Google Ads in 2026: What Actually Changed

2026 is the year the Google Ads interface stopped being the product. The account — its data, its economics, its levers — is increasingly operated through AI layers: Google's own matching (AI Max), Google's own surfaces (AI Overviews, AI Mode), Google's own agent (Ads Advisor, now Ask Advisor), and the open protocol that lets any agent in (MCP). Here is what actually changed, with dates, and what each change asks of the person still responsible for the budget.

The year in one table

When	What changed
Oct 2025	Google open-sources a read-only Google Ads API MCP server
Dec 2025	Ads Advisor reaches full English-language availability
Feb 2026	Amazon ships an Ads MCP server in open beta
Apr 15, 2026	AI Max for Search goes GA after 11 months of beta
Apr 29, 2026	Meta launches hosted Ads MCP connectors in open beta
May 13, 2026	TikTok announces its Ads MCP server at TikTok World
May 2026	Google Marketing Live: Ask Advisor unifies Google's agent surface
Jun 15, 2026	DSA creation restored; voluntary migration window opens
Sep 2026	Auto-created assets and campaign-level broad match fold into AI Max
Jan 2027	DSA creation removed
Feb 2027	Remaining DSA campaigns begin auto-upgrading

AI Max leaves beta, and the DSA clock starts

AI Max for Search reached general availability on 15 April 2026, after eleven months of open beta. The suite bundles three things onto a standard Search campaign: matching that reaches beyond your keyword list, generated headline variants steered by text guidelines, and final URL expansion that can route clicks to pages you did not designate. Each is individually controllable; together they amount to Google asking for the matching decisions you used to encode in keywords.

Google's GA-era claim is an average of 7% more conversions or conversion value at similar CPA/ROAS for the full feature suite versus its matching alone — internal data, retail excluded, no sample size disclosed. Independent analyses have been harsher; Monks found 99% of impressions in a roughly 30,000-term AI Max sample generated zero conversions. Both numbers can be true at once, which tells you the feature's variance is the story: outcomes depend on the account's conversion data and on whether anyone polices the expansion. The full case for and against sits in our AI Max opt-in decision guide.

The forcing function is Dynamic Search Ads. Google announced DSAs will upgrade into AI Max, then — after advertiser pushback — pushed the automatic migration from September 2026 to February 2027, restoring DSA creation from 15 June 2026 for the interim. Two smaller transitions kept their September 2026 date: automatically created assets and the campaign-level broad match setting both fold into AI Max. If you run DSAs, the migration plan is now a scheduled project, not a think-piece.

Ads moved into AI Overviews and AI Mode

Google now shows Search and Shopping ads above, below or inside AI Overviews, on mobile and desktop, in twelve English-language countries. Serving considers both the user's query and the content of the AI Overview itself — a genuinely new rule, because your ad can be matched to an answer, not just a question. Ads in AI Mode are live in US testing, matched on conversational context that evolves across a session.

Two operator-relevant facts follow. First, there is no opt-in: eligibility flows from existing Search, Shopping and Performance Max campaigns automatically. Second, there is no clean measurement yet — AI Overview placements are folded into standard Search reporting rather than broken out. Third-party tracking suggests ads now appear in roughly a quarter of AI Overview results, up sharply through 2025.

Until segmentation arrives, the honest measurement posture is inferential: watch for query mixes drifting longer and more conversational in the search terms report, watch top-of-page share on affected campaigns, and annotate the account each time Google announces an eligibility expansion so later analysis has landmarks. What you can control is readiness, which is an account-hygiene problem — covered in how ads show up in AI search and the ads-in-AI-Overviews glossary entry.

Google's own agent arrived in the interface

Ads Advisor — announced at Google Marketing Live 2025 — reached full English-language availability in December 2025 and gained agentic safety features in April 2026. At GML 2026, Google folded it into Ask Advisor, a Gemini-based agent spanning Google Ads, Analytics and Merchant Center: it diagnoses performance, proposes changes, and can build campaigns from your Merchant Center data, rolling out through late 2026.

An agent inside the product is genuinely useful, and its incentives are genuinely Google's. It optimises toward Google's definition of a healthy account — the same definition behind optimisation score and auto-apply recommendations — and it works only inside Google's walls: it cannot join your ads data to your CRM, your margins or your other platforms, and its suggestions inherit the seller's view of what an account should buy. The practical question for 2026 is not whether to use an agent but which agents hold which jobs: Google's advisor for in-product suggestions and navigation, and an independent agent across your own stack for the analysis and writes you want aligned to your economics, in the AI client your team already uses. The two are not exclusive — but only one of them works for you in the strict sense of the word.

Every major ad platform shipped an MCP server

The protocol argument ended this year. Google open-sourced a read-only Google Ads API MCP server in October 2025 — three tools: list accounts, run GAQL, read metadata. Amazon followed in February 2026, Meta in April with hosted connectors, TikTok in May. Four platforms, four official answers to the same question: how should AI agents talk to an ad account?

The answers differ in exactly the way you would predict from incentives. Google's server reads but never writes; Meta ships writes with a paused-by-default guardrail; TikTok pitches full campaign lifecycle. What none of them provides for Google Ads is the part between reading and regretting: write tools with an approval prompt, an audit trail, scoping, and no ability to delete — which is the layer a hosted Google Ads MCP connector exists to be. The full breakdown, with a capability table, is in our analysis of the official ad-platform servers.

What quietly deprecated this year

The removals matter as much as the launches, because each one deletes a control surface someone's account depends on:

• Dynamic Search Ads — creation removed January 2027, auto-upgrade from February 2027. The dated sunset of the year.
• Automatically created assets — folds into AI Max text customisation in September 2026. If ACA is on anywhere, that behaviour moves under a new toggle with new reporting.
• Campaign-level broad match setting — same September 2026 fold-in. The keyword-level match decision survives; the campaign-wide switch does not.

None of these breaks an account on its own. All three move matching authority from settings you chose once toward systems that expand on their own — which raises the value of the one control Google never removes: your negatives, your budgets, and your attention.

What did not change

A list of everything new flatters the new, so state the constants. The auction still prices what your data and your competitors' bids say it is worth — no 2026 feature exempts anyone from auction economics. Conversion tracking still decides whether every automated system above it optimises toward truth or noise; most "AI Max failed" stories are still tracking stories underneath. The search terms report is still where reality shows up first, whatever generated the match. And accountability did not move: every new surface and agent ships with the same fine print, which is that the budget owner answers for the outcome. The operators having a calm 2026 are the ones who noticed that the fundamentals list did not shrink — the interface around it did.

The operator's 2026 checklist

Six actions cover the year's changes for a working account:

1. Date your DSA exit. Inventory DSA dependencies now; migrate inside the voluntary window rather than in February's queue.
2. Decide AI Max per campaign, not per account. Fence the first test with hard negatives and a weekly query review.
3. Check ACA and campaign-level broad match toggles before September moves them for you.
4. Assume AI-surface serving. Audit assets and landing pages as if a machine will quote them mid-answer — because one will.
5. Assign the agents. Let Google's advisor advise inside the product; keep analysis and writes with an agent that answers to you, with approval and audit built in.
6. Re-read your reports. AI-surface placements are folded into Search metrics; segment what you can and annotate what you cannot.

Each item on that list is an afternoon, not a programme — and every one of them is smaller with an agent connected to the account doing the reading: the DSA inventory, the toggle census, the asset audit and the annotation habit are all prompts before they are projects.

The thread through all of it: Google is automating the interface faster than it is automating accountability. Matching expands itself, agents propose on both sides of the aisle, ads serve on surfaces without their own reporting row — and the budget still has exactly one owner. The operator's job in 2026 is to hold the second while adopting the first on your own terms.