AI Max for Search: Keywordless Search, Text Customization and URL Expansion

by Francis Rozange | Apr 4, 2026 | Google Ads

Category: Google Ads | Reading time: 24 minutes | Last updated: April 2026

AI Max for Search is the feature Google has been pushing to almost every Search advertiser since the global rollout finished in late 2025. The pitch is simple: turn it on and your campaigns get keywordless query matching, AI-generated headlines and descriptions, and automatic landing-page selection, all powered by the same Gemini stack that runs Performance Max. The reality is more textured. Some accounts post the gains Google quotes in its case studies. Others watch their branded CPCs double overnight, see their tone of voice flattened by generative copy, or discover three weeks in that Smart Bidding has been training on a polluted signal. This guide covers what AI Max actually is in 2026, the three components in detail, the controls Google has shipped, and a 30-day evaluation playbook that will tell you whether the feature is paying you back.

What AI Max for Search actually is

AI Max is not a campaign type. It is a feature layer that sits on top of an existing Search campaign, toggled on at the campaign level under “AI Max for Search campaigns” in the campaign settings panel. When you enable it, Google bundles three behaviors that previously lived in separate features (broad match expansion, dynamic ad customization, and final URL expansion) into a single AI-driven layer that uses your landing pages and your ad assets as the inputs.

The release timeline matters because it explains why the feature behaves differently across accounts. Google announced AI Max at Google Marketing Live in May 2025 (see the official announcement on the Google Ads blog) and ran a closed beta through summer 2025. Global rollout to Search advertisers in supported markets finished in late 2025, with text guidelines, the most important brand-safety control, moving from limited beta to general availability in February 2026 according to Search Engine Land’s coverage. Accounts that opted in during the beta now have months of historical signal feeding the model. Accounts switching it on for the first time in 2026 are starting from cold and will see less stable performance for the first two to four weeks while the system collects baseline data.

The architectural distinction worth understanding: AI Max keeps the campaign’s existing keyword list, bidding strategy, location targeting, and conversion goals intact. What changes is how Google decides which queries to match, which copy to render, and which URL to send the click to. The keyword list becomes a steering signal rather than a hard constraint. Your landing-page content and your ad assets become first-class training inputs rather than passive elements. The result is a campaign that behaves more like Performance Max in its automation surface, but stays inside the Search network with the visibility tools (search terms report, keyword report, asset report) that Performance Max does not give you.

The three components in detail

Keywordless search term matching

Traditional Search campaigns work because you tell Google which queries you want to bid on. You upload a keyword list, you set match types, you add negatives, and Google matches your ads to queries that fall inside the boundaries you defined. AI Max inverts the input. You still upload keywords, but they function as one signal among several rather than the only filter. The system also reads the content of your final URLs, the headlines and descriptions across your responsive search ads, your sitelinks and structured snippets, and the historical conversion patterns of the campaign. From those inputs it builds a semantic model of what your business actually sells, then matches that model against the live query stream.

The practical consequence is that AI Max can match queries that share no surface tokens with any keyword in your account. A campaign for a project-management SaaS with the keyword “team task tracker” will match queries like “kanban tool for distributed teams” or “agile sprint planning software” without those phrases existing anywhere in your keyword list, because the landing-page content and the ad copy describe a product that satisfies that intent. Google’s documentation calls this “search term matching beyond keywords.” Search Engine Journal calls it “broad match on Gemini.” Both descriptions are accurate.

The performance numbers Google publishes for this component should be read carefully. The official Google Ads Help article on AI Max cites a 14% lift in conversions at similar cost-per-conversion as the median across opted-in advertisers, with up to 27% for advertisers who were previously running mostly exact and phrase match. These are the figures Google’s salespeople will quote in onboarding calls. The methodology behind them is not fully disclosed (the test cohort selection, the comparison baseline, and the attribution window are all opaque), so they should be treated as a directional signal rather than a contractual guarantee. Tinuiti’s published commentary on AI Max suggests the median lift on accounts they manage is closer to the low end of Google’s range, with high variance: some accounts see 25% gains, others see CPA inflation that has to be reversed within the first month.

One pattern is consistent across third-party analyses. The lift from AI Max is largest on accounts that were previously over-restricted on match type. If your account is 80% exact and phrase match, AI Max opens up a query surface you were genuinely missing. If your account is already 60% or more broad match with strong negatives, the incremental query coverage from AI Max is modest, and the gains will be in the single digits.

Text customization

The second component is generative ad copy. Once AI Max is on, Google’s Gemini-based generator can produce additional headlines and descriptions on top of the assets you uploaded, drawing from your landing-page content, your existing ad copy, and the query that triggered the impression. Each impression gets a copy variant tuned to the specific query, which is a meaningful departure from responsive search ads where Google rotates a fixed set of advertiser-uploaded assets.

The control surface here is text guidelines, the feature that moved from beta to general availability in February 2026. Text guidelines let you constrain the generator in two ways: term exclusions (a list of words or phrases the AI must never use, capped at 25 entries per campaign) and messaging restrictions (natural-language instructions that shape tone, claims, and required disclosures, capped at 40 entries). The messaging restrictions are the more important of the two because they let you encode brand voice in a way the generator can apply consistently. A luxury skincare brand can write “always position the product around clinical efficacy and dermatologist endorsement, never use price-led messaging or urgency tactics,” and the generator will respect that constraint across thousands of generated variants.

Google’s published case studies on text customization include the BYD electric-vehicle account in Greater China, which Google says ran AI Max with text guidelines and reported a 24% lift in leads at 26% lower cost per lead. The L’Oreal fragrance line is another case Google cites, with a reported doubling of conversion rate at 31% lower cost per conversion. Both numbers come from Google’s own marketing materials and have not been independently verified by third-party analysts. They are useful as illustrations of what the feature can produce in best-case conditions, not as benchmarks every advertiser should expect to hit.

The failure mode for text customization is brand-voice drift. The generator pulls from landing-page content and from your existing ad copy, so if your landing pages are weak or generic, the output will be weak and generic. Worse, in regulated verticals (financial services, healthcare, legal), the generator can produce claims that are not in your approved messaging library and that your compliance team will not sign off on. Until text guidelines are documented and tested for your account, AI Max in regulated verticals is a compliance risk.

Final URL expansion

The third component is automatic landing-page selection. Without URL expansion, every ad in a campaign points to the final URL set on the ad or the ad group. With URL expansion enabled, Google can swap the destination at click time to any page on your domain that the system judges more relevant to the specific query. A query for “wireless headphones with noise cancellation” might be redirected from the homepage you uploaded to the dedicated noise-cancelling product collection page, even if you never explicitly attached that URL to the campaign.

The control surface for URL expansion is the URL inclusion and exclusion list. You can constrain the system to only specific URL paths (for example, only pages under /products/), exclude specific paths (the corporate /careers/ section, the /blog/ pages, the legal pages), or both. The exclusion list is the more critical of the two for most advertisers, because the failure mode of URL expansion is sending paid clicks to pages that were never built to convert: blog articles, support pages, archived product listings, country variants of your site that do not match the user’s location.

For e-commerce sites with deep product catalogs, URL expansion is the component with the largest upside. A retailer with 5,000 product pages cannot manually map every long-tail query to a specific landing page, and the homepage or category page they would have used as a fallback converts at a fraction of the rate of the right product detail page. URL expansion automates this mapping using the same semantic model that drives keywordless matching. For B2B service businesses with five or six landing pages, the upside is much smaller and the risk of misrouting is higher; in that segment, URL expansion is often best left off or constrained to a tight inclusion list.

How AI Max differs from Performance Max and Dynamic Search Ads

The three Google products that overlap on automation are Performance Max, Dynamic Search Ads, and AI Max. Understanding what each one actually does removes most of the confusion around when to use which.

Performance Max runs across all Google inventory: Search, Display, YouTube, Discover, Gmail, Maps, and Shopping. It is a cross-channel campaign type with its own goal-based bidding, its own asset groups, and its own audience signals. The trade-off is visibility: Performance Max gives you aggregated performance per asset group and a search themes report, but you cannot see individual search queries the way you can in a Search campaign, you cannot run a Search-specific bidding strategy, and you cannot apply traditional negative keywords without contacting Google support to add them at the account level (account-level negative keyword lists for Performance Max became available in 2023 and remain the only mechanism).

AI Max is Search-only. It runs on the Search network and Search partners, gives you a full search terms report, lets you apply negative keywords at the campaign and account level the way any Search campaign does, and stays under the Search-specific bidding strategies (Maximize Conversions, Target CPA, Target ROAS, Maximize Conversion Value). It is the right tool when you want automation but you also want to see what the automation is doing.

Dynamic Search Ads (DSA) is the older feature AI Max is in the process of replacing. DSA crawls your website and matches ads to queries based on the page content, with auto-generated headlines drawn from the page. It uses page content alone, with no input from your ad assets and no semantic abstraction beyond keyword matching against the crawled text. AI Max does the same job and more, using both the page content and the ad assets, with semantic intent matching that DSA cannot reproduce. Google has signaled that DSA will be retired over the next few release cycles. New campaigns should not be built on DSA in 2026; existing DSA campaigns should be evaluated for migration to AI Max-enabled Search campaigns.

The recommended pairing in 2026, the one Google’s account managers push and Tinuiti’s published guidance broadly endorses, is Performance Max for cross-channel reach plus a separate Search campaign with AI Max enabled for high-intent search demand. The Search campaign captures the high-converting branded and high-intent non-brand queries with full visibility and tight controls. Performance Max captures the long-tail and the cross-channel demand. The split is typically 40 to 60% of paid budget on Performance Max and 30 to 50% on Search-with-AI-Max for e-commerce, with B2B accounts skewing more toward the Search side because the long-tail Display and YouTube placements convert poorly for considered B2B purchases.

The opt-in process

Enabling AI Max takes about ten minutes once the prerequisites are in place. The prerequisites are non-trivial and worth confirming before you flip the toggle.

Conversion tracking must be working. AI Max is a Smart Bidding feature in disguise; without conversion data, the matching model has nothing to optimize against and will spread budget across queries indiscriminately. Confirm that your conversion actions are firing, that they have stable values where applicable, and that the campaign you are about to enable AI Max on has at least 30 conversions in the past 30 days. Below that volume the learning period stretches out and the early performance data will be too noisy to evaluate.

Brand exclusions must be set up first. This is the single most consequential pre-flight check. Without brand exclusions, AI Max will match against branded queries (your company name, your product names, founder names that overlap with branded search) and bid on them at the campaign’s non-brand CPC tolerance. Your branded Search campaign and your AI Max-enabled non-brand campaign will compete in the same auction. The branded clicks will go to the higher bidder, which is usually the AI Max campaign because its target CPA is calibrated against the inflated branded conversion rates. The result is that you pay 3 to 5x your normal branded CPC, and your branded campaign starves. The fix is to set brand exclusions in the campaign settings panel under “Brand restrictions,” listing every brand variation, founder name, and product name that overlaps with branded search.

Text guidelines should be drafted before launch. If your brand has a defined tone of voice, encode it as messaging restrictions before AI Max starts generating copy. If you wait until generated copy is live and then add restrictions, you will spend two weeks fixing brand voice issues in the search results page while the generator catches up.

Once the prerequisites are in place, the activation flow is: open the campaign, navigate to Settings, locate the “AI Max for Search campaigns” section, toggle the master switch on, and review the three sub-toggles for search term matching, text customization, and URL expansion. The conservative pattern is to enable search term matching and text customization in the first wave, leave URL expansion off, and add it after two to four weeks once you have stable baseline data on the other two components. The aggressive pattern is to enable all three at once and accept that the first two weeks of data will be noisy. Both patterns work; the choice depends on how much CPA volatility your account can absorb.

Controls in detail

Brand inclusions and exclusions

The brand controls are the most important guardrail. Brand inclusions let you tell the system “this campaign should bid on these brand terms” (typically used on a dedicated branded Search campaign). Brand exclusions tell the system “this campaign must never bid on these brand terms” (used on every non-brand campaign with AI Max enabled). The lists work at the campaign level and accept brand-name variants, common misspellings, founder names, and product names that double as branded search.

The pattern that fails most often is partial brand exclusion: listing the company name but forgetting product lines that have become branded queries on their own. An apparel brand that excludes “Acme” but not “Acme Original” or “Acme Pro” will see those product-name queries leak into the non-brand campaign at non-brand CPCs. The fix is to maintain a comprehensive brand-list document outside Google Ads (typically in a shared spreadsheet) and audit the brand exclusion list against it monthly.

URL inclusions and exclusions

URL controls govern where AI Max is allowed to send the click. The inclusion list (when populated) restricts URL expansion to specific paths. The exclusion list (more commonly used) blocks specific paths. Both work as path prefixes, so an exclusion of /blog/ blocks every URL under that path.

The standard exclusion baseline for an e-commerce account is /blog/, /careers/, /about/, /press/, /investors/, /support/, /help/, /legal/, /privacy/, /terms/, /sitemap/, and any country or language path that does not match the campaign’s geographic targeting. For a B2B SaaS account, the same list applies plus /docs/ (technical documentation should not be a paid landing page) and any /partners/ section. The exclusion list grows over time from URL performance reports; any URL that receives clicks but no conversions over four weeks should be a candidate for exclusion.

Locations of interest

AI Max also exposes locations-of-interest targeting, which lets you target users who have shown interest in specific locations regardless of where they currently are. A hotel chain in Bali can target users who have searched for Bali on Google, even if those users are sitting in Munich at the moment. The feature exists on Performance Max and on AI Max-enabled Search campaigns, and is found under campaign settings > Locations > Location options > “People in or regularly in” combined with locations-of-interest signals.

The use cases are travel, hospitality, multi-location services with a destination component (skiing, conference venues), and luxury goods with a destination-shopping pattern. For most other verticals, location-of-interest targeting expands the audience without improving conversion rate, so it is left off.

When AI Max helps

E-commerce with deep catalogs. The combination of keywordless matching and URL expansion is genuinely valuable for retailers with thousands of SKUs. Manual keyword-to-URL mapping does not scale past a few hundred products; AI Max automates the mapping using semantic similarity between the query and the product page content. This is the segment where Google’s published lift figures are most defensible.

Multi-language and multi-region campaigns. The Gemini-based matching model handles language and intent variation across markets better than keyword-based matching, which depends on the quality of the keyword list per language. A campaign running in 12 languages with a single set of asset signals and translated landing pages will get better intent coverage from AI Max than from a manually expanded keyword list per language.

Accounts moving from over-restricted match types. If your previous match-type strategy was 80% exact and phrase, AI Max opens a query surface you were genuinely missing. The Google-quoted 27% lift figure applies to this case specifically; other patterns will see smaller gains.

Accounts with thin keyword coverage on a high-intent product. A new product line where the keyword research is incomplete benefits from AI Max because the system can find queries the human researcher missed. The trade-off is that the spend pattern in the first two weeks is volatile while the system explores; budget for that volatility before turning it on.

Service businesses with multiple solution-specific landing pages. A B2B vendor with separate pages for, say, payroll, scheduling, and time tracking can let URL expansion route queries to the right page automatically. The conversion rate uplift from query-to-page matching is typically larger than the matching uplift itself in this segment.

When AI Max backfires

B2B with strict messaging requirements. If your sales team has a controlled vocabulary, your competitors are named in your sales process but not in your ad copy, or your buyer expects a specific value proposition framing, generative copy will misfire. Text guidelines help but are not airtight, and the cost of a single off-message ad reaching a procurement-committee buyer is higher than the cost of leaving AI Max off.

Regulated industries without mature compliance review. Financial services, healthcare, legal services, gambling, and political advertising all run into the same pattern: generative copy can produce claims that violate compliance rules even when the source material is clean. Until your text guidelines have been reviewed by compliance and tested through a structured pilot, AI Max in these verticals is a regulatory exposure.

Brand-controlled accounts with a tight tone of voice. Luxury, fashion, and editorial brands often have a tone of voice that the generator will flatten. The generator optimizes for click-through and conversion, both of which reward urgency and specificity, both of which dilute brand voice. A campaign for a heritage watch brand running with full text customization will produce ads that read like discount retailer copy unless the messaging restrictions are unusually strict.

Niche keywords with low search volume. AI Max is a data-hungry feature. A campaign generating five conversions a month does not have the signal density for the matching model to converge. The lift figures Google quotes do not apply to thin-volume campaigns; the actual outcome is typically random walk inside the existing CPA range with no measurable improvement.

Campaigns with poor conversion tracking. If your conversion events are misconfigured (every page view marked primary, missing values, broken Enhanced Conversions), AI Max optimizes against the noise. The visible symptom is bid behavior that ignores the campaign’s stated target. The fix is upstream of AI Max: rebuild the conversion tracking, validate the values, then turn AI Max back on.

Brand-new products or brand-new campaigns. The 30-day conversion baseline matters. A new campaign with no conversion history starts cold and burns budget for two to four weeks while the matching model collects baseline data. The pattern that works for new campaigns is to launch on traditional Search with phrase and broad match, accumulate 60 to 90 days of conversion history, then enable AI Max once the bidding model has stable signal.

Conversion measurement implications

AI Max changes how conversions get attributed across the campaign in three subtle ways that affect reporting and bidding.

First, the search terms report becomes the primary visibility tool, not the keyword report. Because the matching model uses keywords as one signal among many, the keyword report shows traffic against the keyword list but understates the actual query surface. The full picture is in the search terms report, which lists the actual queries that triggered impressions. Reviewing this report weekly is mandatory for AI Max campaigns; it is the only way to catch query drift, irrelevant matching, and brand-exclusion gaps.

Second, conversion attribution is data-driven by default for AI Max-enabled campaigns. Last-click attribution is no longer compatible with the matching model in 2026. If your account is still on last-click for legacy reasons, switch to data-driven before enabling AI Max. Otherwise the bidding model will optimize against an attribution model that disagrees with the matching model, and the campaign will underperform.

Third, isolating AI Max performance from the rest of the account requires discipline. The naive comparison (AI Max-enabled campaign CPA vs. non-AI Max campaign CPA) is misleading because brand exclusions move branded volume out of the AI Max campaign and into the dedicated branded campaign, artificially inflating the AI Max CPA and deflating the branded CPA. The correct comparison is the blended CPA across the brand-and-non-brand pair before AI Max vs. after AI Max, with budget held constant. WordStream’s coverage of post-rollout analysis flags this as the most common analytical mistake in 2026 campaign reviews.

30-day evaluation playbook

Week 1: launch and stabilize. Enable AI Max with search term matching and text customization on, URL expansion off. Set brand exclusions, text guidelines, and an initial URL exclusion list (the standard /blog/, /careers/, /support/ baseline). Daily review of the search terms report for the first seven days, with negative keywords added aggressively for any irrelevant query that gets impressions. The goal of week 1 is not lift; it is a clean query surface and confidence that the matching model has the right inputs.

Week 2: add URL expansion and review generated copy. Turn URL expansion on with the inclusion or exclusion list configured. Review the URL performance report at the end of the week and exclude any path receiving clicks but no conversions. Pull the asset performance report, review the generated headlines and descriptions, and tighten text guidelines for any pattern that is off-brand.

Week 3: compare to baseline. Pull the campaign’s CPA, conversion rate, conversion volume, impression share, and search terms diversity for week 3 against the same metrics for the four weeks before AI Max was enabled. Hold budget constant for the comparison. Document deltas. Pull the same metrics for the paired branded campaign, because branded performance often shifts when AI Max is enabled on the non-brand side.

Week 4: scale, hold, or roll back. Three decision branches: scale (CPA stable or lower, conversion volume up, blended CPA across brand and non-brand pair improved by 5% or more), hold (mixed results, run for another two weeks before deciding), roll back (CPA up by more than 15%, conversion rate down, brand voice issues unresolved). The roll-back branch is the one most teams skip; it should be a documented option from day one, with the criteria written down before launch so the decision is mechanical rather than political.

Common mistakes

Skipping brand exclusions. The single most expensive mistake in AI Max deployment. Documented impact across multiple agency case studies: branded CPC inflation of 3 to 5x within the first week, branded campaign starvation, blended CPA increases of 20 to 35% before the gap is identified.

Enabling all three components on day one without a baseline. If search term matching, text customization, and URL expansion all go live simultaneously, isolating which component drove the change in performance is impossible. The fix is staged enablement: matching plus text in week 1, URL expansion added in week 2, with clean before-and-after comparisons at each transition.

Treating the search terms report as optional. AI Max generates a wider query surface than traditional Search campaigns, which means the rate of irrelevant matching is higher in the first weeks. Without a weekly search-terms review and aggressive negative-keyword maintenance, budget leaks into queries that should never have triggered impressions.

Shipping text customization without text guidelines. The generator will pull from landing pages and existing ad copy. If those sources have any messaging that does not match current brand voice, the generator will reproduce the off-voice messaging at scale. Text guidelines should be drafted, reviewed, and locked in before AI Max is enabled, not after the first off-brand ad ships.

Comparing AI Max CPA to non-AI Max CPA without controlling for brand exclusion shifts. Discussed above. The correct comparison is blended CPA across the brand-and-non-brand pair. Without that control, the analysis is misleading in both directions: AI Max looks worse when branded volume moves out of the campaign, and the branded campaign looks better than it actually is because it is now consuming volume that was previously distributed.

Using AI Max as a substitute for fixing weak landing pages. If the underlying landing-page conversion rate is low, AI Max will not fix it. URL expansion can route to the best of a weak set, but it cannot make weak pages convert. The landing-page audit should happen before AI Max enablement, not in response to disappointing AI Max results.

Forgetting that AI Max is a Smart Bidding feature in disguise. The bidding strategy on the campaign is doing most of the heavy lifting. If the bid strategy is misconfigured (target CPA set too low, target ROAS unrealistic, learning mode never exited), AI Max cannot rescue the campaign. Validate the bid strategy first, validate the conversion tracking second, then layer AI Max on top.

Conclusion

AI Max for Search is a real shift in how Google Ads Search campaigns work, not a marketing rebrand of broad match. The keywordless matching, generative copy, and URL expansion are functionally distinct from the features they replace, and the best implementations get measurable lift over the previous baseline. The worst implementations burn budget for a month while the matching model trains against polluted signal, generates off-brand copy because text guidelines were skipped, and compete against the account’s own branded campaign because brand exclusions were not configured.

The accounts that win with AI Max do four things consistently: they fix conversion tracking before they enable the feature, they configure brand exclusions and text guidelines before launch, they run a structured 30-day evaluation with documented decision criteria, and they treat the search terms report as a weekly artifact rather than a quarterly review. The features that took years to build into Search (negative keywords, brand controls, conversion tracking, attribution discipline) still matter under AI Max. The automation layer sits on top of those fundamentals; it does not replace them.

The platform is moving in this direction whether or not any individual advertiser opts in. DSA is being retired. Performance Max is the default for new accounts. Smart Bidding is the only viable bidding strategy at scale. AI Max is the Search-side expression of the same architectural shift. Learning to operate it well in 2026 is not optional for any team that wants to keep running Search campaigns in the version of Google Ads that exists three years from now.