Google Ads AI Max for Search: The 2026 Setup Guide for UK Advertisers

Google Ads AI Max is a settings bundle inside existing Search campaigns that adds keywordless query matching, AI-written ad text and automatic landing page selection. Eligible UK campaigns migrate automatically from 1 September 2026. Conversion-based Smart Bidding is required, and the brand, term and URL controls stay yours.

Most UK advertisers found out about Google Ads AI Max the same way: an in-account notice, a vague promise of more conversions, and no clear sense of what they were agreeing to. That vagueness is now a deadline problem. From 1 September 2026 Google begins migrating eligible Search campaigns to AI Max whether or not anyone in your account clicks accept. If you run campaign-level broad match or automatically created assets, that includes you, and there is no tick box marked “no thanks”. Left alone, the change can quietly rewrite your ad copy, send paid traffic to pages you never nominated, and push spend into queries you have spent years excluding. Handled properly, it reaches genuine demand that exact-match keyword lists were never going to find. This guide covers what AI Max actually does, which controls survive the upgrade, how to configure it setting by setting, and how to test it on a modest UK budget before it tests itself on you.

What AI Max actually is, and how it differs from Performance Max

The naming of Google Ads AI Max has caused more confusion than any single feature. AI Max sounds like a sibling of Performance Max, and plenty of UK advertisers have assumed it works the same way. It does not. AI Max is not a campaign type. It is a set of switches you turn on inside a Search campaign you already have, and you can turn most of them off again. Performance Max, by contrast, is a distinct campaign type that takes over placement across Search, Display, YouTube, Gmail, Discover and Maps, and gives you an asset group instead of ad groups. That distinction decides how much visibility you keep. Your Search campaign keeps its ad groups, its keywords, its negative keyword lists and its search terms report. A Performance Max campaign gives you none of those in the same form.

The practical differences that matter to a media buyer

Three differences change day-to-day management more than anything else in the feature list. Inventory is the first. AI Max stays on the Search network and its partners, so a campaign you built for search intent keeps serving against search intent. Performance Max spreads budget across channels according to where Google’s models expect conversions, which is why so many UK accounts have watched Performance Max spend drift towards cheap Display inventory. Structure is the second. Keywords still exist under AI Max, still carry match types, and still report performance individually. AI Max layers keywordless matching on top of them rather than replacing them, so the account structure you have spent time building survives. Reporting is the third, and it is where AI Max is meaningfully better. You keep a search terms report. Google has also added reporting that shows which queries arrived through keywordless matching rather than your keywords, which is the single most useful thing for judging whether the feature is earning its place.

Where the two overlap

Both rely on conversion-based Smart Bidding, and both get worse when your conversion data is thin or badly configured. Both use generative AI to assemble creative from your existing assets and landing pages. Both let Google choose a landing page for you if you allow it. If your conversion tracking is unreliable, neither will work, and AI Max will fail faster because it is spending against search intent at search prices. Fixing measurement before switching anything on is not a nice-to-have. It is the whole basis on which the system makes decisions. Teams running both should also watch for overlap. A Performance Max campaign and an AI Max Search campaign in the same account will compete for the same queries, and Google’s ranking rules decide which one serves. Auditing that overlap before September is worth an hour of anyone’s time. Our Google Ads management work starts there for exactly that reason.

The September 2026 auto-migration, and what happens if you do nothing

This is the part of the story with a date attached, and it is close. Google will run automatic migration of eligible Search campaigns to AI Max through September 2026, covering campaigns that use campaign-level broad match or automatically created assets. Search term matching and text customisation come switched on. There is no formal opt-out mechanism. That last point has been widely misreported, so it is worth stating precisely. You cannot decline the migration, but you can make a campaign ineligible for it. Those are different things, and the second one is entirely within your control.

How to make a campaign ineligible before 1 September

Eligibility hangs on two settings. Remove either one before the migration window and the campaign is not swept up. The first is campaign-level broad match. If a campaign has it enabled and you switch it off, reverting to keyword-level match types, the campaign no longer qualifies. The second is automatically created assets. Disabling ACA at campaign level has the same effect. Neither of those is a free action. Campaign-level broad match is doing work in most accounts that use it, and turning it off to dodge a migration may cost more volume than the migration ever would. Treat this as a deliberate decision with a trade-off, not a default defensive move.

What Dynamic Search Ads advertisers need to know

Dynamic Search Ads were originally scheduled to fold into AI Max in September alongside everything else. Google pushed that back to February 2027, roughly five months later than first announced. During September you will see in-account prompts encouraging a voluntary DSA upgrade. Those are invitations, not migrations. A DSA campaign left alone will keep running as a DSA campaign into 2027. The extra runway is genuinely useful. DSA advertisers rely on Google crawling and matching against site content, which is the closest existing analogue to Final URL expansion, so they have the most to gain from testing the transition deliberately rather than being moved mid-quarter.

A twelve-day checklist

For UK accounts reading this in the second half of August, the useful work is small and specific.

1. List every Search campaign using campaign-level broad match or automatically created assets. Those are your migration candidates.
2. Export 90 days of search terms for each one. After migration you will want a clean before-and-after, and reconstructing it later is painful.
3. Check that every campaign in scope uses a conversion-based Smart Bidding strategy. Campaigns on manual CPC or maximise clicks will not get the intended behaviour from AI Max features.
4. Screenshot or export current ad copy. Text customisation can rewrite headlines and descriptions, and having the original to compare against makes the first performance review far easier.
5. Decide, campaign by campaign, whether you want the migration or want to make the campaign ineligible. Write the decision down with the reason.

Warning: Accounts in regulated sectors should treat step 5 as urgent rather than routine. Financial services, legal and healthcare advertisers face compliance exposure from AI-generated ad copy that a consumer retailer simply does not, and the controls that manage that risk need configuring before migration, not after.

The first week after migration: a triage routine

Preparation is one job and the week after the switch is another. Accounts that handle the second badly lose more money than accounts that skipped the first.

Day one: confirm what actually changed

Open each migrated campaign and read the settings rather than assuming the migration did what the notice described. Check which components are live. Search term matching and text customisation come switched on. Confirm whether Final URL expansion was also enabled, because that is the one with the widest range of outcomes and you want to know before it has been running a fortnight. Check your bidding strategy survived unchanged. Migrations occasionally coincide with other automated recommendations being applied, and separating the two afterwards is difficult. Note the date and time in a shared document. Every performance comparison you make for the next two months will reference that timestamp, and reconstructing it from change history is slower than writing it down now.

Days one to three: watch queries, not costs

Cost data in the first three days tells you almost nothing, because Smart Bidding is relearning against a wider pool. Query data tells you a great deal immediately. Pull the search terms report daily and filter to keywordless matches. You are looking for three categories: competitor brand terms, queries that are clearly irrelevant to your business, and genuinely useful queries your keywords would never have caught. Add negatives the same day you spot them. A negative added on day two saves budget across the whole month, while the same negative added on day twenty has already let the spend happen. Resist the urge to judge performance. Teams that pull a migration in the first three days are almost always reacting to the learning period rather than to a real problem.

Days three to seven: read the generated copy

If text customisation is live, your ads are no longer the ads you wrote. Export the generated headlines and descriptions and read every line. Three things to check for. Claims your business cannot substantiate, which for regulated advertisers is the material risk. Tone that does not match how you talk about yourself. Factual errors about your products, services or coverage area. Where you find problems, tighten your text guidelines and add the offending language to term exclusions rather than switching the whole feature off. Switching text customisation off also disables Final URL expansion, which may not be what you want. Keep the export. Comparing week one against week four shows you whether the generated copy is converging on something sensible or drifting further from your brand.

Days three to seven: check where the traffic landed

Open the landing page report and sort by cost rather than by clicks or conversions. Any page taking meaningful spend that you did not nominate is Final URL expansion at work. Some of those will be sensible choices you had not thought of. Others will be a careers page or an old blog post. For anything unwanted, the immediate lever is switching Final URL expansion off. The longer-term fix is removing those pages from the index or improving them, and that decision belongs to whoever owns the site rather than whoever owns the ad account.

End of week one: the go or no-go conversation

By day seven you have enough to make one decision, and it is not whether performance improved. The question is whether the query pool looks like demand you want. If the new queries are relevant and your negatives are holding, continue and let the campaign learn properly. If the query pool is dominated by competitor terms or irrelevant traffic that your exclusions are not catching, intervene now rather than waiting for the four week mark. Judge week one on query relevance and copy quality. Judge week four on cost. Getting those two the wrong way round is the most common reason a migration gets reversed before anyone knows whether it worked.

What to tell whoever signs off the budget

Set the expectation before the numbers arrive, because a surprised finance director makes worse decisions than a briefed one. Tell them cost per acquisition will probably rise for one to two weeks. Tell them the meaningful read comes at day forty-five. Tell them what you will do if it has not recovered by then. Writing that down in advance also protects you from your own optimism later, when a mediocre result is easier to explain away than to act on.

The three components that make up AI Max

AI Max is usually described as one feature. It is three, they behave differently, and they are not fully independent of each other.

Search term matching

Search term matching adds what Google calls keywordless technology alongside your existing keywords. In practice the campaign becomes eligible for queries that no keyword in your account would have matched, chosen by Google’s models based on your landing pages, existing assets and conversion history. Enabling AI Max switches this on at campaign level. You can toggle it off per ad group, which is more useful than it sounds, because it lets you run a controlled comparison inside a single campaign rather than across two. Your negative keywords still apply. This is the reassurance most advertisers need and the one most often left out of the vendor guides. Query expansion happens within your exclusions, not around them.

Text customisation

Text customisation uses your existing ad text, your landing page copy and generative AI to assemble headlines and descriptions matched to the individual query. Ads become dynamic rather than fixed. For a retailer with hundreds of near-identical product pages, this solves a real problem at a scale no copywriter would want to handle manually. For a professional services firm whose ad copy has been through a compliance review, it introduces language nobody has approved. The output is constrained by the text guidelines and term exclusions covered later in this guide, and those constraints are the difference between a useful feature and a liability.

Final URL expansion

Final URL expansion lets Google send a click to whichever page on your site it judges the best match for the query, overriding the final URL set on the ad. This is the component with the most upside and the most risk. A site with deep, well-structured content gets clicks routed to genuinely better pages. A site with thin category pages, orphaned legacy content or an unmaintained blog archive gets paid traffic delivered to pages that were never built to convert. There is a dependency worth committing to memory. Final URL expansion requires text customisation to be enabled, and switching text customisation off switches Final URL expansion off with it. You cannot run URL expansion with your original locked ad copy, which rules out the combination that cautious advertisers usually want first.

The Smart Bidding requirement

All three components need a conversion-based Smart Bidding strategy to function as designed. Target CPA, Target ROAS, maximise conversions and maximise conversion value all qualify. Manual CPC and maximise clicks do not. The reasoning is straightforward. Keywordless matching produces a far wider query pool, and the only thing preventing that pool from draining budget into irrelevant traffic is a bidding system that can price each query by its likelihood of converting. Accounts with weak conversion signals get the wide query pool without the mechanism that makes it safe. That combination is the single most common way AI Max loses money, and it is worth reading twice before switching anything on.

Google’s published uplift figures, and what independent testing found

Google’s headline claim is specific. Advertisers who activate AI Max in Search campaigns typically see around 14% more conversions or conversion value at a similar CPA or ROAS, rising to roughly 27% for campaigns still relying mostly on exact and phrase match keywords. The full announcement sits on Google’s own ads blog. Those numbers are plausible and worth taking seriously. They are also averages published by the company selling the feature, drawn from a self-selecting group of advertisers who chose to switch it on.

What independent analysis shows

The picture from the wider industry is more mixed, and more useful for planning. An analysis of more than 250 retail campaigns found a median revenue lift of about 13%, close to Google’s figure, alongside a median CPA increase of roughly 16%. More revenue at a higher cost per acquisition is a real outcome, and whether it is a good one depends entirely on your margin. The spread matters more than the median. ROAS outcomes across that dataset ran from roughly +42% at the top to about -35% at the bottom. Averages hide the fact that AI Max is a high-variance change, and the variance is not random. It tracks account quality. A separate review of 23 documented tests, written up by Search Engine Land, found an average uplift of about 7% in conversion value from queries the campaign had never previously targeted. That is the mechanism working exactly as described: incremental demand, not cannibalised demand. Practitioner sentiment runs colder. One poll of working PPC professionals found only about 16% reporting good performance, with the remaining 84% neutral or negative. Sentiment polls are not performance data, and early adopters test on their hardest accounts, but a gap that wide between vendor claims and operator experience is worth respecting.

The failure modes behind the bad results

Three structural problems account for most of the disappointing outcomes, and all three are diagnosable before you start. Broad match cannibalisation appeared in up to 63% of cases in one analysis. The campaign takes credit for conversions it would have won anyway through existing keywords, so reported performance improves while incremental revenue does not move. Competitor brand hijacking is the second. In one documented case, competitor brand queries consumed 69% of a campaign’s Search impressions after AI Max was enabled. Without brand exclusions configured, keywordless matching will find competitor terms because they look like high-intent queries to a model that does not know your commercial position. Reporting overload is the third and the most avoidable. Query volume expands sharply, and teams that reviewed search terms monthly find that cadence no longer works.

How to test the claims on a small UK budget

Most UK SMEs are not running the six-figure monthly budgets that produce clean test data in three weeks. The approach has to fit the budget. Pick your second-best campaign, not your best. The best campaign carries too much revenue risk, and the worst is usually underperforming for reasons AI Max will not fix. Use the ad group level toggle to keep a control. Enable search term matching on half your ad groups and leave the rest untouched, so seasonality and market movement affect both sides equally. Set the measurement window before you start. Two weeks is too short for Smart Bidding to settle after a targeting change. Four weeks is the practical minimum, and six is better if conversion volume is under about 30 per month. Write down the number that would make you turn it off before you turn it on. Deciding what counts as failure after you have seen the data is how underperforming tests get extended indefinitely.

Turning AI Max on: the setup, setting by setting

The configuration itself takes about ten minutes. The preparation should take considerably longer.

Before you touch a setting

Confirm three things first, because each one determines whether the rest is worth doing. Conversion tracking must be accurate and recent. Check that your primary conversion action fires correctly, that you are not double-counting, and that the values passed back reflect real commercial outcomes rather than form submissions of unknown quality. Teams routing leads through a CRM should confirm the HubSpot side is passing qualified outcomes back to Google rather than raw enquiry counts. Your bidding strategy must be conversion-based. If you are on manual CPC, migrate to Target CPA or maximise conversions and let it stabilise for a fortnight before layering AI Max on top. Changing two variables at once produces data you cannot read. Your negative keyword lists must be current. Keywordless matching operates inside your exclusions, so those lists are now doing more work than they were.

The settings, in order

Work through these in sequence rather than switching everything on at once.

1. Open the Search campaign and find the AI Max section in campaign settings. Enabling AI Max activates search term matching by default.
2. Decide on text customisation. Leaving it off keeps your approved ad copy intact and also keeps Final URL expansion unavailable. Switching it on opens both.
3. Decide on Final URL expansion separately, and only if text customisation is on. For a first test, most advertisers should leave this off.
4. Add term exclusions, up to 25, covering anything that must never appear in generated copy.
5. Configure brand exclusions at campaign level and brand inclusions at ad group level where relevant.
6. Set locations of interest at ad group level if geographic intent matters to your business, which for most UK service businesses it does.
7. Apply text guidelines describing what the AI may and may not say about your business.
8. Save, then check the ad group level toggles to confirm your control groups are genuinely excluded.

The sequencing that produces readable data

Enable one component, measure for the full window, then add the next. Search term matching first, because it carries the least brand risk and produces the clearest signal about whether there is untapped demand. Text customisation second, once you have seen what the wider query pool looks like and can judge whether generated copy would help or hurt against those queries. Final URL expansion last, and only after auditing which pages Google might select. Switching all three on simultaneously gives you one number and no way to attribute it, which is how most of the neutral and negative results in the sentiment polls were produced. Advertisers building a Search account from scratch should get the fundamentals right first. Our guide on how to set up Google Ads covers the campaign structure that AI Max then sits on top of, and a weak structure underneath will not be rescued by an AI layer above it.

Term exclusions, brand controls and text guidelines

These are the controls that decide whether AI Max is a manageable tool or an uncontrolled one, and they are the least discussed part of the feature.

Term exclusions

You get up to 25 term exclusions per campaign. These are words and phrases the AI must never place in a headline, description or sitelink. The enforcement extends negative keyword logic into the text generation layer, which is a meaningful change. A negative keyword stops your ad showing against a query. A term exclusion stops the words appearing in your ad. Twenty-five is not many, so spend them on genuine risk rather than stylistic preference. Regulated claim language, superlatives you cannot substantiate, competitor names, discontinued product names and anything your legal team has previously asked you to remove all belong on the list. “Cheap”, “guaranteed”, “best” and “free” earn their place in most UK accounts.

Brand exclusions and inclusions

Brand exclusions sit at campaign level and stop your ads serving alongside or against brands you specify. Brand inclusions sit at ad group level and do the reverse, concentrating targeting on users interested in named brands. The competitor hijacking problem described earlier is precisely what brand exclusions exist to prevent. If you enable search term matching without configuring brand exclusions, you should expect competitor brand queries to appear in your search terms report within days. Whether that is a problem depends on your strategy. Some UK advertisers bid on competitor terms deliberately and profitably. The failure is not competitor traffic, it is competitor traffic arriving unplanned at a cost per acquisition nobody chose.

Locations of interest

Locations of interest operates at ad group level and targets geographic intent within the query rather than the physical location of the searcher. For a Suffolk business serving East Anglia, this is the difference between reaching someone in Ipswich searching generally and reaching someone in London searching specifically for a supplier in Ipswich. Both are valuable. They are not the same audience and often should not sit in the same ad group. Service businesses with defined catchment areas should configure this before migration rather than after. Getting it wrong wastes budget on geographic intent you cannot serve.

Text guidelines for regulated sectors

Text guidelines let you describe, in plain language, what the AI may and may not say about your business. Think of them as a brief rather than a filter. Financial services, legal and healthcare advertisers carry the most exposure here, because generated copy can drift into claims that breach FCA, SRA or ASA expectations without anyone approving a word of it. The exposure is real and it is yours, not Google’s. Practical guidance for those sectors: write guidelines that name the specific claim types you cannot make, list your regulatory constraints explicitly, and pair them with term exclusions covering the language your compliance function has already flagged. Then review generated copy weekly for the first month rather than trusting the configuration.

Tip: Export your generated headlines and descriptions at the end of week one and read every line. It is tedious, it takes about twenty minutes, and it is the only way to find out what the system actually writes about your business rather than what you assume it will write.

Final URL expansion: when it helps and when it sends traffic to the wrong page

Final URL expansion deserves its own section because it is the component most likely to produce a bad outcome that nobody notices for a month.

What it actually does

The setting allows Google to override the final URL on your ad and send the click to a different page on your site, chosen for relevance to the individual query. Google’s documentation on Final URL expansion describes the intent as improving engagement by matching intent to content. Your whole site becomes the landing page inventory. Every published page, including ones you had forgotten about.

The sites where it works

Deep, well-maintained sites with genuine content variety benefit most. Ecommerce catalogues with thousands of product pages are the clearest case, because no keyword-to-URL mapping a human builds will ever be as granular as the catalogue itself. Publishers and large content sites see similar gains. So do multi-service businesses whose service pages are genuinely distinct and individually optimised for conversion. The common factor is that a randomly selected page is still a decent page. If that holds for your site, expansion is likely to help.

The sites where it does damage

Thin sites with a handful of pages have nothing useful to expand into, so the feature adds risk without upside. Sites with a large legacy blog archive are the classic failure case, and it catches out more UK businesses than any other profile. Ten years of posts, many outdated, none built to convert, all crawlable. Paid traffic lands on a 2019 article with no clear next step, bounces, and your cost per conversion climbs while your click volume looks healthy. Businesses with pages that must not receive paid traffic form the third group. Careers pages, investor information, terms and conditions, old campaign landing pages still live for reference. Anything indexable is a candidate destination, and “we assumed Google wouldn’t send traffic there” is not a control.

The audit to run before enabling it

Half an hour of work prevents most of the problems. Crawl your site and list every indexable page. Sort by last modified date and flag anything older than about two years that has not been reviewed. For each flagged page, decide: update it, remove it from the index, or accept paid traffic landing on it. There is no fourth option once expansion is on. Check the conversion path on your top twenty pages by organic traffic, since those are the pages Google is most likely to consider relevant. A page with no form, no phone number and no obvious next action will convert paid traffic at close to zero regardless of how well it matches the query. If that describes several of your pages, the fix is conversion rate optimisation before expansion, not instead of it. Then monitor the landing page report weekly for the first month. Pages you did not nominate appearing with meaningful spend and no conversions is the signal to intervene.

Governance for larger accounts

Teams managing accounts across several stakeholders need a rule, not a habit. The workable version: any page that receives paid traffic must have an owner, a conversion action and a review date. Pages that fail those tests get excluded from the index or improved. Both are more productive than discovering in a quarterly review that 15% of paid clicks went somewhere nobody was responsible for.

The 17 August 2026 bidding change and what budget-limited campaigns must do

This landed two days before this guide was written, and it interacts with AI Max in a way most coverage has treated separately.

What changed

Google updated how it bids on campaigns that are limited by budget and use a target-based strategy: Target CPA, Target ROAS, or Target CPC on Demand Gen. Previously, budget-limited campaigns frequently over-performed against their stated target. A campaign with a £40 Target CPA and a budget too small to capture available demand would often deliver at £30, because the bidding system concentrated spend on the very cheapest available conversions. From 17 August 2026, the bidding systems deliver performance closer to the stated target even as budgets shift. Google’s FAQ on the target-based bidding changes sets out the mechanics.

Why this matters more than it sounds

Plenty of UK advertisers have been running on accidental over-performance for years without knowing it. Their stated target was aspirational, their budget was the real constraint, and the gap between the two was doing the work. Those accounts will see cost per acquisition rise towards the stated target. Nothing has broken. The campaign is now doing what it was told to do, and what it was told to do was wrong. If your actual CPA has consistently sat well below your Target CPA, your target is not a target. It is a ceiling you never approached, and it needs resetting to reflect what you actually want to pay.

The interaction with AI Max

Here is the part that gets missed. AI Max widens the query pool, which increases the volume of available conversions, which changes whether a campaign is budget-limited at all. A campaign that was budget-limited before AI Max may find itself with more demand than budget afterwards, pushing it further into the territory this change affects. Two variables move at once, and separating their effects afterwards is difficult. The practical sequencing for anyone testing AI Max this autumn: let the bidding change settle first, establish a new performance baseline over two to three weeks, then introduce AI Max. Running both changes together in September will produce a number nobody can attribute.

The preparation work

Google made a Bid Target Adjustment Tool available from 6 July 2026, giving roughly six weeks of warning. If you have not used it, the manual version is straightforward. Pull 90 days of performance for every campaign on a target-based strategy. Compare actual CPA or ROAS against the stated target. Where actual performance sits more than about 15% better than target, the target is wrong and needs bringing in line with reality. Where actual performance sits close to target, nothing changes. Where the campaign is not budget-limited, this change does not apply and no action is needed. Checking which of your campaigns are actually budget-limited is the first step, and a surprising number of accounts have never looked.

Reporting: reading AI Max data properly

Query volume expands, ad copy varies by impression, and landing pages multiply. Old reporting habits stop working.

The search terms report is still your primary tool

It survives the migration, and Google has added the ability to distinguish queries matched by keywords from queries matched by keywordless technology. That distinction is the most important number in the whole feature. Queries arriving through keywordless matching that your keywords would never have caught represent genuinely incremental reach. Queries arriving through keywordless matching that duplicate your existing keywords represent cannibalisation. Review weekly for the first month, then fortnightly. Monthly review is too slow when the query pool has widened, because a month of unnoticed competitor traffic is a month of wasted budget.

Reading the asset and creative data

Text customisation means your ads differ by impression, so asset-level reporting tells you less than it used to about what the reader actually saw. Focus on the generated combinations rather than individual asset performance. Export them, read them, and check for claims your business cannot support and tone that does not match your brand. Keep a record of the original copy. Comparing generated variants against the approved baseline is the only way to judge whether text customisation is improving relevance or diluting positioning.

Landing page reporting

Sort by cost, not by clicks. A page taking meaningful spend with no conversions is the specific failure mode Final URL expansion produces, and cost ordering surfaces it immediately. Watch for pages appearing that you never nominated. Some will be pleasant surprises. Others will be the 2019 blog post.

Connecting paid data to actual revenue

Platform conversions are not revenue, and AI Max makes the gap wider by finding queries further from the point of purchase. A wider top of funnel produces more conversions of lower average quality unless something downstream is measuring quality. Passing lead outcomes back from your CRM closes that gap. Deal stage, deal value and won or lost status tell you which queries produce customers rather than which produce form fills. We cover the mechanics in measuring ad ROI with HubSpot, and the principle applies regardless of which CRM you run. Without that feedback loop, AI Max optimises towards whatever your conversion action counts, which for most UK lead generation accounts is enquiries of wildly varying quality. Smart Bidding will happily buy you more of the worst ones if that is what you have told it to value.

Making the weekly review survivable

A wider query pool creates more review work, and review work that takes two hours a week does not get done in month three. Building the habit around a fixed, short routine is what makes it stick. Set a standing thirty-minute slot. Open three reports in the same order every time: search terms filtered to keywordless matches only, landing pages sorted by cost, and generated ad copy from the past seven days. Record what you changed and why in a shared document. Six weeks into a test, the question “when did we add that negative?” comes up constantly, and reconstructing the answer from change history is slow. Several PPC teams have published scripts that pull the keywordless versus keyword split automatically, which removes the most tedious part of the routine. Anything that shortens the review makes it more likely to happen at all.

When not to use AI Max

Most coverage of AI Max is written by people with an interest in you switching it on. The honest position is that several common UK account profiles should not, at least not yet.

Accounts with fewer than 30 conversions a month

Smart Bidding needs conversion volume to learn, and AI Max leans harder on Smart Bidding than a standard keyword campaign does. Below roughly 30 conversions per month, the models are working from too little signal to price a much wider query pool sensibly. The result is predictable: spend spreads across queries the system cannot yet evaluate, and cost per acquisition climbs while the algorithm gathers data you are paying for. The fix is not AI Max. It is more conversion volume, which usually means fixing tracking to capture micro-conversions, or widening the conversion definition to something that fires often enough to teach the system something.

Highly regulated advertisers without compliance sign-off

Generated ad copy that has not been through your compliance process is a risk that sits entirely with you. FCA-regulated firms, solicitors under SRA rules and healthcare advertisers all face real consequences for claims they did not write and did not approve. Text guidelines and term exclusions reduce that risk substantially. They do not eliminate it. Advertisers in these sectors can still run search term matching, which does not touch ad copy, while leaving text customisation off. That combination captures most of the reach benefit with none of the copy risk, and it is the configuration we would recommend for any regulated client testing this in 2026.

Sites with thin or poorly maintained content

Final URL expansion needs somewhere good to expand to. A five-page brochure site does not have that, and a site with a decade of unmaintained archive content has something worse than nothing. Search term matching and text customisation still work on thin sites. Expansion should stay off until the content underneath is worth sending paid traffic to.

Accounts in the middle of another significant change

Migrating bidding strategies, rebuilding campaign structure, replacing a website or changing conversion tracking are all changes that need their own measurement window. Layering AI Max on top produces data nobody can interpret. Wait for the other change to settle, establish the new baseline, then test.

Businesses whose margin cannot absorb a CPA increase

The independent data showed median revenue up around 13% with median CPA up around 16%. For a business with healthy margins and room to buy growth, that trade is often worth making. For a business operating on thin margins where cost per acquisition is the binding constraint, it is not. More revenue at a worse CPA is only progress if your unit economics say it is, and that arithmetic is yours to do, not Google’s. Run the numbers before the test, not after. If a 16% CPA increase would put you underwater, you need a much tighter success criterion and a much shorter leash.

What AI Max changes about keyword strategy

Keywords are not finished, despite a lot of commentary suggesting otherwise. Their job has changed.

Keywords as direction rather than boundary

For twenty years a keyword list defined the edge of what a campaign could reach. Under AI Max it describes the centre of gravity instead, and the system reaches outward from there. That shift makes some traditional practice less valuable. Exhaustive long-tail keyword lists built to capture every variant are largely redundant when keywordless matching finds those variants anyway. It makes other practice more valuable. Negative keywords now carry more weight than at any point in the last decade, because they define the boundary that keywords used to define.

Where to spend the time you used to spend on keyword research

Three areas repay attention more than expanding your keyword list. Negative keyword lists come first. Audit them properly, share them across campaigns, and add to them weekly from the search terms report during the first month. Landing page quality comes second. The system chooses which of your pages to send traffic to and generates copy from what it finds there, so page content is now an input to targeting rather than just a destination. Conversion data quality comes third and matters most. Every decision the system makes traces back to what you told it a conversion is worth.

Match types in a keywordless account

Exact and phrase match still control bidding priority and still report separately, which is why Google’s own uplift figures are highest for accounts still relying on them: those accounts have the most unreached demand. Keeping a tight exact match campaign alongside an AI Max campaign remains a sensible structure for accounts with clear, high-value head terms. The exact campaign protects your core terms at a controlled cost. The AI Max campaign finds what sits beyond them. Structure your account so you can always answer the question “would we have won this conversion anyway?” That is the question the cannibalisation data says most advertisers cannot answer.

Brand terms need their own treatment

Brand campaigns should generally stay out of AI Max. Brand queries convert at high rates and low costs, and adding them to a wider query pool contaminates the performance data for everything else. Keep brand in a separate campaign with exact match keywords and its own budget. It is the cleanest way to stop brand performance flattering a test that is not actually working.

Building negative keyword lists for a keywordless account

Negative keywords have quietly become the most important asset in a Google Ads account, and most UK accounts are working from lists built years ago for a different kind of campaign.

Why the old lists are not enough

A negative keyword list built alongside exact and phrase match keywords only ever needed to catch the near misses. The match types did the heavy lifting. Keywordless matching removes that protection. Your exclusions now define the outer edge of the campaign, which means they have to anticipate query types you have never seen in your account before. The practical consequence: a list that was adequate in 2024 is probably leaking money now, and the leak is invisible unless you go looking in the search terms report.

The five lists worth maintaining separately

Splitting negatives into purpose-built lists makes them far easier to keep current, and shared lists apply across campaigns without duplicating work. An irrelevance list, covering words that indicate a completely different intent. Jobs, careers, salary, training, courses, free, DIY, and anything indicating someone wants to do the thing themselves rather than buy it. A competitor list, naming every competitor you can think of plus their common misspellings. Keep this even if you also use brand exclusions, since the two controls work differently. A geography list, covering places you cannot serve. For a business covering East Anglia, that means excluding the major cities and regions where you would not take the work. A research-intent list, catching queries from people gathering information rather than buying. What is, definition, meaning, examples, template, calculator, and similar. A product-mismatch list, covering adjacent products you do not sell. This is the list that grows fastest once keywordless matching starts exploring, because the system reasonably assumes adjacent categories are relevant.

How to build the first version quickly

Start from your own data rather than a generic list off the internet. Export twelve months of search terms and sort by cost. Read the bottom two hundred by conversion rate, which is where the waste already lives even before any migration. Pull your GA4 landing page data for pages receiving paid traffic with high bounce rates, since those often reveal a mismatch between query and page. Ask your sales team what unqualified enquiries have in common. They will name three or four patterns immediately, and those patterns usually map straight onto negative keywords nobody in marketing had thought of. That process takes about two hours and typically produces a better list than a month of incremental additions.

Match types for negatives

Negative match types work differently from positive ones and the difference matters more now. Negative broad match blocks any query containing all your terms in any order, which is the strongest and the easiest to over-apply. Negative phrase match blocks queries containing your terms in order, which is usually the right default. Negative exact match blocks only the precise query, which is useful for a specific high-cost term you want gone without affecting anything nearby. Start with phrase for most additions, and reserve broad for terms that are unambiguously wrong in every context.

The weekly routine

Fifteen minutes a week, filtered to keywordless matches only, sorted by cost. Add anything clearly wrong. Flag anything ambiguous for a fortnightly discussion rather than acting on it immediately, since some odd-looking queries convert perfectly well. Move additions into the shared lists rather than the campaign, so the work compounds across the account instead of being repeated. Track how many negatives you add each week. When that number stops falling after a month or so, the campaign has found its boundary and you can drop to a fortnightly review.

The risk of going too far

Over-exclusion is a real failure mode and it is harder to spot than over-spending, because the symptom is an absence rather than a cost. Aggressive broad match negatives can block queries you wanted, and nothing in the interface tells you that a query was excluded rather than never searched. Watch impression share lost to rank and to budget as you add negatives. A campaign whose impression volume falls faster than its cost has probably been cut too hard. Review your negative lists annually and remove anything that no longer reflects what the business sells. Lists tend to accumulate and almost never get pruned.

AI Max against the alternatives

The choice is rarely AI Max or nothing. Four options are genuinely available to a UK advertiser this autumn, and each suits a different profile.

Staying on exact and phrase match

Still a defensible position, and the one we recommend for most regulated clients. You keep total control over which queries you reach and exactly what your ads say. The cost is unreached demand, and Google’s own figures suggest it is substantial: the 27% uplift claim applies specifically to accounts still relying mostly on exact and phrase, because those accounts have the most headroom. Making a campaign ineligible for the September migration puts you here by default, so it is worth choosing deliberately rather than arriving by omission.

Campaign-level broad match without AI Max

An underrated middle position. Broad match gives you a wide query pool while your ad copy stays exactly as written and your landing pages stay exactly as nominated. The catch is that keeping broad match enabled is one of the two triggers for automatic migration, so holding this position through September means actively managing it. For advertisers whose main objection to AI Max is generative ad copy rather than query expansion, this is the configuration worth arguing for.

Performance Max

Suits ecommerce with a strong product feed and genuine appetite for Display, YouTube and Discover inventory. It does not suit most UK B2B service businesses, whose buyers are searching with intent rather than being interrupted. Running Performance Max and AI Max in the same account against overlapping queries is the specific mistake to avoid. Google decides which serves, your reporting shows each in isolation, and the internal competition is invisible unless you go looking.

Dynamic Search Ads

Still available until February 2027, and still the right answer for large catalogue sites that have built their approach around Google crawling and matching site content. The migration delay gives DSA advertisers roughly eighteen months of runway, which is enough to test AI Max properly alongside rather than under time pressure. The genuinely bad outcome is defaulting into any of these four without deciding. September will make that decision for a lot of UK accounts, and the accounts that fare worst will be the ones that never framed it as a choice.

Ten mistakes UK advertisers are making with AI Max

These come from the documented test data, from the sentiment polls, and from the accounts we have reviewed since general availability in April.

1. Switching all three components on at once. One number, no attribution, no way to know which part helped.
2. Leaving brand exclusions unconfigured. Competitor traffic arrives within days, and in one documented case took 69% of Search impressions.
3. Testing on the best-performing campaign. Too much revenue at risk, and the campaign least likely to have untapped demand.
4. Judging results at two weeks. Smart Bidding has not finished relearning, so the data reflects transition rather than steady state.
5. Ignoring the keywordless versus keyword split in search terms. Without it, cannibalisation looks identical to growth.
6. Enabling Final URL expansion without auditing the site. Old blog posts and careers pages start taking paid clicks.
7. Treating platform conversions as revenue. A wider funnel produces more leads of lower average quality unless the CRM is feeding quality back.
8. Spending term exclusions on style rather than risk. Twenty-five slots used on tone preferences instead of compliance language.
9. Running AI Max and Performance Max against the same queries. Internal competition that neither report shows clearly.
10. Having no pre-agreed failure criterion. Tests with no defined stopping point run until someone loses patience, which is usually well past the point the data was clear.

The pattern across all ten is the same: AI Max punishes accounts that were already loosely managed and rewards accounts that were already disciplined. It is an amplifier rather than a fix.

A 30-day test plan for a small UK budget

This is built for an account spending somewhere between £1,000 and £5,000 a month, which covers most UK SMEs running paid search. Larger accounts can compress the timeline. Smaller ones should extend it.

Week zero: preparation

Do this work before the test starts, not during it. Verify conversion tracking end to end. Submit a test enquiry, confirm it appears in Google Ads with the right value, and confirm it reaches your CRM. Roughly a third of the accounts we audit have at least one broken or double-counting conversion action. Export 90 days of search terms, landing pages and ad copy. This is your baseline and you cannot reconstruct it later. Choose the test campaign. Second-best performer, conversion volume above about 30 a month, conversion-based bidding already in place and stable. Split the campaign’s ad groups into test and control. Aim for roughly equal conversion volume on each side rather than an equal number of ad groups. Write down your success and failure criteria with numbers attached. For example: continue if incremental conversions rise by more than 8% with CPA no more than 12% higher; stop if CPA rises more than 20% at any point after day fourteen.

Days 1 to 7: search term matching only

Enable AI Max with search term matching on the test ad groups. Leave text customisation and Final URL expansion off. Check search terms daily. You are looking for three things: competitor brand queries, obviously irrelevant queries, and genuinely new relevant queries your keywords would have missed. Add negatives daily during this week. The first week sets the boundary for everything that follows, and negatives added on day three save budget for the remaining twenty-seven days. Expect performance to wobble. Smart Bidding is relearning against a wider pool and the first few days rarely look good.

Days 8 to 14: stabilisation

Reduce search term review to every other day. Keep adding negatives, but the volume should be dropping noticeably. Do not judge performance yet. Comparing week two against your baseline will mislead you, because the system is still in transition. Use this week to check the keywordless versus keyword split in your search terms report. If most of the new volume duplicates existing keywords, cannibalisation is the likely outcome and you should be sceptical of any apparent improvement.

Days 15 to 21: first assessment and second component

Now compare test against control for the first time. Look at incremental conversions, CPA, and the proportion of conversions coming from genuinely new queries. If the numbers are within your continue criteria, enable text customisation on the test ad groups. If they are outside your stop criteria, turn AI Max off and write up what you learned. Where you proceed with text customisation, export generated copy at the end of the week and read every line. Check for unsupportable claims, wrong tone and anything your compliance function would question.

Days 22 to 30: final assessment

Let it run. Resist the urge to make changes in the final week, because changes made now contaminate the result you are about to measure. On day 30, compare test against control across incremental conversions, CPA, conversion quality from CRM data, and generated copy quality. Make one of three decisions: roll out to the rest of the account, keep testing with a modified configuration, or turn it off. All three are legitimate outcomes and the third is not a failure.

What to do about Final URL expansion

Thirty days is not long enough to test all three components properly. Final URL expansion should be a separate test in month two, and only after the site audit described earlier. Running it as a fourth variable inside a 30-day window guarantees an unreadable result.

Budgets, timelines and what to expect in the first 90 days

Expectations are where most AI Max tests go wrong before a single setting changes.

The first fortnight usually looks worse

Every targeting change resets the learning period, and a change this broad resets it more than most. Cost per acquisition typically rises for seven to fourteen days before settling. Teams that have not been warned about this often pull the test in week two, which is exactly when the data is least meaningful. Budget for a fortnight of noise and say so to whoever signs off the spend.

Month two is where the real number appears

By day 45 to 60 you have a steady-state figure that means something. This is the point to make a rollout decision across the account rather than a single campaign. Accounts with lower conversion volume need longer. At 30 conversions a month, a statistically meaningful difference takes closer to 90 days than 30, and pretending otherwise produces confident decisions built on noise.

Budget implications

AI Max does not require more budget, but it changes what your existing budget does. A wider query pool with the same budget means thinner coverage across more queries. Accounts that were comfortably meeting demand may find themselves budget-limited after enabling it, which then interacts with the August bidding change described earlier. Watch your impression share lost to budget in the first fortnight. A sharp rise means the query pool has outgrown your budget, and you either need more budget or tighter targeting.

The 90-day view

By month three you should be able to answer four questions with data: whether incremental demand exists at acceptable cost, whether generated copy helps or hurts, whether expansion improves or damages landing page performance, and whether lead quality has held up. Businesses that treat AI Max as a permanent programme rather than a one-off switch get better results, because the negative keyword work, copy review and page auditing are ongoing rather than a setup task. That is broadly how we approach lead generation campaigns generally, and paid search is no exception.

How GLO approaches AI Max before switching it on

We do not start with the platform. Starting with the platform is how accounts end up with a well-configured campaign pointed at the wrong audience.

The Growth Strategy comes first

Before recommending any tool, we build a detailed picture of the business: brand archetypes, tone of voice, commercial objectives, the challenges getting in the way, and the audiences most likely to buy. For each audience we develop vertical profiles covering the job titles that sign off purchases, the pain points driving the decision, the objections that slow conversion, and the lead magnets that attract enquiries in the first place. That work determines what a good query looks like for this specific business, which is the thing AI Max cannot infer on its own. A model optimising towards your conversion action has no idea that enquiries from one job title are worth six times more than another unless your data tells it so.

Then the CRM architecture

Conversion quality only reaches Google if something is measuring it. We build the CRM structure before the campaigns, so lead status and lifecycle stages reflect real commercial progress rather than activity. A lead that reaches Meeting Booked is worth more than one that stalls at Attempted to Contact, and feeding that distinction back into bidding is what stops a wider query pool degrading into a wider pool of poor leads. This is the step most accounts skip, and it is the step that decides whether AI Max helps or hurts.

Then the campaign configuration

Only at that point do the settings in this guide get touched, and by then most of the decisions have already been made. Which pages should receive traffic, what the copy may claim, which brands to exclude and what a conversion is genuinely worth are all outputs of the strategy work rather than judgement calls made in the settings screen. The configuration takes ten minutes. Everything that makes the configuration correct takes considerably longer, and it is the part that separates the accounts seeing +42% ROAS from the ones seeing -35%.

A worked example: a £3,000 a month UK service business

Abstract guidance on Google Ads AI Max only goes so far, so here is the arithmetic on a profile we see constantly: a commercial services business in the East of England, spending £3,000 a month on paid search, selling a service with an average contract value around £4,000.

The starting position

The account runs three Search campaigns. Brand, core services on exact and phrase match, and a broader prospecting campaign using campaign-level broad match with automatically created assets enabled. Monthly numbers before any change: £3,000 spend, 62 conversions, £48.39 cost per conversion. Of those 62 enquiries, the CRM shows 14 reach a booked meeting and 4 become customers. The real cost per customer is therefore £750, against a £4,000 contract value. That is a working account, not a broken one. The prospecting campaign is the migration candidate, because it uses both of the settings that trigger eligibility. It takes £1,200 of the £3,000 and produces 28 of the 62 conversions at £42.86 each.

What the numbers hide

The CRM data changes the picture completely. Of the 28 conversions from the prospecting campaign, only 3 reach a booked meeting and 1 becomes a customer. The core services campaign, at £1,100 spend and 21 conversions, produces 9 meetings and 3 customers. Cost per customer is £1,200 on prospecting against £367 on core services, which the Google Ads interface shows as £42.86 versus £52.38 in the opposite direction. Platform data ranked these campaigns backwards, and this is the single most common measurement failure in UK lead generation accounts. Anyone testing AI Max on this account using platform conversions alone would draw the wrong conclusion whatever the result.

The decision

Two defensible options exist and the choice depends on appetite. The first is to make the prospecting campaign ineligible by disabling automatically created assets before September, and test AI Max deliberately on the core services campaign instead, where lead quality is demonstrably better and the CRM feedback is more reliable. The second is to let the prospecting campaign migrate, accept it as a live test of whether keywordless matching finds better demand than campaign-level broad match was finding, and hold it to a lead quality criterion rather than a cost per conversion one. We would take the second, with conditions. The prospecting campaign is already the weakest performer on quality, so the downside is contained, and campaign-level broad match is a fair comparison point for keywordless matching.

The criteria set in advance

Continue if meetings booked from the prospecting campaign rise above 5 per month at a cost per meeting under £300. Stop if cost per meeting exceeds £450 at any point after day fourteen, or if generated ad copy produces claims the business cannot support. Note what is absent from both: cost per conversion. Judging this test on platform conversions would reward exactly the behaviour that is already costing this business money.

What actually happened in comparable accounts

Across the accounts we have moved through this pattern, the common outcome is a modest rise in conversion volume, a rise in cost per conversion, and a lead quality result that splits sharply by how well the CRM feedback was configured beforehand. Where deal-stage data flows back into Google Ads, the wider query pool tends to find genuinely new demand and bidding steers towards it. Where only form fills are measured, volume rises and quality drifts down, and the account looks better in the interface while the sales team gets busier for no additional revenue. That split is not really about AI Max. It is about whether the account was measuring the right thing before AI Max arrived, which is why the preparation work matters far more than the configuration.

Troubleshooting the five problems that show up first

These are the issues that appear in the first fortnight, in roughly the order they tend to appear.

Cost per acquisition spikes in week one

Almost always the learning period rather than a fault. A targeting change this broad resets Smart Bidding, and seven to fourteen days of higher costs is the normal pattern. Check that nothing else changed at the same time, confirm your budget has not been raised, and wait. Intervening in week one restarts the clock and guarantees a longer disruption. If costs are still running high past day twenty-one, the problem is real. Look first at query relevance in the search terms report, then at whether the campaign has become budget-limited.

Competitor brand queries appearing in search terms

Expected behaviour when brand exclusions are not configured, and it happens fast. Add brand exclusions at campaign level immediately. Add the specific competitor names as negative keywords as well, since the two controls work differently and using both is more reliable than either alone. Decide deliberately whether you want any competitor traffic. Some UK advertisers run competitor terms profitably, but it should be a strategy rather than an accident.

Generated ad copy that misrepresents the business

Usually a symptom of thin or outdated landing page content, because text customisation generates from what it finds on the page. Fix the page content first. Then tighten text guidelines to name the specific claims that are off limits, and add the problem language to your term exclusions. If the copy problem is severe or carries regulatory risk, switch text customisation off while you fix the underlying pages. That also switches off Final URL expansion, which is usually acceptable during remediation.

Traffic landing on pages you never nominated

Final URL expansion doing what it does. Sort the landing page report by cost and look at everything taking spend without conversions. Short-term fix: switch off Final URL expansion. Longer-term fix: remove the offending pages from the index, or improve them until they are worth receiving paid traffic. Old blog archives are the usual culprit in UK accounts, and the honest answer is often that those posts should have been pruned or refreshed years ago.

Conversion volume up, sales pipeline flat

The most serious of the five, and the slowest to surface because it takes a full sales cycle to become visible. The cause is nearly always a conversion action that counts enquiries without regard to quality, combined with a wider query pool reaching people further from a buying decision. The fix is downstream of Google Ads. Pass deal stage and deal value back from your CRM so bidding optimises towards outcomes rather than form fills, then give the system four to six weeks to respond to the better signal. If you take one thing from this guide, make it this: a wider query pool is only an improvement if something in your stack can tell good demand from bad. AI Max does not know the difference. Your CRM does.

Frequently asked questions

Is AI Max the same as Performance Max?

No. AI Max is a group of settings inside an existing Search campaign, while Performance Max is a separate campaign type that serves across Search, Display, YouTube, Gmail, Discover and Maps. With AI Max you keep your ad groups, keywords, negative keyword lists and search terms report, and you can switch most components off again. Performance Max replaces that structure with asset groups and gives you far less visibility over placement and query data. The similarity is that both depend on conversion-based Smart Bidding and both use generative AI to assemble creative, so both suffer badly when conversion tracking is unreliable.

Can I opt out of the September 2026 migration?

Not directly. There is no opt-out control, but eligibility is based on two settings, and removing either one before 1 September 2026 makes the campaign ineligible. Those settings are campaign-level broad match and automatically created assets. Disabling either will keep a campaign out of the automatic migration. Both are doing useful work in most accounts that use them, so treat this as a considered trade-off rather than a reflex. Dynamic Search Ads campaigns are a separate case: their migration was pushed back to February 2027, so a DSA campaign left alone will keep running in its current form well into next year.

Do my negative keywords still work with AI Max?

Yes, and they matter more than before. Keywordless query matching operates inside your exclusions rather than around them, so a negative keyword still prevents your ad serving against that query. Because the query pool is much wider, your negative lists are now doing the job that keyword match types used to do in defining the boundary of a campaign. Review them before enabling Google Ads AI Max, share lists across campaigns where appropriate, and add to them daily during the first week of any test. Term exclusions are a separate control that stops specified words appearing in AI-generated ad copy, which is a different job from stopping ads serving.

How much conversion volume do I need before AI Max makes sense?

As a working threshold, around 30 conversions per campaign per month. Below that, Smart Bidding has too little signal to price a much wider query pool, so budget spreads across queries the system cannot yet evaluate and cost per acquisition rises while it learns. Accounts under that threshold usually get better results from fixing conversion tracking to capture more valid conversion actions, or from tightening targeting, than from adding AI Max. If you do test below 30 conversions a month, extend the measurement window to 90 days rather than 30, because a shorter window will produce a confident-looking result built on noise.

Will AI Max rewrite ad copy my compliance team approved?

Only if text customisation is switched on. That component uses your existing ads, your landing page content and generative AI to produce headlines and descriptions matched to individual queries, which means live ad text that nobody has signed off. Regulated advertisers can run search term matching on its own and leave text customisation off, which captures most of the reach benefit with none of the copy risk. If you do enable it, configure up to 25 term exclusions covering prohibited claim language, write text guidelines naming your regulatory constraints explicitly, and export and read generated copy weekly for at least the first month.

Where this leaves UK advertisers

AI Max is arriving in most UK accounts whether or not anyone planned for it, and the twelve days before September are better spent preparing than debating. The practical priorities:

• Decide campaign by campaign whether you want the migration or want to make the campaign ineligible, and record the reasoning
• Export search terms, ad copy and landing page data now, while a clean baseline still exists
• Configure brand exclusions and term exclusions before migration rather than after the first competitor query appears
• Keep Final URL expansion off until you have audited what Google might select from your site
• Fix conversion quality feedback from your CRM, because everything the system does traces back to what you told it a conversion is worth
• Set your stop criteria in numbers before the test starts

Accounts that were disciplined before AI Max tend to do well with it. Accounts that were not tend to find out quickly. There is a wider point worth making about the direction of travel. Google has spent three years moving control from the advertiser to the model, and Google Ads AI Max is the clearest expression of that so far in Search. Arguing about whether that is welcome is not a good use of anyone’s time, because it is happening regardless. The productive response is to get very good at the controls that remain, and those controls are more substantial than the commentary suggests. Negative keywords, brand exclusions, term exclusions, text guidelines, locations of interest and the decision about which pages are indexable at all add up to real influence over what the system does. Advertisers who treat those as configuration details will be disappointed. Advertisers who treat them as the new craft of paid search will do well. What has not changed is the fundamental: knowing precisely who you are trying to reach, what they are worth, and what happens after they enquire. Every automated system built in the last decade has rewarded businesses that could answer those three questions and punished businesses that could not. If you would like a second pair of eyes on your account before the migration window opens, book a Growth Strategy session and we will go through it with you.