Google AI Max: design a test around qualified leads
Build an AI Max experiment with a stable control, qualified-lead metrics, URL checks and a decision sheet. Includes an illustrative cost comparison.
1. Start with a decision, not a toggle
AI Max adds search matching and creative features within Google Search campaigns. Its settings can change which searches you reach, which messages people see and which pages they visit. A useful experiment therefore starts with a specific question. Will this configuration produce more qualified demand at an acceptable acquisition cost? That question is much easier to act on than whether automation is generally good.
Agree on a qualified lead with the sales team. A US or Canadian service business might require an eligible service area, the right project type and a confirmed need. Keep those rules identical for both groups. Avoid changing the definition when one group generates more forms. Otherwise, the experiment becomes a comparison of reporting choices.
2. Establish a reliable starting point
Choose a campaign with a stable offer, working conversion tracking and a usable history. Check for duplicate form events, test submissions and low-value events included in the primary goal. Make a real test journey and confirm the lead reaches the expected system. If the business cannot distinguish a form submission from a qualified opportunity, resolve that gap before judging the new targeting.
Save the campaign configuration and document brand controls, geography, destination pages and tracking templates. Google specifically describes compatibility problems between dynamic destinations and tracking templates. Test representative expanded URLs with their parameters. Review any claim that requires exact wording. Automated copy still needs to reflect what the business actually offers and can deliver.
3. Register the rules before traffic starts
Use an eligible Google Ads experiment to compare treatment and control during the same period. Record traffic allocation and the exact feature settings. Keep the offer, lead handling and landing experience stable unless they are the intended variable. If an unavoidable operational change occurs, log its date and affected group so it does not disappear from the interpretation.
Choose one primary outcome and a small set of safeguards. For lead generation, cost per qualified lead can be the outcome; unacceptable destinations and overspending can be stop conditions. Also state when evidence will be inconclusive. A test with immature sales statuses is not a clean win or loss. Decide how long leads normally need to be assessed before opening the final report.
4. An illustrative example: lower CPL, worse qualification
The following USD figures are invented for teaching; they are not client results or industry benchmarks. Both groups spend $2,000. The control generates 40 submissions and 20 qualified leads. The treatment generates 50 submissions and 15 qualified leads. Cost per submission improves, while the cost of reaching an appropriate prospect rises.
Divide spend by qualified leads: $100 for control versus $133.33 for treatment. That difference warrants investigation, but the table alone does not establish statistical confidence. Compare pending statuses, response delays and lead maturity. If one group contains many unreviewed inquiries, qualification is not yet comparable. The useful next action is a query and journey review, not an immediate claim about all AI Max campaigns.
| Illustrative metric | Control | AI Max |
|---|---|---|
| Spend, USD | 2,000 | 2,000 |
| Submissions | 40 | 50 |
| Qualified leads | 20 | 15 |
| Cost per submission, USD | 50 | 40 |
| Cost per qualified lead, USD | 100 | 133.33 |
5. Inspect the complete customer journey
Read the search query alongside the ad message and actual destination. An appropriate query can still lead to an irrelevant page. A generated message can create expectations that the service page does not support. Separate those failures: blocking a query will not repair a broken form, and rewriting a headline will not fix an ineligible service area.
Collect examples from both high-spend and high-quality segments. Record the observed problem, your explanation and the evidence needed to test it. The ad copy review tool on this site analyzes text you paste and suggests prioritized edits. It does not connect to Google Ads or validate CRM qualification. Use it for editorial preparation, then confirm the suggested wording against your offer and account data.
6. Make the decision reproducible
Close the experiment with one of three decisions: adopt, revise and repeat, or collect more evidence. Adoption requires acceptable economics, mature outcomes and no critical message or destination failures. If gains are concentrated in one segment, preserve that qualification in the decision. An account-wide average can hide the precise audience that made the test useful.
Common mistakes include choosing the winner on clicks, stopping after an early swing, ignoring brand traffic and changing the success metric after results arrive. Save settings, exports, change history and the final reasoning together. The downloadable decision sheet gives the next person enough context to understand why a configuration stayed enabled. It also makes a later reversal easier to explain without rewriting the history.
AI Max experiment decision sheet
Fill in the rules before launch and complete the result after outcomes mature.
Download the working CSV template
Open in Excel or Google Sheets. Replace example rows and empty fields with your own data.
Frequently asked questions
How long should an AI Max experiment run?
Long enough to collect useful evidence and allow the normal lead qualification cycle to finish. There is no universal calendar duration. Define the evaluation rule before launch and account for delayed outcomes.
Can I test only selected AI Max features?
AI Max has individual feature controls. Check what your account supports and record the exact treatment configuration. A label such as AI Max enabled does not describe all relevant settings.
Should generated ad copy receive a separate review?
Yes. Review factual claims, offer restrictions and agreement with the destination page. A clear message can still be inaccurate, and an automated asset remains part of your advertising.