Section 1
The artifact: the page-priority restructure
Sort every page on your site into one of four tiers by a single test: can a language model deliver this page's value inside the results page without sending anyone to you? The more completely it can, the lower the tier. Then invest inversely to that. Your effort, your internal links, and your homepage real estate flow to the top tiers. The bottom tier gets pruned, consolidated, or demoted. Tier 0. Transactional. The pages a summary cannot stand in for. These resolve an action, not a fact. A model can describe how booking works. It cannot take the booking for you. • Booking or scheduling page, with a real calendar or call action above the fold. • Quote or estimate request page. • Service-area pages: one genuine page per town or zip you actually serve, naming the area, with local proof. • Contact page with phone, hours, and a map. Checklist for Tier 0: every one of these loads in one click from the homepage. Phone number is tap-to-call on mobile. The primary action is visible without scrolling. Forms ask for the fewest fields that still qualify the lead. No Tier 0 page is more than two clicks deep from any other page on the site. Tier 1. Proof. The pages that answer "why you" after the machine has answered "what." The Overview can explain the service. It cannot vouch for you specifically. Proof is the value it structurally cannot generate for a business that has not earned it. • Reviews and testimonials, shown on the pages where decisions happen, not quarantined on a single page. • Case studies and before/after work, with real names, locations, and numbers. • Credentials, licenses, insurance, guarantees, warranties. • Team and about content that establishes a real person and a real address. Checklist for Tier 1: proof appears on every service page and every service-area page, not only on a dedicated testimonials page nobody visits. At least one piece of proof is specific to each service you sell. Credentials are stated as verifiable facts, license numbers and dates, not adjectives. Tier 2. Commercial-investigation. The pages that catch the searcher still comparing. These sit between fact and hire. Some of this intent still reaches a page, because "which is better for my house, X or Y" often shows a summary but also a searcher who wants to talk to someone. • Service pages, one per service, written to convert not to educate. • Transparent pricing or price-range pages, which the answer box tends to cite and which pull the click when a searcher wants specifics for their situation. • Comparison pages framed around the decision, ending in a call to action. Checklist for Tier 2: each service page ends in a Tier 0 action. Pricing is concrete enough to be useful, ranges and what moves them, because vague pricing helps no one and specific pricing earns the click from a summary. Every Tier 2 page links to the relevant proof. Tier 3. Informational. The blog the answer box now eats. Demote, prune, consolidate. This is the tier most sites over-invested in, and it is the tier the machine has taken. Do not delete it reflexively, but stop feeding it. • Audit every informational post against three outcomes: does it still drive any measured traffic or conversions, does it earn AI citations that send referral sessions, or does it do neither. • Neither: prune it or fold it into a stronger page (consolidate ten thin "tips" posts into one strong guide, then redirect the rest). • Cites but rarely converts: keep it lean, add a proof block and a Tier 0 action, and stop expanding it. • Still converts: leave it, but do not write more like it on the assumption the old funnel works. Checklist for Tier 3: no informational post is one of your homepage's primary links. Internal links from Tier 3 point up to Tier 0 and Tier 1, not sideways to more blog posts. Your publishing calendar has shifted its hours from new informational posts to strengthening Tier 0 through Tier 2.
Section 2
How to run it, in order
The order matters because the early moves gate the value of the later ones. First, inventory and tier. Pull every URL from your sitemap or a crawl into a sheet and assign each a tier using the test above. Add three columns from your analytics: last-12-month traffic, last-12-month conversions, and whether the page earns any AI or referral citations. This is the evidence you will make cut decisions on, so pull it before you form opinions. Second, fix Tier 0. Before you touch a single blog post, make sure the transactional pages are one click deep, mobile-first, and fast. This is where recovered attention has to land. A rebuilt content map that funnels people to a slow, buried booking page recovers nothing. Third, spread the proof. Move Tier 1 proof onto Tier 2 pages. The searcher who arrives already knowing the "what" from a summary needs the "why you," and it should be on the page, not a click away. Fourth, prune and consolidate Tier 3. Now use the evidence. Redirect the dead posts into stronger pages so you keep whatever authority they hold, rather than deleting them into 404s. Consolidation concentrates ranking signals that thin posts scatter. Fifth, rewire internal links. Links are how you tell Google, and increasingly the answer engines, what matters. Point them up the tiers. A site whose internal links all flow into a blog is telling the machine the blog is the point. It is not anymore.
Section 3
Why this holds, in one note
Two models justify the tiering, and it is worth naming their limits rather than pretending the map is permanent. The first is comparative statics: move one variable, "Google now answers informational queries in-box," and trace where value goes. It drains out of Tier 3 and pools in the surfaces prose cannot replace, the transactional action and the specific proof. The tiering just follows the value. The limit: this assumes the interface stays put. If Google begins brokering bookings inside the results page, Tier 0 gets exposed too, and the safe ground moves again. The second is a behavioral read of your competitors. The consensus reaction to AI search is to write more content to win citations, which pours effort back into the exact tier the machine has taken. That leaves the transactional-first rebuild underpriced, because most operators are running the other way. The limit: some informational content genuinely still earns citations that convert, so "prune the blog" is a default to test against your own data, not a law to apply blind.
Section 4
What the evidence does and does not support
The Pew click-through figures are solid: a real browsing panel of 900 US adults and roughly 69,000 searches, published July 2025. Treat the 8 percent and 1 percent as reliable direction. The broader "40 to 70 percent traffic loss" numbers that circulate come largely from SEO-vendor blogs and vary wildly by industry, so use them to understand the shape of the problem, not to forecast your own decline. The only decline that matters for this rebuild is the one in your own analytics, which is exactly why the inventory step pulls your numbers first. And the tiering assumes you serve a definable market where booking and proximity matter. A national e-commerce or pure-content business would tier its pages differently.
Section 5
The fitness test
Run this rebuild if you are a local or regional service business, your site still leads with a blog, and your transactional pages are buried more than one click deep. For you the restructure is close to free and the upside is direct: you stop maintaining pages the machine has taken and start funneling recovered attention to pages that book jobs. Do not run it wholesale if your informational content still drives measured conversions or citation-led referral sessions, or if content is genuinely your product. For you the answer is selective pruning guided by the evidence columns, not a blanket demotion. Either way, pull the inventory before you cut. The tiers tell you where value went. Your analytics tell you which specific pages to move, and that is the only cut list worth acting on. Sources: Pew Research Center, July 2025; Search Engine Land on the Pew study; The Digital Bloom 2025 zero-click analysis.