There are two ways to move a Search account into the AI Max era. One is to do nothing, let the September 2026 auto-upgrade arrive, and inherit whatever defaults Google assigns. The other is to treat the migration as a project: test it, instrument it, set the controls yourself, and decide campaign by campaign. The first path is less work this quarter. It is also how budgets quietly leak.
This is part two of a two-part series. Part one, AI Max and the Death of Dynamic Search Ads, covered what AI Max is and why Google built it. The short version: AI Max is a setting inside a standard Search campaign that finds queries without keywords, writes the ad copy itself, and can pick the landing page. It does the three jobs that used to define a search marketer's week. Google is retiring Dynamic Search Ads, automatically created assets, and campaign-level broad match into it.
This post is the practical part. It is a checklist for a paid search practitioner moving real campaigns, real budget, and a real boss who will ask what happened to the numbers.
Origin: why a migration plan beats a deadline
The instinct with a Google deprecation is to wait. Deadlines slip, features change, and the auto-upgrade promises to mirror your old setup so nothing breaks on day one. All true. None of it is a reason to wait.
The reason to move early is control over defaults. When a Dynamic Search Ads campaign auto-upgrades, all three AI Max features switch on: search term matching, text customization, and final URL expansion. If you migrate voluntarily first, using Google's one-click upgrade tools, you choose which features run and at what level. You can dial text customization off on a regulated campaign. You can leave final URL expansion off where your landing page logic is deliberate. Wait, and that choice is made for you.
There is a second reason, and it is about evidence. Once a campaign upgrades, the old DSA query and performance data starts to blur, and practitioners running migrations report that historical DSA benchmarks not exported before the upgrade can become partially unavailable. To know whether AI Max helped, you need a clean before, and the before only exists if you save it now. Export 90 days of CPA, ROAS, cost per lead, impression share, and the search term report for every affected campaign, and keep it somewhere outside the Google Ads interface.
Present: how to test AI Max without betting the account
The single most repeated piece of advice from independent practitioners is the same: do not convert the whole account. Test first, on a subset, against a controlled comparison. AI Max results vary enough that a blind switch is a gamble.
The cleanest tool is the AI Max experiment. It is a built-in A/B test that splits traffic inside a single existing campaign. Part of the traffic runs with AI Max off, the control. The rest runs with it on, the trial. Because the traffic stays in one campaign rather than a separate copy, the learning period is shorter and the comparison is fairer. To start one, go to Experiments in the Campaigns menu, add a new experiment, and choose campaign features and settings, then AI Max for Search.
A few rules keep the test honest.
Pick representative campaigns, not your best or your worst. One or two mid-volume campaigns with steady conversion data tell you more than a low-traffic experiment that never reaches significance.
Run it for at least four weeks. Smart Bidding needs a learning period, and a two-week read mostly measures noise. Plan the window so the test does not end inside a seasonal spike.
Know what the experiment cannot test. AI Max experiments do not support every setting. Per Google's documentation, you cannot run one if the campaign already uses text customization, brand inclusions or exclusions, ad group location targeting, a portfolio bid strategy, or shared budgets. The experiment by default tests search term matching and asset optimization. Treat the result as a signal about the core engine, not a verdict on every control.
Watch the search term report from day one, not week four. This is where irrelevant query expansion shows up first, so read it every few days for the opening weeks. You are looking for queries unrelated to your business, competitor brand names, and poorly converting traffic from the Search Partner Network. Catching that early means you add negatives before the spend compounds, not after.
Present: the controls that still exist, and how to use them
The story that AI Max is an uncontrollable black box is wrong, and believing it leads to the worst outcome, using none of the controls. AI Max keeps the structure of a Search campaign, keywords, ad groups, negative keywords, and a search term report, then layers automation on top. The guardrails are real. They are just off by default, which means setting them is your job.
Negative keywords still work, and they matter more here, not less. AI Max matches queries far more broadly than a keyword list, so the negative list is doing more work. Build it before the campaign runs, not reactively. An account-level negative list covering jobs, careers, free, DIY, adult terms, and known competitor names stops a category of waste on day one.
Brand controls are the fix for the most documented failure mode. AI Max has a habit of chasing competitor terms. One analysis surfaced by Lunio found a campaign sending 69 percent of its impressions to competitor brand terms, more than 8,000 impressions against 77 from broad match. The system reads your deliberate decision to avoid a rival's name as an untapped gap. Brand exclusions tell it otherwise. In Tools, build a brand list, then apply it as an exclusion in the AI Max settings panel at campaign or ad group level. Brand inclusions do the reverse, fencing a campaign to your own brand. If you run a separate brand campaign, this is how you stop AI Max from blurring brand and non-brand traffic.
URL controls govern where final URL expansion can send a click. With expansion on, AI Max can route traffic to any indexable page on your domain: an out-of-stock product, a support article, a gated asset, a stale page you forgot existed. URL exclusions, set at the campaign level, keep paid traffic off pages that should never be a landing page. URL inclusions, set at the ad group level, do the tighter job of restricting delivery to a chosen set of pages. Per Google's setup guide, both live in the AI Max section of campaign and ad group settings.
Location settings are easy to get wrong and quietly expensive. The location-of-presence option decides whether your ad shows to people physically in your targeted area or also to people merely searching about it. Left on presence or interest, a campaign for a Chicago business can serve to someone searching for the Philadelphia version. For a local advertiser, setting location to presence only is often the difference between qualified and wasted clicks.
The AI Brief is the newest control and the one that changes the job most. Announced at the end of April 2026, it lets you steer AI Max in plain language across three areas: messaging guidelines for what ads should and should not say, matching guidelines for which searches to prioritize, and audience guidelines for tailoring the message to a segment. It also adds something earlier controls lacked, a preview: sample assets and sample searches you can review before the campaign goes live, then adjust the wording. Existing text guidelines roll into the Brief's messaging section. The shift is from defensive blocking toward positive instruction. Term exclusions, a hard list of words the system may never use, still exist for absolute no-go terms like a regulated phrase or a competitor name. Use both: the Brief for tone and direction, term exclusions for the words that can never appear.
Present: the metrics change, so the job changes
This is the part that trips up experienced practitioners. The old optimization loop was keyword bids and match types. The new loop is inputs and guardrails. You are not tuning a bid on a keyword. You are setting the negatives, the brand lists, the URL rules, and the Brief, then judging the output.
Two reporting habits matter most.
Read the search term report as a waste audit, not a curiosity. AI Max gives you a dedicated view, search terms and landing pages from AI Max, which shows the query, the headline served, and the landing page together. Going through it weekly is the core maintenance task now. Underperforming search terms can be excluded as negative keywords; bad landing pages can be excluded as negative URLs, directly from that report.
Judge AI Max on incremental conversions, not reported ones. This is the most important measurement point in the migration, and it is where the numbers mislead. Adalysis found that AI Max often claims credit for queries your existing exact and phrase keywords were already winning, because it treats all keywords as broad match and ignores the normal match-type hierarchy. Its reported totals are not incremental gains; they bundle traffic you already had. The only honest read is the experiment, the lift of the trial arm over the control arm. For the deeper logic of why reported and incremental results diverge, see incrementality without the jargon. For AI Max, the rule is simple: trust the experiment delta, distrust the dashboard total.
On the Dynamic Search Ads deadline itself, do not assume AI Max is a clean swap. DSA was a catch-all for queries keywords missed. AI Max is built to interpret intent, a different job, and as PMG and others note, it does not fully replace DSA's safety-net coverage. Before the deadline, document what each DSA campaign actually did. Some were exploratory, some were efficient long-tail drivers, some were pure catch-all, and those roles do not transfer evenly. Audit query reports, expand keyword coverage where high-value queries deserve intentional targeting, and tidy your landing pages, because AI Max reads your site the way DSA did and rewards a clean one.
Future and impact: the mistakes, and the honest reality
A short list of the migration mistakes that show up most.
Turning everything on at once. The whole-account switch with no experiment is the fastest way to a bad month and a hard conversation. Stage it.
Skipping brand controls. Leave brand exclusions unset and you should expect the competitor-term drift, because it is the default behavior, not the exception.
Ignoring the search term report. The report is the steering wheel now. An account where nobody reads it is on autopilot toward waste.
Judging it too fast. Two weeks is a learning period, not a result. Reading the experiment before four weeks, or before it clears a seasonal spike, produces a confident wrong answer.
Promising a number. Google publishes several. The one tied to the DSA upgrade is that the full feature suite delivers, on average, 7 percent more conversions or conversion value at a similar cost per acquisition versus search term matching alone. Google's launch materials cite a higher typical uplift of 14 percent for activating AI Max in a Search campaign, rising to 27 percent for campaigns mostly built on exact and phrase keywords. Those bigger numbers measure a different baseline, and the 14 percent figure explicitly excludes retail advertisers. Pick the figure that matches the campaign in front of you, and do not hand a client the optimistic one.
Here is the honest reality, and it is the reason measurement discipline is the whole game. AI Max is not uniformly good or uniformly bad. The results genuinely split. An analysis of more than 250 retail campaigns by Smarter Ecommerce, reported by Search Engine Land, found median revenue up 13 percent but median cost per acquisition also up 16 percent, with the ROAS outcomes spread from a 35 percent loss to a 42 percent gain. Same feature, opposite results, depending on the account. Because Google's 14 percent benchmark excludes retail, an ecommerce account should weight that independent number more heavily. Practitioner testing finds AI Max does better on campaigns already heavy with phrase and exact keywords and worse where the structure is loose.
You cannot know in advance which side your account lands on. That is why the playbook is a playbook and not a recommendation to adopt or refuse. Test on a subset. Hold a controlled comparison. Set the brand, URL, location, and negative controls before you commit. Read the search term report like it is your job, because now it is. Judge the system on the experiment delta, not the dashboard. Do that, and the September deadline becomes a date on a calendar instead of a surprise in a report.
Council summary
This post argues that the AI Max migration is a project to be run, not a deadline to be survived: move before the September 2026 auto-upgrade so you, not Google's defaults, decide which features run and at what level. It is concrete where it counts, with a working method (test on representative campaigns inside an AI Max experiment, hold a control arm, read the search term report from day one) and a clear-eyed tour of the controls that still exist, from brand exclusions to URL rules to the new AI Brief. Its sharpest point is a measurement one: AI Max's reported conversions bundle traffic your exact and phrase keywords already won, so the only honest read of lift is the experiment delta, not the dashboard total. The reader's takeaway is durable: you cannot know in advance whether your account will land on the 42 percent ROAS gain or the 35 percent loss, so the discipline of staging, instrumenting, and judging on incrementality is the migration.
Comments