Google AI marketing tools: ads, audiences and predictions
Short answer Google's AI finds customers in two places. In Google Ads, Smart Bidding, Performance Max and Demand Gen decide who sees your ad, using your conversions and customer lists as hints. In Google Analytics 4, predictive audiences guess who will buy soon, but only with 1,000 buyers and 1,000 non-buyers a week. Most small businesses should feed the ad side first.
Want help applying this to your business? Talk it through on a free 30-minute call.
Book a callWhich Google AI tools find and predict customers?
Five tools do the targeting and predicting. Each one runs on data you give it; without that data, it guesses.
| Tool | What its AI does | Where | What it needs from you | Small business fit |
|---|---|---|---|---|
| Smart Bidding | Bids higher on searches likely to become a lead | Google Ads campaigns | Conversion tracking that counts real leads | Yes, once leads are tracked |
| Audience signals | Treats your lists as hints, then looks wider | Performance Max | A customer list or website visitor list | Yes |
| Lookalike segments | Finds new people who resemble your list | Demand Gen only | A list with 100+ matched people | Sometimes |
| Customer Match | Matches your customer emails and phones to Google accounts | Google Ads | Customers who gave you their details | Yes, with limits |
| Predictive audiences | Guesses who will buy or stop visiting in 7 days | Google Analytics 4, shared to Google Ads | Purchase events at large volume | Rarely |
Sources: Google Ads Help on Smart Bidding and Performance Max audience signals, the Google Ads API guide to Lookalike segments, the Customer Match policy and Analytics Help on predictive audiences.
The pattern: the ad tools work with small data; the analytics predictions don't.
How does Google's AI decide who sees my ad?
It decides per auction, the split-second sale of each ad slot. Google says Smart Bidding uses its AI "to optimize for conversions or conversion value in every auction" (Google Ads Help: About Smart Bidding). A conversion is the action you count as a win, such as a call or a booking form.
Your audience lists change how much that AI listens to you, depending on one setting:
- Observation. Reach stays the same. Google says lists added this way "will be used as signals for Smart Bidding strategies" (Google Ads Help: About "Targeting" and "Observation" settings).
- Targeting. Ads show only to people on the list. That narrows reach a lot and, for Customer Match lists, isn't open to most small accounts (see below).
In Performance Max, the campaign type that runs one campaign across Google's channels, your lists are only a starting point:
"Performance Max may show ads to relevant audiences outside of your signals if they have a strong likelihood of converting" (Google Ads Help: Performance Max audience signals)
On Search, the AI also picks which searches your ad matches. Dynamic Search Ads, which matched searches to your site's pages, are being folded into AI Max; Google pushed their automatic upgrade to February 2027 (Google Ads & Commerce Blog: We're upgrading Dynamic Search Ads to AI Max). AI Max for Search campaigns covers whether to turn it on.
So the AI's real target is your conversion goal. If you count page views as conversions, it finds people who view pages. Performance Max for a local business covers when that campaign type pays off, and Google Ads automation covers which bids to hand over.
Can Google find people like my customers?
Yes, in two ways, both built from your own list.
Lookalike segments find new people who share traits with a "seed" list, such as past customers. Google's rules (Google Ads API: Lookalike audience segments):
- Demand Gen campaigns only. Demand Gen shows image and video ads on YouTube, Discover, Gmail, Maps and the Display Network (Google Ads Help: About Demand Gen campaigns). Since April 1, 2026, Google requires a budget of at least US$5 a day, or the local equivalent, for new or edited campaigns (Google Ads Developer Blog).
- Your seed lists together need at least 100 active matched people.
- You pick how wide: narrow, balanced or broad, plus the countries.
Customer Match uploads your customers' emails or phone numbers so Google can recognize them. The Customer Match policy sets what each account can do:
| Use | Any account in good standing | 90+ days and over US$50,000 lifetime spend |
|---|---|---|
| Observation and exclusions | Yes | Yes |
| Targeting only the list | No | Yes |
A list stays usable only with at least 100 members added or updated in the last 540 days, and only data you collected yourself counts (Google Ads policy: Customer Match policy). Google also says manual bidding ignores these lists (Google Ads Help: About Customer Match). Upload steps, consent rules and the "new customers only" setting are in Google Ads for AI-driven search.
Website visitor lists (Google calls them "your data segments", formerly remarketing) need at least 100 active visitors in the last 30 days to show on Search (Google Ads Help: Your data segment requirements).
One hard limit: some businesses can't use any of this. Google's personalized advertising policy says:
"Advertisers promoting products and services that fall within sensitive interest categories are unable to use advertiser-curated audiences." (Google Ads policy: Personalized advertising)
Health is one of those categories, and so is negative financial status, which names bankruptcy services (Google Ads policy: Negative financial status). The blocked list includes Customer Match, your data segments and Lookalike segments. Those businesses rely on keywords, location and Smart Bidding instead.
Can Google Analytics predict who will buy?
GA4 can, but the bar is high. It offers ready-made predictive audiences such as "Likely 7-day purchasers", "Likely 7-day churning users" (people likely to stop visiting) and "Predicted 28-day top spenders" (Analytics Help: Predictive audiences). Each is built from a predictive metric: purchase probability, churn probability or predicted revenue.
To turn them on, your property must qualify:
"In the last 28 days, over a seven-day period, at least 1,000 returning users must have triggered the relevant predictive condition (purchase or churn) and at least 1,000 returning users must not." (Analytics Help: Predictive metrics)
The templates are built on purchase events (Analytics Help: Predictive audiences). A plumber or law firm that counts form fills and calls, not online purchases, won't see them. An online store with steady sales might.
If you qualify, Google says predictive audiences "are automatically shared with any advertising accounts you have linked to your property", including Google Ads (Analytics Help: Predictive audiences). If your property later stops qualifying, those audiences stop adding new people.
What GA4 gives everyone else: on August 10, 2026, Google announced AI summaries at the top of the Analytics homepage and Ask Advisor, an AI assistant that answers questions about your campaign data (Google Ads & Commerce Blog). Ask Advisor works in English-language accounts and asks your approval before it changes anything (Google Ads: Ask Advisor). These explain what happened; they don't predict buyers. For the events to track instead, see GA4 key events for a small business.
What should I set up first?
- Count real leads as conversions. Calls over a set length, booked appointments, paid jobs. Every tool above aims at whatever you count.
- Upload a customer list for Observation and exclusions, if your category allows it.
- Build a website visitor list in Google Ads or GA4, so you can show ads to people who left without booking.
- Add both as signals in Performance Max, or as seed lists in Demand Gen once a list has 100+ matched people.
- Skip predictive audiences unless you sell online and pass 1,000 buyers a week.
Example
A roofer with 400 past customers, US$2,500 a month
The roofer uploads 400 customer emails to Customer Match. Google matches enough to pass 100 members, so the list goes into Performance Max as an audience signal and as an exclusion on a "new customers" campaign. GA4 predictions never turn on: the site has no purchase events.
An online kitchenware shop, C$30,000 a month in sales
The shop sends purchase events with value to GA4 and has about 1,500 buyers a week. "Likely 7-day purchasers" turns on and flows to Google Ads, where it is added under Observation for its Search campaigns.
A bankruptcy law firm with 2,000 past clients
Bankruptcy services fall under a sensitive category, so the firm can't use Customer Match, visitor lists or Lookalike segments. It runs Search ads on service and town keywords with Smart Bidding aimed at booked consultations.
FAQ
Is Google AI targeting the same as Smart Bidding?
No. Smart Bidding sets the bid for each auction. Targeting tools, such as audience signals and Lookalike segments, shape who the AI considers. Both chase the conversion you track.
Do I need Google Analytics for Google's AI ad targeting?
No. Google Ads tracks conversions and builds lists on its own. GA4 adds predictive audiences, but only for properties with 1,000 buyers and 1,000 non-buyers in a week.
Why can't I target my Customer Match list directly?
Google limits "Targeting" to accounts with 90 days of history and over US$50,000 lifetime spend. Smaller accounts can still use the list for Observation and exclusions.
Does Google use my customer list to find similar people on Search?
Lookalike segments work only in Demand Gen campaigns. On Search and Performance Max, your list is a signal Smart Bidding uses, not a lookalike audience.
Different for you?
- You don't track leads yet. Fix that before any AI targeting: Tracking a lead from Google Ads to a booked job.
- You're choosing between AI tools for writing ads and reports. See Best AI tools for Google Ads.
- You run Facebook and Instagram ads too. Meta's version of customer lists is in Facebook Custom Audiences.
- You want someone to set up the lists and signals. See /paid-ads/.