The biggest AI Max mistake is not a bidding mistake or a budget mistake. It is letting the system choose from a site that was not designed for automated routing. Final URL Expansion is both the most powerful and the most dangerous feature in AI Max for exactly the same reason: it extends reach by allowing Google to route traffic to pages it judges most relevant to the user's intent, regardless of which URL you specified when building the campaign.
For a well-structured site with strong commercial pages and a clean page feed, this is genuinely useful. For a site with thin content, outdated pages, informational blog posts, careers sections, or multiple service pages of uneven quality, it is a way to efficiently spend budget on pages that convert at a fraction of the rate of your best-performing URLs.
Why routing control matters more now
AI Max uses your site more aggressively than previous campaign types did. In a standard Search campaign, the URL you specified was where traffic went. In AI Max, the URL you specify is a starting point. The system may override it if it believes another page on your site is a better match for the specific query.
This means the quality of every page on your site becomes a paid search concern. Weak pages, outdated pages, informational content written for SEO but not designed to convert, duplicate service pages covering similar territory, and pages with poor UX or slow load times all become problems at scale when AI Max treats them as eligible landing destinations.
The right way to think about page feeds
Page feeds are not about making the AI Max system smaller. They are about preventing bad routing while preserving the reach benefits that make the campaign format valuable. The goal is to define a curated universe of pages that can receive paid traffic confidently, so the system can expand within that universe without wasting spend on pages that were never designed to handle commercial intent.
A well-built page feed combined with a clear exclusion list gives AI Max genuine freedom to route intelligently across your commercial inventory without access to the pages where that traffic will be wasted.
Simple page classification: four URL buckets
Bucket 1: Priority commercial pages
Core service pages, product category pages, top-performing collection pages, dedicated landing pages built for specific intent, and high-converting product detail pages. These should be included in the page feed and prioritized aggressively. Any query that aligns with these pages should route here.
Bucket 2: Conditional pages
Pages that are commercially relevant but may need monitoring: comparison pages, pricing pages, case study pages, and detailed product guides. Include these but track their conversion rate separately so underperformers can be moved to Bucket 3 as data accumulates.
Bucket 3: Exclude from paid routing
Careers pages, press pages, legal pages, contact and support pages, thin informational blog posts, low-intent tag archives, out-of-stock collections, outdated or deprecated pages, and any page with a bounce rate that suggests poor intent match. These should be explicitly excluded so the system cannot route paid traffic to them.
Bucket 4: Fix before use
Pages that have commercial potential but are currently underperforming due to content quality, UX issues, or missing conversion elements. These should be excluded from the page feed until the underlying issues are resolved, then added to Bucket 1 or 2 after fixing.
Lead generation example: B2B services site
A B2B services company had built a substantial content library over three years, including service pages, case studies, articles, guides, and a resource center. When AI Max began routing traffic based on intent matching, a significant portion of spend landed on informational articles that had been optimized for organic search but contained no conversion elements, no clear CTA, and no form. Form submission rates for those pages were near zero. The fix was to build a page feed containing only the six core service pages, four case study pages, and two dedicated campaign landing pages, and to exclude all blog and resource content. Cost per qualified lead dropped by 34 percent within four weeks as routing concentrated on pages built for the task.
Ecommerce example: category and PDP routing
An ecommerce retailer with several thousand SKUs used AI Max expansion without a page feed, which meant the system could route traffic to out-of-stock product pages, support FAQs, account management pages, and seasonal collections from prior years that had never been removed from the site. Build a page feed containing current category pages, active curated collections, and PDPs for products with inventory. Exclude account pages, support pages, blog content, and any collection or product page with zero in-stock variants. For the retailer, this reduced bounce rate from AI Max traffic by 19 percent and improved ROAS within the first 30 days.
How to measure routing quality
| Metric | What to look for |
|---|---|
| Page-level conversion rate | Identify which pages in the feed are converting and which are not |
| Qualified lead rate by landing page | For lead gen, whether pages in the feed generate real pipeline |
| Revenue per landing page | For ecommerce, whether routed pages generate proportionate revenue |
| Bounce rate by route class | Whether excluded pages are still receiving traffic through gaps in the exclusion list |
| Assisted conversion paths | Whether pages that do not convert directly contribute to later conversions |
| Pages receiving spend that should not | Regular audit of the URL report to catch unexpected routing |
Common exclusion failures
- –Building a page feed but not explicitly excluding the pages that should never receive traffic: the system may still route to them
- –Including too many informational pages in Bucket 2 without tracking their conversion rate separately
- –Failing to update the page feed when new content is published or old pages are removed
- –Setting up exclusions once and never reviewing the URL report to confirm they are holding
- –Treating the page feed as a set-and-forget configuration rather than a maintained list