Microsoft AI Max Is On by Default. Give It a Test Plan.

A default setting is still a strategy decision. It just arrives wearing sweatpants.

Microsoft Advertising made AI Max available for Search campaigns on August 27. New Search campaigns will have it enabled by default. Accounts using scheduled Google Import can also inherit AI Max settings from corresponding Google Ads campaigns.

That deserves a deliberate test, especially if nobody on the team remembers checking the box.

Automation should earn a wider lane with evidence. A toggle is not evidence.

What AI Max can change

Microsoft’s launch announcement describes three capabilities:

• Search term matching can reach queries beyond the campaign’s keyword list by using keywords, ads, landing pages, and other context.
• Text customization can generate and test additional ad messaging from existing assets and website content.
• Final URL expansion can choose a landing page it considers more relevant instead of sending every click to the URL you supplied.

These are separate operating risks. A campaign can reach the wrong query, say the wrong thing, or send the right person to the wrong page. A single conversion-rate chart will not tell you which one happened.

First, audit the settings that may travel

Before launching a test, export the current Microsoft campaign settings and inspect scheduled Google Import jobs. Microsoft says AI Max settings can now carry over from Google Ads, and corresponding Microsoft campaigns can keep inheriting them on future scheduled imports.

Record five items for every test campaign:

1. AI Max on or off;
2. search term matching on or off;
3. text customization on or off;
4. final URL expansion on or off, including URL exclusions;
5. the import job that can change any of those settings.

Do this again after the next scheduled import. A clean launch screenshot is not useful if tomorrow’s sync quietly changes the experiment.

Make three risk lists before spending

Microsoft gives advertisers brand inclusions and exclusions, negative keywords, term exclusions for generated copy, and URL exclusions. Use them as a preflight, not as cleanup after an awkward screenshot.

Query risks: competitor names, job-seeker intent, do-it-yourself research, consumer support questions, irrelevant geographies, and adjacent services the business does not provide.

Copy risks: unapproved discounts, guarantees, regulated claims, superlatives, competitor references, outdated product names, and language that changes the scope of the offer.

Routing risks: careers, privacy, investor, support, old campaign, thin location, and gated pages. Also exclude any page whose form, price, inventory, or proof is not ready for paid traffic.

The official Search campaign setup documentation says term exclusions can be applied to generated text and recommendations, while URL exclusions can keep final URL expansion away from specific pages. The controls exist. The team still has to name the boundaries.

Do not test all three features as one mystery box

A useful field test needs a stable control campaign and a narrow change. Start with a campaign that has enough conversion volume to learn, a clear offer, accurate tracking, and landing pages you trust.

Then sequence the tests:

1. Query expansion first. Hold approved ads and landing-page routing steady. Evaluate the added search terms and the outcomes they produce.
2. Generated copy second. Keep query and routing conditions understood. Review every generated asset and compare claims, click quality, and conversion quality.
3. URL expansion third. Use approved copy and inspect which query, ad, and landing page combinations the system chooses.

Microsoft says the suite works best together. That may be true for performance. It is terrible advice for diagnosis on day one. Learn where each feature helps and where it gets cute before combining them.

Build the measurement sheet around decisions

Microsoft reports an average conversion lift from advertiser-run experiments. Useful context, but somebody else’s average cannot approve your rollout.

For each test arm, track:

• spend, impressions, clicks, conversions, and conversion value;
• qualified leads, sales, cancellations, or another downstream business outcome;
• new search terms grouped as useful, irrelevant, risky, or unclear;
• generated assets marked approved, edited, removed, or escalated;
• landing pages selected, with conversion and lead-quality results by page;
• brand, legal, pricing, or customer-experience exceptions.

Microsoft says expanded matching appears with an “AI optimized” label in search-term and keyword reporting. Generated copy appears in the asset report. URL routing appears in a search-term landing-page report. Pull all three. If the report cannot answer which query produced which promise on which page, the test log should.

Our marketing causality check is useful here. Choose the decision and evidence window before results arrive. Otherwise, the winning metric tends to become whichever number looks friendliest on Friday.

Use stop rules that a busy team will actually follow

Pause or narrow the feature when:

• a generated claim is false, unapproved, or materially misleading;
• a protected brand or excluded term appears;
• traffic repeatedly lands on an excluded, broken, or unsuitable page;
• irrelevant query spend exceeds the amount approved for discovery;
• lead volume rises while qualified outcomes fall beyond the agreed limit;
• a Google Import changes the tested settings without review.

Name the person who can pause the test. Give that person the report path, the account access, and the threshold. “The team will monitor it” is how everyone monitors the same dashboard while nobody touches the switch.

The two-week field plan

1. Export the campaign and import settings before launch.
2. Document query, copy, brand, and URL boundaries.
3. Choose one AI Max capability to test first.
4. Preserve a credible control and avoid unrelated campaign edits.
5. Review search terms, assets, and routed pages every business day during the test.
6. Tag downstream quality, not just platform conversions.
7. Apply the written stop rules when a boundary fails.
8. At the end, choose expand, revise, hold, or disable. Record why.

AI Max may find good queries your keyword list missed. It may write a useful variation. It may rescue a click from a mediocre landing page. Fine. Let it prove each claim in your account, under your rules, with your customers.

About the Author

Brian Mundt

Director of Content Strategy

Brian helps marketing teams turn automated platform settings into bounded tests, plain decisions, and evidence that can survive the next budget meeting.