Section 1
The artifact: the page triage sheet
You are building a one-tab spreadsheet with a row per page and six columns. Everything you need is free and comes from Google Search Console. Here is the sheet. Twenty minutes gets you through your top fifteen to twenty pages, which is almost always where 90 percent of the value sits. You do not need to audit the long tail.
Section 2
Step 1: pull the pages that actually dropped (5 minutes)
In Search Console, open the Pages report, switch the date range to Compare, and set it to the last three months against the same three months a year ago. Same season, so a normal winter lull does not masquerade as AI damage. Sort by the biggest click losses. Copy your top fifteen to twenty declining pages into the sheet and fill in the Clicks-lost column. Ignore pages that gained or held; they are not the patient on the table.
Section 3
Step 2: tag the intent of each page (5 minutes)
This is the judgment that makes the audit honest. For each page, ask what the visitor wanted when they arrived, and tag it: • Informational. The page teaches something: "how a heat pump works," "signs your roof needs replacing," "what SEER rating means." The visitor wanted to learn. This is exactly what the AI box is built to answer. • Commercial. The page helps someone compare before hiring: "cost to replace a roof," "tankless vs tank water heater." Real buying value, partially exposed to the box. • Transactional. The page exists to get someone to hire you: service-area pages, "emergency [service] [city]," quote and booking pages, your reviews. The visitor wanted to act. Be strict. Most home-service blogs are stuffed with informational pages that owners are emotionally attached to because they took work to write. The work is sunk. What matters now is whether the visitor was ever a buyer.
Section 4
Step 3: confirm the AI box and check whether it books (7 minutes)
For each page, do two quick things. First, run the page's main query in an incognito window and note whether a full AI Overview appears that answers the question. Mark AIO present Yes or No. This confirms the cause instead of assuming it. Second, check your CRM or call tracking: has this page ever actually produced a lead? Mark Books jobs Yes or No. A page can lose thousands of clicks and never once have booked a job, and that single fact changes everything about whether you should care.
Section 5
Step 4: assign the verdict (3 minutes)
Now the triage. Run each row through this table. The emotional trap this defuses is the informational-dead row. Those pages often show the scariest click losses, and every instinct says rescue them. The audit gives you permission to let them die, because losing traffic that never converted is not damage. It is you no longer paying to host visitors who were never going to hire you.
Section 6
What to do with each verdict
Defend pages get your attention first. Make sure the phone number, the quote form, and the booking path are obvious and fast, and that reviews and proof are front and center. These pages losing a ranking spot is a genuine emergency; treat it like one. Rebuild pages need the thing the AI box structurally cannot reproduce: an interactive estimator, a real quote flow, local proof, named reviews, something that requires the visitor to come to you rather than read a summary. A prose paragraph the box can absorb is a page you will keep losing. An interaction it must send the user to is a page you keep. Let go pages you simply stop investing in. Leave them up if they cost nothing; they are harmless brand exposure. Do not rewrite them, do not measure your SEO by them, and do not let their red numbers set the mood for the whole audit.
Section 7
A worked example: a plumber's top eight pages
Abstract rules are easy to nod at and hard to apply, so here is the audit run against a realistic set. Imagine a plumbing firm whose Search Console compare view surfaces these eight declining pages. The point is to watch how two pages with nearly identical click losses end up with opposite verdicts. Look at the first and third rows. The how-a-water-heater-works post lost 1,900 clicks, the biggest drop on the sheet, and the verdict is Let go. The water-heater-installation page lost slightly fewer, 1,850, and the verdict is Defend. If this owner had reacted to the raw click chart, the encyclopedia post would have screamed for attention because its number is the largest and its percentage loss is total. It never booked a single job. Its 1,900 lost clicks were people learning how a tank heats water, and the answer box doing that teaching for free is a cost this firm no longer has to carry. Meanwhile the installation page, which actually produces booked jobs, deserves the first hour of work, and the two commercial pages that book at least sometimes get the interactive or proof treatment that the box cannot replicate. The three informational posts, together responsible for the largest share of the scary red on the chart, get nothing but a shrug. That is the entire value of the audit in one screen: it inverts the priority the traffic chart would have handed you.
Section 8
Why this triage beats a full rebuild
Two ideas justify doing it this way rather than rewriting everything. The first is comparative statics: you isolate one variable, the AI box, by holding season and query set steady in the compare window, so you attribute the loss to the right cause instead of blaming AI for a slow quarter. The limit is that Search Console samples and rounds, so treat the click numbers as strong estimates, not audited figures. The second is a base-rate correction. The panic reflex over-weights the vivid big-percentage losses and ignores that most of them were low-intent pages that never booked. The Books-jobs column is the discipline that counteracts it, forcing every page to prove it earned a customer before it earns your time. The limit is that CRM attribution for home services is imperfect; a page may have influenced a call it never got credit for. So weight the verdict toward the pattern across your pages, not toward any single uncertain row.
Section 9
The fitness test
Run this audit if your traffic dropped, you have a Search Console history that straddles the AI rollout, and you are about to spend real time or money reacting. The twenty minutes buys you a ranked list that keeps you from rewriting fifteen encyclopedia pages nobody was going to hire from, and points your effort at the handful of pages that actually book jobs. Skip it only if you have very few pages or very little search traffic, in which case there is nothing to triage and your leads are coming from somewhere else worth auditing instead. For everyone with a content-heavy service site, this is the cheapest hour you will spend this quarter, because its main output is permission to stop working on the pages that never mattered. Sources: Pew Research Center on AI summary clicks; Google Search Central on the Performance report; Search Engine Land on AI Overviews and click loss. Your Search Console export and CRM are the ground truth; treat any industry-average traffic-loss figure as a sanity check only.