On April 15, 2026, Google announced the mandatory retirement of Dynamic Search Ads, Automatically Created Assets, and campaign-level broad match — three legacy configurations that collectively represent the last remaining points where human intent (expressed through keywords, landing page selections, and asset guidelines) could meaningfully constrain what Google's systems did with an advertiser's money. The migration to AI Max for Search is not optional for the affected campaigns. Google confirmed: there is no formal opt-out for eligible campaigns. Those who do not migrate voluntarily will be auto-upgraded starting September, with the process expected to complete by month's end.
The official framing is an upgrade. Google's data claims 7 percent more conversions at similar CPA for advertisers using the full AI Max feature suite, rising to 14 percent for non-retail advertisers. The company calls it a natural evolution — Dynamic Search Ads, introduced in 2011 as Google's first major keywordless search format, was built on page-crawl architecture that matched user queries against a static index of advertiser landing pages. It worked when queries were short and intent was obvious. It does not work when queries run twenty words long because AI Overviews taught users to ask questions instead of keywords.
This is an accurate summary of the technical gap. What the official framing omits is what independent testing reveals about the upgrade's actual performance distribution — and what it means that Google is now the sole decision-maker in a market that spent fifteen years accommodating human input as a check on platform optimization.
Dynamic Search Ads, at its core, was a crawling system. Googlebot would index an advertiser's website, generate thematic categories from the site's structure, and dynamically produce headlines based on each page's HTML title tags. Advertisers supplied the description lines and selected which categories or page feeds to target. The system was effective for large, fast-changing inventories — retailers with thousands of SKUs, real estate listings, travel packages — where maintaining keyword campaigns was operationally impossible. Early DSA beta results showed 5 to 10 percent conversion lifts; one documented case study (ApartmentHomeLiving.com) reported materially higher conversion efficiency compared to keyword-based alternatives, though the specific metric requires source verification.
What DSA was not, by design, was adaptive. It matched queries to pages. It did not incorporate real-time behavioral signals, cross-session learning, or auction context. It was a lookup function dressed as an ad product.
AI Max replaces the lookup with a prediction engine. It still ingests website content, but it layers over it Google's full complement of real-time intent signals — what users searched in the previous session, what they clicked in the one before that, what device they're on, what time of day it is, what auction is actually happening right now. The headline generation is handled by text customization models that write copy specific to each query rather than pulling from a pre-written set. The landing page matching — what DSA called category targeting — is replaced by Final URL Expansion, which can route a user to any page on the advertiser's domain, including pages that were never intended to be landing pages.
The three features that carry over (or are enabled by default) for migrating campaigns: search term matching (broadened reach), text customization (AI-generated asset optimization), and Final URL Expansion (dynamic landing page routing). Advertisers who wait until September get all three, simultaneously, with Google's defaults — not their own.
Google's published figures — 7 percent more conversions, 14 percent for non-retail — come from internal data comparing AI Max with full features against AI Max with partial features. The baseline is not legacy DSA. The comparison is not what the migrating advertiser was actually running.
Independent research paints a more ragged picture. A Smarter Ecommerce analysis of more than 250 retail campaigns found a median revenue uplift of 13 percent — genuine new volume — accompanied by a median CPA increase of 16 percent. The two numbers sound contradictory only until you notice they aren't: AI Max is finding additional conversions at a higher cost-per-conversion than the campaigns it replaced. The median outcome is more spend producing more conversions at worse efficiency. The ROAS spread across studied campaigns ranged from plus-42 percent to minus-35 percent. The median was not the mean; the distribution was wide.
A four-month controlled test conducted by an independent practitioner compared AI Max against phrase match within the same account. AI Max averaged $100.37 per conversion. Phrase match averaged $43.97. The same product, the same auction, the same budget allocation — the algorithm premium was 128 percent.
One agency's analysis of 30,000 search terms found that 99 percent of AI Max impressions generated zero conversions. An unspecified but documented account saw AI Max expand into competitor brand terms, consuming 69 percent of total Search impressions — not because the advertiser targeted those terms, but because Final URL Expansion interpreted brand-adjacent queries as inventory opportunities. Brand controls exist in AI Max, but they must be explicitly configured, and the default configuration does not include them.
These are not representative samples in the academic sense. They are also not disputed by Google, which has not published independent research rebutting them.
The structural issue is broad match. AI Max enables search term matching by default, which expands query coverage using broad match semantics even when no keywords are present. The feature is designed to find queries that existing keywords miss. In practice, independent audits found that up to 63 percent of AI Max's "new" queries were already covered by existing exact or phrase match keywords — the expansion was not finding incremental intent, it was recycling the existing pool under different attribution.
The mechanism is straightforward: when AI Max serves an ad on a query that matches an existing keyword, it receives the conversion credit even if the original keyword would have served the same ad. The conversion moves from the keyword column to the AI Max column. The advertiser's historical performance data becomes unreliable for the purpose it was designed for — understanding which inputs actually produced outcomes.
Search term reports that once contained hundreds of rows now contain tens of thousands. The audit surface for waste expands by an order of magnitude. Agencies managing multiple accounts describe the reporting load as operationally unsustainable without automated parsing tools.
The Search Partner Network adds a compounding problem. AI Max distributes across partners by default when enabled. One documented case showed 500,000 monthly impressions on partners at a 0.07 percent conversion rate — against a 3.04 percent conversion rate on Google Search for the same campaigns. Partners generated volume; partners did not generate efficiency.
Google gave advertisers approximately five and a half months between the April announcement and the September auto-migration. This is 40 percent shorter than the window Google provided for the Smart Shopping to Performance Max migration in 2022, which itself was widely criticized as compressed. Agencies that waited for the deadline on PMax received Google's default configurations — and spent the following months untangling misfires.
The advertisers who will be least harmed are those already running parallel tests, gathering their own performance baselines, and preparing controlled migrations on their own timeline before September. The advertisers who will be most harmed are those who discover in October that Final URL Expansion routed significant spend to landing pages that were never reviewed for brand safety or conversion quality — because no one had time to audit them before the switch flipped.
Google's official guidance is to migrate voluntarily. The practical effect of this advice is to move the deadline forward, not to change the migration's terms. The voluntary path lets the advertiser choose which features to enable and in what order. The mandatory path gives them Google's defaults. The incentive structure is not neutral.
The $243 billion figure is Google's 2026 search ad revenue projection. It is not in dispute. What passes through AI Max — and what Google extracts in the process — depends on the interaction between the algorithm's optimization behavior and advertiser incentives.
Academic research complicates the assumption that more automation is uniformly efficient for the market. A 2025 Journal of Marketing Analytics study found that automated bidding does not improve average keyword efficiency compared to manual bidding; manual bidding rewards keywords more productive of transactions, revenue, and clicks. A 2025 Yale Cowles working paper found that more sophisticated non-uniform bid-scaling — exactly the kind of per-auction optimization Smart Bidding implements — leads to lower aggregated welfare in first-price auction markets. The mechanism: better autobidders implement finer-grained strategies that fragment the auction, reducing competitive pressure on each individual bid and enabling platforms to extract higher CPMs without improving advertiser outcomes.
Performance Max, Google's other major automation product, provides a relevant precedent. It peaked at 82 percent of Google ad spend share in May 2024. By early 2025 it had declined roughly 6 percentage points as advertisers rediscovered that removing human visibility into where ads actually served — the defining characteristic of PMax — made optimization at scale genuinely difficult. Enhanced CPC, a hybrid manual-automatic bidding strategy, was deprecated in March 2025. The direction of travel has been consistent: remove intermediate controls, force advertisers toward the extremes, call it evolution.
AI Max is the latest crossing. The September deadline does not pause for the learning curve.
The search advertising market spent fifteen years developing a shared vocabulary between human intent and platform execution. Keywords were imperfect but legible. Negative keyword lists were blunt but comprehensible. DSA was opaque by comparison but still rooted in site content that advertisers could audit. The transition to AI Max replaces that vocabulary with a model. The model is not published. Its decisions are not fully explainable. Its optimization target is declared (conversions at CPA/ROAS) but its path to achieving that target is not controllable in the way keyword campaigns were controllable.
For Google, the migration completes a transition that has been underway since at least 2022: the shift of search advertising from a channel where advertisers expressed intent and platforms served matching supply, to a channel where platforms interpret intent, generate inventory, allocate budget, and report outcomes — all within systems the advertiser can influence but not direct.
The advertisers who will do well in AI Max are those with the largest datasets, the most history with Google's systems, and the smallest need to explain to a CFO why the CPA is what it is. The advertisers who will struggle are those in categories where the model's incentive to find volume diverges from the advertiser's incentive to find efficient conversions — which is to say, most of them.
The September deadline is real. The algorithm is already moving.