Section 1
The artifact: the zero-click damage report
The model rests on one idea. Not all lost clicks are equal, because not all queries were ever going to convert. An informational query you lost to the answer box may have been worth almost nothing. A hire-intent query you lost is an emergency. So the report classifies every query first, then measures the before-and-after decline within each class, then weights the loss by what that class was actually worth to you. One tab, a handful of columns, and three summary rows. The source data. Everything comes from Google Search Console, which is free and reports your real impressions, clicks, and click-through rate per query. Export two periods: a "before" window from before AI Overviews expanded across your queries, and an "after" window from the most recent comparable period. Use matching lengths and seasons, quarter over the same quarter a year earlier is cleanest, so you are not blaming AI for a slow January. The columns. One row per query (or per query group). Build these columns: The four query classes. This is the judgment that makes the report honest. Tag each query: • Transactional. Hire-intent: "emergency plumber near me," "book roof inspection." These were your money queries. • Commercial-investigation. Comparing before hiring: "best tankless water heater," "cost to repaint a house." Real value, partially at risk. • Informational. Pure fact-seeking: "how does a heat pump work." Top-of-funnel at best, often worth little directly. • Branded. Your name: "Acme Plumbing reviews." High-intent, and a class the answer box treats differently. The value-per-click logic. Do not use a made-up number. Derive it. Value per click for a class equals your average job value, times your close rate from a click in that class, times the conversion rate from click to lead for that class. If you lack per-class rates, use a defensible split: assign transactional clicks the highest value (they were closest to a booking), commercial-investigation a fraction of that, informational a small nominal value, and branded high because those searchers were already looking for you. The point is not false precision. It is to stop counting a lost informational click as if it were a lost customer, which is exactly the error the scary headlines make.
Section 2
The three summary rows that tell the story
Underneath the query rows, compute three things. They are the actual deliverable. Row 1. Total dollar loss, and its concentration. Sum the dollar-loss column. Then sum it within each query class. The headline is rarely "we lost X dollars." It is "90 percent of our dollar loss is concentrated in two classes," or the more common and more reassuring finding, "most of our lost clicks were informational and worth little, while our transactional clicks barely moved." Either result is actionable. A blended traffic number hides both. Row 2. The CTR delta where AIO is present versus absent. Filter to queries where AIO present is Yes, and compare average CTR before and after. Then do the same for the No rows. This isolates the answer box's effect from every other cause of traffic change. If your CTR fell hard on AIO-present queries and held on AIO-absent ones, you have your culprit in your own data, not in a blog's average. If CTR fell on both, something else is also going on and AI is not your whole problem. Row 3. The rebuild priority list. Sort the query rows by dollar loss, descending, filtered to Transactional and Commercial-investigation only. This is your rebuild queue. The top rows are the high-value queries the answer box actually cost you, and they are where the transactional-first rebuild, the interactive tools, and the map-pack work should point first. Ignore the informational rows near the bottom no matter how many clicks they shed. Losing clicks that were never going to convert is not damage. It is noise you were paying to host.
Section 3
How to run it, and how often
Build it once with a year-over-year comparison. Then re-run it quarterly, appending each period so you can watch the trend, because AI Overview coverage is still expanding and this quarter's safe query is next quarter's summarized one. Tag the AIO-present column fresh each time by spot-checking your top queries in an incognito search, since coverage changes. The whole thing is one afternoon to build and an hour a quarter to refresh, and it replaces every secondhand statistic with your own.
Section 4
Why this model is the right one, in one note
Two models justify the structure. Comparative statics is the backbone: you hold everything else as steady as you can (same season, same query set) and move one variable, the presence of the answer box, to isolate its effect. That is why the year-over-year window and the AIO-present split matter. Without them you are attributing every wiggle in traffic to AI, which is how the panic numbers get inflated. The limit: Search Console is imperfect, it samples and rounds, and it does not tell you why CTR fell, only that it did. Treat the output as a strong estimate, not an audited ledger. A behavioral correction sits underneath the value-per-click column. The scary industry numbers commit a base-rate error, counting all lost clicks as equally valuable, which over-weights the vivid big-percentage loss and ignores that most of it was low-intent traffic that never booked. Weighting by class is the discipline that counteracts it. The limit: your per-class value estimates are themselves assumptions, so keep them conservative and label them, and let the concentration pattern, not the absolute dollar figure, drive decisions.
Section 5
What the evidence does and does not support
The point of this artifact is to stop importing other people's numbers, so use the public figures only to sanity-check, not to fill your cells. The reliable external anchor is Pew: a real browsing panel found clicks in 8 percent of AI-summary searches versus 15 percent without, so roughly a halving of CTR where a summary appears is a defensible expectation to compare your AIO-present rows against. The widely quoted "40 to 70 percent traffic loss" range comes largely from SEO-vendor reports and varies enormously by industry, so it belongs in the "others saw this" column, not in your forecast. Your Search Console export is the ground truth. Where your numbers disagree with the headlines, trust your numbers, because the headline was never measuring your business.
Section 6
The fitness test
Build the damage report if you have a Search Console history that straddles the AI Overview rollout and you are about to spend money reacting to it. For you the report is the cheapest possible insurance against spending in the wrong place: it tells you whether AI actually hit your money queries or just cleared out low-value informational traffic, and it ranks the rebuild so your effort lands where the dollars were lost. Skip it, or simplify it, if you have too little search traffic for the query-level data to be stable, or if search was never a meaningful channel for you, in which case the honest answer is that AI Overviews did not take much you were counting on. Either way, do not act on a number you did not measure. Build the report, read the concentration, and rebuild from the top of the priority list down. Sources: Pew Research Center on AI summary clicks; Search Engine Land on the Pew study; The Digital Bloom 2025 zero-click crisis analysis; Contently on what the 2026 traffic data shows.