← Back to blog

AI Max in Google Ads — what Google doesn't say

Scale of AI Max adoption — and the DSA sunset

Google has been pushing AI Max since early 2026 as the "evolution" of Search campaigns. In parallel, it is sunsetting Dynamic Search Ads (DSA) — campaigns that for years did one specific thing well: routing granular queries to matching product pages. AI Max is becoming the only path to "smart matching" in Search.

In practice: every Search campaign with AI Max enabled has "smart match" unlocked — the algorithm adds its own queries based on semantic matching, contextual relevance, intent prediction. It sounds good. But there's one problem Google does not highlight: you trade transparency for convenience.

Advertisers are not buying this narrative without pushback. ppc.land documented an open backlash — agencies publicly criticise the removal of DSA and being forced into AI Max without quality guarantees. The reason is simple: DSA did one thing, and did it predictably. AI Max promises everything, delivers inconsistently.

What Google promised (and what it actually delivered)

Google's messaging is consistent across several areas:

• Wider reach — discovering new queries
• Better relevance — the algorithm matches more accurately
• Less work — the algorithm handles matching for you
• Scale — faster campaign expansion without manual keyword work
• Better outcomes — more converting traffic

What the data actually shows (based on 2026 reports):

What Google promised	What the hard numbers show
Wider reach	+20% impressions (industry data 2026)
Better relevance	Only 16% of advertisers report good performance
Less work	Reporting overload — search term reports too large for manual control
Scale (revenue)	+13% revenue (250-campaign study, almcorp)
Better outcomes (efficiency)	+16% higher CPA in the same study
Lower costs (implied)	+15–25% higher CPC in competitive niches

Conclusion: Scale = yes. Efficiency = no. In other words — if you pay per conversion, AI Max costs you more for the same outcome. If you pay for traffic alone, you'll get more, but of questionable quality.

5 things Google stays silent about

1. Black box — no transparency in algorithm decisions

Google provides search term reports and asset performance ratings. It does not provide a breakdown of how the algorithm actually decides: why this query, why this audience, why this landing page. This is a structural limitation, not a bug to be fixed. Your money goes where the algorithm decides, and you only see the outcome.

2. Broad match cannibalization

AI Max often matches queries already covered by exact match in the same campaign. The result: you pay twice for the same thing — once on exact, once on AI Max (sometimes at a higher CPC because AI Max has a lower Quality Score on that query). Attribution lands more often on AI Max (because it's the "smart match"), making exact match look worse than it really is.

3. Competitor brand hijacking

The algorithm discovers queries containing competitor brand names. Targeting "free audit [competitor]" is not a strategic decision — it's "expansion". Google doesn't ask whether you want it. Competitors see your ads on their brand search, business relationships deteriorate, and ROI on those clicks is low.

4. Granular queries → generic landing pages

DSA did this predictably: a granular query "trekking boots meindl 43" landed on a specific PDP. AI Max does it worse — often sending traffic to the category page "trekking boots". Conversion drops because the user lands on a "half-relevant" page. Google does not explain why the algorithm decides this way.

5. Attribution credit theft

AI Max takes credit for conversions that exact match already delivered. You turn AI Max off — exact match looks like it "dropped", when in reality it's just going back to what it was always delivering. This distorts scaling decisions: you think AI Max is generating value when in reality it isn't.

What the hard numbers show

A study of 250 Search campaigns with AI Max enabled vs disabled (almcorp, 2026):

• +13% revenue — scale works
• +16% higher CPA — efficiency drops
• Net ROAS (after higher ad costs) — drops on most accounts

LinkedIn poll (Adriaan Dekker, Google Ads specialist, March 2026):

• Only 16% of advertisers report "good performance" with AI Max
• The other 84% — neutral or worse results

This is an industry poll (self-selection bias, sample of a few hundred PPC specialists), not peer-reviewed research. The number shows sentiment direction, not a hard benchmark — but it lines up with other observations in the sector.

Additionally, from industry reports (Asthatechnology 2026, Lunio): +20% impressions, but +15–25% higher CPC in competitive niches.

What this means in practice depends on account size:

Budget tiers below are in PLN (Polish zloty). Rough USD/EUR equivalent: divide by ~4.

Account size	Actual AI Max impact
Small budget (< 3,000 PLN/month ≈ $750)	The algorithm burns through money in 2 weeks before it "learns"
Mid (3,000–15,000 PLN/month ≈ $750–$3,750)	Traffic growth visible, but efficiency drops. Whether it's worth it depends on what you measure
Large (15,000+ PLN/month ≈ $3,750+)	Scale carries more value, data stabilizes faster. But still +16% CPA without transparency

When AI Max is NOT for you

Turn off or don't enable AI Max when:

1. You have a small budget (< 3,000 PLN/month ≈ $750) — the algorithm doesn't have enough data to learn fast. You'll pay for its training.
2. You have a strong keyword structure with good exact / phrase match — AI Max will cannibalize it, not add value.
3. Regulated industry (finance, healthcare, legal) — AI Max generates matches that can take your ads out of compliance. Legal audit becomes a problem.
4. High-ticket B2B — the algorithm understands "CTO audit" vs "google ads audit" less well — it lets marketing-intent traffic onto a sales-intent budget.
5. Specific PDPs in e-commerce — DSA did query → PDP routing better. AI Max sends more often to category pages, conversion drops.
6. Limited LP infrastructure — you have 5 landing pages, not 50. AI Max has nowhere to route wider traffic.
7. Tracking not buttoned up — without Enhanced Conversions / OCI you don't know what converts. AI Max learns on weak signal, burns budget.

What to do instead (or in parallel)

If you go into AI Max — go in consciously. How to set up AI Max and its controls step by step, before you turn it on, is covered in my article How AI Max in Google Ads works and how to set it up. A practical playbook:

1. Strong exact + phrase match base first. Without this there's no point discussing AI Max — your fundamentals are weak, the algorithm will cannibalize everything.
2. Enable AI Max on one test campaign (not all at once). Give the algorithm 30 days, compare CPC / CTR / conv rate / CPA vs an identical campaign without AI Max. If CPA grows > 15% without ROAS growth — turn it off. Note: 30 days is the minimum sensible learning window, but with a budget < 3,000 PLN/month (~$750, see table above) the training cost may exceed the gain from scale. The realistic threshold for this test to make sense is around 5,000 PLN/month (~$1,250).
3. Daily search terms review during the first 4 weeks. Exclude fast — before the algorithm cements bad matches into its learning history.
4. Global negative keyword lists — categorical exclusions (competitors, "free", "job", "career", informational queries) before enabling AI Max. Otherwise you start on the back foot.
5. Monitor CPA per network / match type — not just overall campaign CPA. AI Max can inflate CPA at stable headline numbers because scale masks efficiency loss in segments.
6. Keep DSA as long as Google allows. DSA is still better for granular query → PDP routing. When Google turns DSA off on your account — that's when you migrate. Not earlier.
7. Monthly audit from the perspective "what is Google deciding for me". The question is not "does AI Max work" — it's "how much transparency am I losing and is it measurable value". For most accounts I observe — it is not.

About the sources: Article based on almcorp's 250-campaign study (2026), Lunio analysis, Asthatechnology — AI Max 2026 reality, ppc.land — DSA sunset backlash, groas.ai — Hidden Risks of AI Max + observations from JP Ads client accounts. Numbers from my practice are labelled "from my experience" — distinguish them from hard research data.