Should you let AI run your ad campaigns? A 30-day test plan

Picture the owner of a landscaping company in Spokane. In early September, an email from Google Ads says the company's Search campaign has been upgraded to AI Max. A week later, Meta's Ads Manager suggests switching the fall cleanup campaign to Advantage+. Both promise more leads for the same money. The owner's question is reasonable: should I let the platforms run this, or keep doing it myself?

The honest answer depends on your tracking, your volume and your margins, and you do not have to guess. Google and Meta both have built-in tools to split a budget and compare two setups fairly. This guide explains what changed this fall, what you actually hand over to the AI, what the platforms claim and what outside data adds, and a 30-day test with a decision rule written before you start.

What changed this fall

Automation in ad platforms is not new; what is new is how far the default has moved. On September 1, Google began automatically moving Search campaigns that used automatically created assets or the campaign-level broad match setting into AI Max for Search. Campaigns that used automatically created assets were upgraded with search term matching and text customization switched on. Dynamic Search Ads, an older automated format, are now scheduled to follow from February 2027. ¹²

Meta is heading the same way. On its second-quarter earnings call, Meta said its Advantage+ campaigns, which automate targeting, placements and budget, had passed an annual revenue run rate of more than $75 billion, and that about nine million small businesses use at least one of its AI ad creative tools. Its new Muse Image model is coming to advertisers through Advantage+ creative to produce ad variations. ³⁴

Even smaller platforms are removing manual levers. From October 1, Microsoft Advertising no longer offers a maximum cost-per-click cap when you create a new standalone campaign that uses automated bidding such as Maximize Conversions. Existing campaigns keep the cap they already have. ⁵

What you hand over when AI runs the campaign

Running a campaign comes down to five decisions: who sees the ad, what it says, where the click lands, how much you bid and where the budget goes. Automated campaigns take some or all of those decisions and make them search by search or person by person, using signals you cannot see. The table shows what moves to the platform and what stays yours.

Two details in Google's documentation matter for small advertisers. AI Max's features need a conversion-based Smart Bidding strategy, so it only makes sense if your conversion tracking is solid. And when final URL expansion picks a different landing page, pinned headlines may not show, because Google assumes the original ad may not fit the new page. In practice, every page on your site becomes a possible landing page, so outdated pages become a risk. ⁷

What the platforms claim, and what outside data adds

Both companies publish strong numbers. Google said at launch that advertisers who switch on AI Max typically see 14% more conversions or conversion value at a similar cost per acquisition or return on ad spend. It has since cited 27% more conversions for advertisers whose campaigns relied mostly on exact and phrase match keywords. ⁸¹⁰

Meta's figures point the same way. A 2025 Meta study, built on a large randomized trial using methods developed with researchers at UC Berkeley, credited Meta's AI-driven ad tools with a 22% improvement in return on ad spend. Meta's product page for Advantage+ sales campaigns cites a 20% lower cost per result on average. ⁹¹¹

The outside picture is thinner and more mixed. The most cited independent data point comes from Lunio, a company that sells click-fraud protection. Across 414 million retail clicks, it found that the share of invalid clicks in AI Max campaigns rose from 2.46% at the end of 2025 to 5.28% in the second quarter of 2026, while standard Search fell from 3.72% to 3.07%. Lunio measured clicks that had already passed Google's own filters, and it has a product to sell, but the direction is a good reason to watch traffic quality closely during any test. ¹²

None of this means the platforms are wrong. Their averages come from millions of accounts, many with far more conversion data than a local business has. Your result depends on how well the system can learn from your account, which is exactly what a test measures.

When automation tends to work, and when it struggles

Automated campaigns learn from conversions. The more often they see the outcome you care about, the faster they find the searches and people that produce it. Google recommends judging Target CPA bidding on at least 30 conversions over the previous 30 days. Meta labels an ad set Learning limited when it is unlikely to get around 50 optimization events in the week after its last significant edit. A business that books eight jobs a month gives either system very little to learn from. ¹³¹⁴

Before you test: fix what the AI will learn from

An automated campaign optimizes toward whatever you tell it counts as a conversion. If a spam form submission counts the same as a booked job, the system will happily find more spam. Before any test, make sure the conversion you optimize for is the one you care about. For lead businesses that usually means importing qualified leads or booked jobs from your CRM back into Google Ads, and sending the same events to Meta through its Conversions API, rather than counting raw form fills.

Then write your guardrails down. In Google Ads, that means negative keywords, brand exclusions, URL exclusions for pages that should never receive ad traffic, and text guidelines that tell AI Max which words and claims to avoid. In Meta, decide which creative enhancements you allow, which placements you exclude and who approves generated variations before they run. Finally, pick one decision metric, such as cost per qualified lead or profit-based return on ad spend, and the improvement that would make you switch. ⁷

The 30-day test, week by week

Both platforms have tools for a fair comparison. Google offers a one-click AI Max experiment that splits a Search campaign's traffic 50/50 between the current setup and AI Max. Meta's A/B tests compare two versions of a strategy, such as an Advantage+ campaign against your manual setup, and make sure nobody sees both. Meta notes that its A/B tests can run for up to 30 days and that tests shorter than 7 days may be inconclusive. A split test beats a before-and-after comparison because both versions run in the same weeks, through the same weather, news and seasonality. ¹⁵¹⁶

The first week is the hardest, because it is tempting to react to early numbers. Resist it. Meta treats pausing an ad set or changing its optimization event, audience or creative as a significant edit that restarts learning, and large budget or bid changes can do the same. Google's automated bidding also needs time to settle. If something is clearly wrong, such as ads appearing for a service you do not offer, add a negative keyword or exclusion and leave everything else alone. ¹⁷

From week two, put twenty minutes a week on the calendar. In Google Ads, read the search terms report and the landing pages AI Max chose, and look at the headlines it wrote. In Meta, preview the generated variations and read the comments people leave on them. In both, check a sample of leads in your CRM. A cheaper lead that never answers the phone is not cheaper.

When day 30 arrives, wait before judging. Many conversions are recorded days after the click, a delay called conversion lag, and lead qualification takes longer still. Give the data one to two weeks to arrive, longer if your sales cycle runs over a month, then compare the two versions on the metric you chose at the start.

Day 30: read the result with a decision rule

Here is what a decision looks like with real arithmetic. Say the landscaping company splits a $3,000 monthly Search budget evenly. The current setup spends $1,500 and brings 30 leads, of which 12 turn into quoted jobs: $50.00 per lead and $125.00 per qualified lead. The AI Max version spends $1,500 and brings 41 leads, of which 13 qualify: $36.59 per lead, but $115.38 per qualified lead, because more of the extra leads were poor fits. Illustrative arithmetic only; these are not results from a real account.

Judged on cost per lead, AI looks like a 27% win. Judged on qualified leads, it is about 8% better, below the 10% threshold the owner wrote down before the test. The right move is not to declare victory. It is to send the qualification data back to the platform so the AI learns which leads matter, extend the test by two weeks and decide then. Writing the rule first means you cannot talk yourself into a result afterward.

What should never be handed over

Even when a test goes well, some decisions stay with people. Prices, offers and promises are yours: an ad that implies a discount or a same-day guarantee you do not offer creates a customer service problem the platform will not solve. Regulated claims in health, finance and legal work need human review every time. Your total budget and the pages you allow ads to send traffic to are business decisions, not optimization settings. And someone should look at what is actually running, including generated text and images, at least once a week.

Start this week

• Check whether any Google Search campaigns were moved to AI Max on or after September 1, and review which features were switched on.
• Confirm the conversion you optimize toward is a qualified lead, booked job or sale, not every form fill.
• Write down your decision metric and the improvement that would make you switch.
• Set negative keywords, brand exclusions, URL exclusions and text guidelines before testing.
• Launch a 50/50 AI Max experiment in Google Ads or an A/B test in Meta, and leave it alone for the first week.
• Book a weekly 20-minute review of search terms, landing pages, generated creative and lead quality.