AI Max, Performance Max and Google Ads in 2026: What to Hand the Machine, and What to Keep

Google spent the last few years quietly moving the steering wheel. In 2026 it has taken both hands. Between AI Max for Search, a more mature Performance Max, Demand Gen sitting in the middle of the funnel, and a Smart Bidding change landing on 17 August, the platform now expects you to hand it more of the day-to-day work than ever before. That is not automatically a bad thing. Handled well, automation does more of the tedious grind and frees you to think. Handled badly, it quietly spends your budget on queries you would never have chosen and calls it progress.

This is a clear-eyed guide to what is worth handing the machine, and what you should keep a firm grip on. It is written for people who actually run the account — whether that is you, your marketing lead, or an agency — and it assumes you would rather understand the levers than be told to “trust the AI”. If some of the terminology is new, the the marketing glossary covers the acronyms as we go.

AI Max for Search: the DSA successor, with guardrails

AI Max for Search is Google's next-generation search product and, in practical terms, the replacement for Dynamic Search Ads. If you run DSAs today, this affects you directly. Google has said voluntary upgrades are available through the summer, with automatic upgrades of eligible campaigns beginning in September 2026 and new DSA creation ending around the same time. Treat that as a plan-ahead date rather than a surprise.

The core idea is that AI Max combines your inputs with broader signals to find queries you did not explicitly target. Three features do the heavy lifting:

• Dynamic text customisation — Google generates additional headlines and descriptions, drawing on your landing pages and existing assets, to match the ad copy to the query.
• Final URL expansion — the system can send traffic to the most relevant page on your site rather than only the URL you specified.
• Broader keyword and search-term matching — it reaches beyond your keyword list to related queries it judges relevant.

Google's own figures suggest campaigns using the full feature set see roughly 7% more conversions or conversion value at comparable cost, versus search-term matching alone. That is Google's number, measured Google's way, so read it as directional. The interesting question is not whether it can find more volume — it can — but whether that volume is the volume you want.

How to keep control

AI Max is more controllable than Performance Max, and the controls are the whole point. Use them from day one rather than bolting them on after the spend has drifted.

• Negative keywords are your main defence. Unlike Performance Max historically, AI Max fully supports negatives at campaign and ad group level. Build comprehensive lists before you switch on broader matching — competitor brands, low-intent and research queries, job-seekers, and irrelevant product variations. This matters far more in B2B, where a “free template” or “salary” query can burn budget fast.
• Keep your own branded and proven assets in place. You can supply your manually written responsive search ad assets, and if the generated copy strays off-message you can disable text customisation entirely and run on your assets alone. Where text guidelines and term exclusions are available in your account, use them to fence off claims, competitor names and off-brand phrasing.
• Control URLs deliberately. Set URL inclusions and exclusions at ad group level so URL expansion cannot send a high-intent click to a blog post or the wrong regional page. Multi-location and multi-market accounts should watch this closely — geographic mismatches between the campaign's target and the landing page have been a recurring complaint.
• Lean on tight account and site structure. AI Max is only as sensible as the signals you feed it. Clean ad groups built around genuine intent, and landing pages that clearly match those intents, keep the automation on the rails. Loose structure is where broad matching goes wrong.
• Mind brand safety. Review search-term reports in the first fortnight aggressively and add negatives as irrelevant queries appear. This is not a set-and-forget switch.

A starting negative-keyword list for a B2B account might look like this:

Campaign-level negatives (broad match):
free
freeware
template
templates
jobs
salary
salaries
course
courses
training
"how to" diy
cheap
reddit

Performance Max in 2026: less black box, more dashboard

Performance Max has matured. The main complaint — that it was an opaque box you fed budget into — has been steadily answered. In 2026 you have meaningfully more visibility and a few more genuine steering controls. If PMax is new to you, the the marketing glossary has short definitions for it and impression share.

What you can now rely on:

• Asset-group-level and channel reporting. You can see how individual asset groups perform, and segment placement data by network — useful for brand-safety visibility and for spotting when Display or video is quietly soaking up conversions that Search would have won anyway.
• Search-term insights and campaign-level negative keywords. You can see more of the queries driving performance and exclude terms at campaign level, which was the single biggest missing lever for years.
• Brand exclusions and new-customer controls. Brand exclusion lists keep PMax off your own or competitors' brand terms, and first-party customer-list exclusions let you point budget at genuine new-customer acquisition rather than re-selling to people already in your CRM.
• Better reporting all round. Budget projections, and demographic breakdowns by age range and gender, make it easier to explain what the campaign is actually doing to a sceptical finance director.

What to watch

Two failure modes are worth naming. The first is Search cannibalisation: Performance Max will happily claim conversions from branded and high-intent searches that your standard Search campaigns would have captured more cheaply, then present that as its own success. Use brand exclusions, keep dedicated Search campaigns for your core terms, and compare like for like. The second is impression share erosion on the queries that matter most — if PMax is pulling budget toward cheap, easy conversions, your visibility on the hard, valuable ones can quietly fall. Watch Search impression share and lost impression share on your priority terms, not just the blended cost per acquisition.

Demand Gen: the mid-funnel that the other two skip

AI Max works the bottom of the funnel, where people are searching with intent. Demand Gen fills the gap above it. It runs across YouTube — including Shorts — Discover and Gmail, putting visually led ads in front of people before they have started searching for you. In 2026 it has picked up shoppable connected-TV formats and better measurement of the branded searches it influences, which helps with the perennial problem of proving that upper-funnel spend does anything.

For most B2B advertisers, Demand Gen is a considered test rather than a core channel. It earns its place when you have a strong creative story, a longer buying cycle where awareness genuinely shifts pipeline, and the patience to judge it on assisted conversions and branded-search lift rather than last-click. If your Search and PMax campaigns are already capturing all the existing demand, Demand Gen is how you create more of it. If they are not, fix those first.

The 17 August 2026 Smart Bidding change

This one deserves a diary entry. From 17 August 2026, Google is changing how Smart Bidding treats budget-constrained campaigns using Target CPA or Target ROAS. Today, a budget-limited campaign often overperforms its target — a £50 Target CPA campaign might quietly deliver conversions at £35 because the budget runs out before the algorithm has to reach for more expensive conversions. After 17 August, Smart Bidding will optimise more closely toward the target you actually set, rather than sitting comfortably below it.

Google (via Ginny Marvin) has clarified what this is and is not:

• It will not automatically change your spend.
• It will not automatically change your targets.
• The stated aim is to make performance fluctuations less surprising, especially when you adjust budgets.

The practical effect: if a campaign currently beats its target and is budget-limited, it may drift up toward that target — winning more volume, but at a higher CPA (or lower ROAS) than you have been used to. If you are happy with your current efficiency, the move is to tighten your targets before the change lands — lower your Target CPA, or raise your Target ROAS, to reflect the performance you are actually getting rather than the looser number you set months ago.

That only works if your targets are grounded in real economics: margin, average order or deal value, close rate and customer lifetime value. This is exactly the kind of number worth calculating properly rather than guessing — the free calculators will help you set a Target ROAS or CPA you can defend before you let Smart Bidding run to it.

The human-in-the-loop principle

The through-line across all of this is simple: automation amplifies strategy, it does not replace it. Google's AI is very good at execution within the boundaries you set — and largely indifferent to whether those boundaries are any good. Point it at a clean account, sharp intent segmentation, targets rooted in margin, and well-tested creative, and it compounds your advantages. Point it at a messy account with vague goals, and it compounds the mess — faster and more expensively than a human ever could.

The parts you cannot hand over are the parts that were always the actual job: how the account is architected, how you separate intent and audiences, what you are willing to pay for a customer, and which messages you put in front of them. Those are decisions, not tasks. Keep them.

Automate vs keep control: a working checklist

Comfortable to automate:

• Headline and description combinations and asset rotation, within your guidelines.
• Real-time bid adjustments toward targets you have set deliberately.
• Query discovery through broader matching — fenced in by negatives.
• URL expansion — with inclusion and exclusion rules in place.
• Cross-channel delivery in Performance Max and Demand Gen, monitored by channel.

Keep firmly in human hands:

• Account and campaign architecture — the structure the AI reads as signal.
• Intent and audience segmentation — what should never be blended together.
• Target-setting — CPA and ROAS grounded in real margin, not defaults, and reviewed ahead of 17 August.
• Negative keywords, brand exclusions and brand safety — your standing defences.
• Creative strategy and testing — the message, the offer, and what “good” looks like.
• Reading the reports that matter — Search cannibalisation, impression share on priority terms, and whether volume is coming from the right queries.

Hand the machine the execution. Keep the judgement. The accounts that win in 2026 are not the ones that automate the most — they are the ones that automate the right things on top of a strategy worth amplifying.

If you would like a straight assessment of what to automate and what to keep control of in your own account — before the August change and the AI Max upgrades land — a free growth plan is a sensible place to start.