AI Max for Search: Opt In, Wait, or Fence It?

AI Max for Search is a trade, stated plainly: you give up some keyword control, and Google's AI expands your campaign into queries, headlines and landing pages you did not specify. Whether that trade pays depends less on the feature than on your account — specifically, whether your tracking can be believed and whether anyone will read what the expansion matched. Decide the conditions first and the opt-in decision mostly makes itself.

What AI Max changes inside an existing Search campaign

Three mechanisms switch on, and each takes a decision away from you:

• Search term matching. The campaign can serve on queries beyond your keyword list — including ones no match type would have reached. Your exact keywords keep priority for identical queries; everything else opens up.
• Text customisation. Google generates headline variants against those new queries, steerable with text guidelines but not written by you.
• Final URL expansion. Google may route clicks to pages it judges more relevant than your designated final URL, unless you exclude URLs or constrain it.

Notice what kind of feature this is: not a bid change you can watch on a chart, but a matching change that alters which auctions you enter at all. That is why the review workload lands on the search terms surface — the place expansion mistakes show up first.

The case for opting in

Google's numbers, with their labels attached: the GA announcement in April 2026 claims an average 7% lift in conversions or conversion value at similar CPA/ROAS for the full AI Max suite versus its matching alone — internal data, retail advertisers excluded, sample size undisclosed. Beta-era claims ran higher: 14% against keyword-only setups, and 27% for campaigns leaning on exact and phrase match. Google also argues expansion unlocks queries that simply did not exist in your keyword universe — the "billions of new searches" pitch.

The structural argument is more persuasive than any single number: search behaviour is drifting into long conversational queries that keyword lists cannot enumerate — a drift Google's AI surfaces are accelerating. Some mechanism has to bridge your account to that demand. AI Max is Google's answer, and for accounts with dense conversion data and clean feeds of intent, expansion has real upside the keyword list cannot reach on its own.

There is also a timing argument for engaging now rather than eventually: adjacent features are converging on AI Max whether you opt in or not — automatically created assets and the campaign-level broad match setting fold into it in September 2026, and Dynamic Search Ads follow on their own schedule. Learning the suite's controls on one fenced campaign this quarter beats meeting them for the first time inside a forced migration.

The case for waiting

The counter-evidence is equally concrete. Monks' widely-cited test found 99% of impressions across roughly 30,000 AI Max search terms produced zero conversions — expansion without discrimination, in their sample. And the failure modes are asymmetric: a keyword you never added costs you nothing, while an expansion into the wrong query costs money until someone catches it.

Wait if any of these is true:

1. Your tracking is dubious. Expansion optimises toward your conversion signal; a polluted signal scales pollution. This one is fixable in weeks, not quarters — repair the conversion actions first and the whole decision improves, for AI Max and everything else downstream of the signal.
2. Brand safety is strict. Text customisation writes headlines a compliance team never saw. Text guidelines narrow this; they do not notarise it.
3. Nobody owns query review. If the search terms report went unread before AI Max, it will be unread after — with more in it.
4. The campaign is fragile. Low budgets and thin conversion volume give the system little to learn from and you little room to absorb waste.

The fenced rollout: one campaign, hard walls, weekly review

Between "everything on" and "never" sits the sane default: enable AI Max on one mid-sized campaign and build a fence before flipping the toggle.

• Pick a campaign with real conversion volume and stable history, so drift is visible against a baseline.
• Pre-load the fence: brand negatives, competitor negatives, known-bad themes from past query mining, and URL exclusions for pages that must never take paid traffic (careers, support, logged-in docs).
• Write text guidelines before the system writes headlines — banned claims, required disclaimers, tone.
• Set a 30-day verdict date with the metric named in advance: cost per conversion at equal-or-better volume against the prior period.

The fence is not pessimism. It converts an open-ended matching change into a bounded experiment with an exit.

One more pre-flight check earns its minute: reconcile the fence with what already polices the campaign. Existing shared negative lists should be attached, not duplicated; existing URL exclusions from any DSA history carry intent worth porting; and if the campaign participates in an experiment or a shared budget, understand how expansion will read in those reports before the toggle, not after. Expansion features are rarely dangerous on day one — they are dangerous on day forty, when the novelty has worn off and nobody assigned the watching. The fence plus a named weekly reviewer is the whole discipline.

Monitoring the expansion with an AI reviewer

The weekly chore that makes AI Max safe is exactly the kind of reading a connected AI agent does well. One prompt, run weekly on the fenced campaign:

Pull this campaign's search terms for the last 7 days. Separate keyword-matched terms from AI Max-expanded terms where identifiable. For the expanded set: cost, conversions, and the ten most expensive terms with zero conversions. Compare the two sets on cost per conversion. Draft the negative keywords you would add, with the level to add them at — and wait for my approval.

The pattern to watch across weeks is the ratio between the two sets. If expanded terms convert near parity with keyword-matched terms, widen the rollout. If the expanded set's conversions concentrate in terms you would happily add as keywords — add them as keywords, at exact, and let your negative list absorb the rest. The agent drafts; every write passes through approval; the account keeps a record of what the fence caught.

Set escalation rules in advance so the review never becomes a judgement call made tired: a week where expanded spend outpaces expanded conversions by your chosen threshold triggers a negatives session; two such weeks trigger narrowing the features; a headline that breaches your text guidelines triggers tightening them the same day. Written rules turn the weekly review into administration — which is the point.

The DSA special case: your upgrade is scheduled

If the campaign in question is a Dynamic Search Ads campaign, the calculus changes — waiting has a deadline. Google removes DSA creation in January 2027 and auto-upgrades remaining DSA campaigns from February 2027, with automatically created assets and the campaign-level broad match setting folding into AI Max in September 2026. For DSA users the question is not whether but when and how deliberately: migrate inside the voluntary window with your URL rules and negatives carried over, rather than in February's bulk queue. The step-by-step is in the DSA to AI Max migration plan, and the year's full context in what changed in Google Ads in 2026.

Opt in when the fence exists. Wait when it does not. Fence one campaign when you are honestly unsure, and let thirty days of your own data replace everyone else's averages — Google's and the sceptics' alike. And if the fence is the part you keep postponing, that is the delegable part — the reading, the ratios, the drafted negatives — which is exactly the job a connected reviewer was built for.