feed them matters more than how often you intervene.

The five levers your team still controls

1. The conversion you count, and what it is worth

The bidding systems optimize toward the action you tell them to count. If that action is a form fill and only some form fills become customers, the system is built to find more people who fill in forms. So the first lever is the signal.

Two kinds of work improve it:

• Send the later outcome back. Google's offline conversion imports (opens in a new tab) let you store the click ID with a lead and send it back when that person converts offline, "by signing a contract, for example"; Google recommends its enhanced conversions for leads, which match hashed first-party data such as email addresses. Meta's Conversions API (opens in a new tab) connects your server, website platform, app or CRM to Meta, and Meta says it lets ads optimize for later actions, including post-purchase actions and customer scores.
• Tell the system what a conversion is worth. Google's conversion value rules (opens in a new tab) adjust the value of a conversion by location, device or audience to reflect margins or lifetime value, and Target ROAS and Maximize conversion value bid to those values. Google's own example is a leads business that values every lead at $5 but knows its Californian leads are worth $10. Some conditions may not be available for housing, employment or credit ads.

Offline conversion imports and the Conversions API share customer data with the platforms, so they need the permission your privacy notice, the law and the platforms' policies require. Google's page tells advertisers to get consent where the law or its policies require it, and Meta says the Conversions API is not designed to bypass data-sharing rules such as Apple's App Tracking Transparency framework.

2. Targets and budgets

Once bidding is automated, Google's guidance on measuring Smart Bidding (opens in a new tab) says "adjusting bid targets and budgets are the best levers to pull to control performance." The trade-off is direct: lowering a target ROAS makes the bid strategy more competitive in auctions and likely to win more conversion volume, and raising it asks the system to prioritize a higher return over volume. The reverse applies to a target CPA.

Google's advice on setting a target is conservative. Report conversion values across the relevant campaigns for four weeks or three conversion cycles, whichever is longer, before switching to value-based bidding, and set the target at or below the ROAS the account has achieved. For a known spike such as a promotion, it offers seasonality adjustments, and for tracking outages, data exclusions, which Google says may help limit the fluctuations an outage causes.

On Meta, the learning phase guidance (opens in a new tab) asks for realistic budgets and warns against frequent budget changes, which can send an ad set back into learning. Meta says a very small or inflated budget gives the delivery system an inaccurate indicator of whom to optimize for.

One allocation neither system makes is the split between platforms. Performance Max moves budget across Google's channels, and Advantage+ campaign budget moves it across the ad sets in a Meta campaign, but how much goes to Google and how much to Meta stays your decision. Make it on your own sales or pipeline records, since each platform's reports describe its own ads.

3. The creative the system can choose from

With bidding and audience selection automated, creative remains an input your team can change and test.

Google's asset group best practices (opens in a new tab) for Performance Max ask for every asset type in each asset group and for variations, including up to 15 headlines, 5 descriptions and 20 images; its video asset page (opens in a new tab) allows up to 15 videos. If you supply no video, Google may generate one from your other assets, though it still recommends uploading at least one yourself. It also recommends waiting two to three weeks after editing assets before deciding to replace the ones that perform poorly.

Meta's creative fatigue (opens in a new tab) page describes what happens when an audience has seen the same creative too many times: people may be less likely to engage, and cost per result can rise. When Meta judges that an audience has seen an ad too many times, Ads Manager marks it "Creative limited" if its cost per result is above what your past ads cost but less than twice as much, and "Creative fatigue" at twice or more. Meta counts recent exposures of the same image or video across other campaigns from your Page. The page says this feature is available only for ad sets with one creative, and not for Advantage+ catalog ads, dynamic creative or Advantage+ app campaigns, so many automated setups will not show these statuses. One of its recommendations is a new ad with an image or video that is materially different from the original, and it notes that keeping the original running may maximize results.

I wrote about this pattern in outbound sales in 2020, explaining Andrew Chen's law of shitty click-through rates: "every marketing tactic is great when it starts, but then everyone starts doing it and people become immune to the tactic." An ad concept goes through a smaller version of the same cycle, and Meta's fatigue statuses are a way of seeing it. Plan for new creative as a steady supply.

There is one tension to manage. Meta's learning phase guidance also warns against high ad volumes, because the system learns less about each of many ads and ad sets. The practical answer is creative that is genuinely different, introduced into a structure that stays simple. Many near-copies spread across many ad sets give the system less to learn from each one.

4. The limits the system must respect

These systems are designed to widen reach, so the limits you set are where your brand and your buyer definition live.

• Audience limits. Advantage+ audience (opens in a new tab) expands beyond your targeting "subject to your non-negotiable constraints". Meta lists location, minimum age, language and custom audience exclusions as the constraints it does not expand. Its developer documentation also limits the minimum age setting to between 18 and 25 when Advantage+ audience is on, so confirm how any age restriction you need is enforced in your account. Other targeting is not on that list, so check that every limit you rely on is one of these before you switch it on.
• Search and brand limits in Performance Max. Google offers brand exclusions and negative keywords (opens in a new tab) at campaign level. The negative keywords apply to Search and Shopping inventory only; Display and video need excluded content keywords and placement exclusions at account level, and Google says it "can't guarantee" every related placement is blocked. The search terms report for Performance Max (opens in a new tab) shows which queries triggered your ads, so you can add irrelevant ones as negatives and write new text assets around the terms that perform well.
• Landing page selection and generated text. With final URL expansion on, Google may send people to a different page on your site and write a matching headline and description. Its Performance Max page states that you remain responsible for the accuracy of that page content and of all dynamically generated assets.

5. Time, and how much evidence you wait for

The systems learn from results, so allow for conversion delay, the time between an ad click and the conversion being reported back to the platform. Judge performance over complete conversion cycles, and remember that significant edits can restart learning.

• Meta says an ad set usually leaves its learning phase after about 50 results in the week after its last significant edit. During learning, performance is less stable and cost per acquisition usually higher, and significant edits restart it.
• Google recommends judging Smart Bidding over periods with at least 30 conversions, such as a month or longer, or 50 for Target ROAS. Its measurement page suggests at least two full conversion cycles and allowing one to two cycles after a major change to bidding or budget before changing anything else.
• Conversions arrive late. Google's guidance on testing a bid strategy says to leave out recent days in which less than 90% of conversions have been reported, and to exclude a ramp-up of two weeks or three conversion cycles, whichever is longer.

If an account produces only a handful of qualified conversions a month, those thresholds are hard to reach. Consolidate similar campaigns or ad sets, or optimize to an earlier stage that happens more often while you keep checking what those leads become. For campaign types that offer Maximize clicks or manual CPC, Google's advice for a campaign without enough conversion history is to use them to build that data before switching to an automated conversion strategy. Performance Max (opens in a new tab) supports only Maximize conversions and Maximize conversion value, so this fallback is not available there.

A rhythm for checking the levers

The platforms' timing guidance points to a slower cadence than daily bid management. This is a working rhythm drawn from that guidance; adjust it to your conversion volume and sales cycle.

Google's custom experiments (opens in a new tab) compare an original campaign with a changed copy. Google lists them for Search, Display, Video and Hotel campaigns; Performance Max and Shopping are not on that list. For a bid strategy test (opens in a new tab), Google's guidance comes down to three steps:

1. Choose an established campaign and an experiment split that would provide at least 30 conversions in the last 30 days, and change one variable.
2. Split traffic and budget between the original and the experiment. Google recommends 50%, and editing either version during the test makes the result harder to read.
3. Exclude a ramp-up of two weeks or three conversion cycles, whichever is longer, then let the experiment run uninterrupted for at least 30 days, leaving out recent days in which less than 90% of conversions have been reported.

Image

Zoom

Original timeline based on Google Ads’ Smart Bidding experiment guidance. Campaign eligibility and conversion volume still matter; recent incomplete conversion data should be excluded.support.google.com (opens in a new tab) · support.google.com (opens in a new tab)

A split test compares two versions of a campaign. Whether the ads cause conversions at all is a different question. Meta's Conversion Lift (opens in a new tab) compares people who could see your ads with a group that could not, and lists among the questions it answers whether your attribution model agrees with a controlled experiment. Meta's guide, read in October 2026, gives a baseline of a campaign started in the past year with at least $5,000 in spend and 500 conversions under a 1-day click, 7-day click or 1-day view attribution setting. The requirement scales for campaigns longer than 90 days (its 180-day example needs $10,000 and 1,000 conversions), and self-serve tests also have signal-quality requirements, so confirm eligibility before planning a study. The AI-driven advertising guide covers lift studies and test design in more detail.

Questions to ask your team or agency

The aim is to make sure the system optimizes toward what your business values, with enough distinct creative and clear limits, and that someone judges it on evidence rather than on a bad week.