Is the decline real and commercially important?

Write a precise observation: which source, dates, market, device and pages changed, and whether the measure is search clicks or website sessions. Use complete, comparable periods and inspect the longer history for seasonality. A partial week compared with a full week can manufacture a crisis.

Compare search-side evidence with your website analytics and actual outcomes. Search Console clicks and analytics sessions use different definitions and collection methods; they need not match exactly. A large new divergence after a consent, tagging or site change deserves investigation before an editorial rewrite.

Check enquiries, accepted opportunities, orders, contribution and customer quality. Losing irrelevant informational visits has a different business consequence from losing a small number of high-intent buyers. Neither a preserved sales total nor a falling traffic total fully answers the other question.

Where should a traffic-loss audit begin?

Start with change dates and affected scope. Was there a migration, redesign, release, domain change, new consent choice, outage, stock change or discontinued offer? Record what changed and what remained comparable.

Then inspect a small, purposeful set of lost pages: a valuable affected page, an affected page of another type and a relatively stable comparison. Check the actual public response, redirects, nominated canonical, indexing eligibility, internal links and the relevant crawler's view. Search Console's indexing and inspection reports can help distinguish an excluded page from a page that is eligible but performing less well.

Prioritise substantial lost clicks and commercial importance, not the largest percentage change on a tiny base. If a recent release demonstrably broke a route, the technical owner should assess a controlled fix or compatible rollback, preserving later changes. That incident repair need not wait for the broader strategic audit.

Google's own traffic-drop guidance covers technical, security, spam, ranking, demand and migration explanations. It recommends inspecting page and query patterns rather than treating one graph shape as a diagnosis. Google's diagnostic guide

Check manual-action and security reports where relevant. The absence of a manual action does not rule out an algorithmic ranking change; a ranking decline does not establish a penalty. Compare update dates as context, alongside the actual affected pages and competing results.

What do impressions, positions and clicks tell you?

Segment by query intent, page, country and device. Falling impressions can reflect weaker visibility or changing demand. A sustained loss of positions on commercially relevant queries strengthens a ranking hypothesis. Stable exposure with lower click-through rate suggests investigating what searchers see and choose, including the title, competing results and search features.

These are diagnostic leads, not mutually exclusive causes. Average position and overall click-through rate can also change because the mix of queries changed. Comparing aggregates without preserving that mix can send the team towards the wrong repair.

How can you separate exposure changes from weaker click-through?

Ask for a matched-row calculation, not another comparison of sitewide averages. For a few commercially important pages, obtain before-and-after query data with the same search type, country, device and complete date windows. Use page-filtered exports or an appropriate analyst report, keeping the canonical page identity explicit. Each matched row is one query and page under those fixed conditions. Treat migrations, new rows and absent rows separately; an absent query is not automatically zero activity.

For every matched row with baseline impressions, calculate an exposure-adjusted expectation: current impressions × previous clicks ÷ previous impressions. This asks how many clicks the current exposure would have supplied at that row's old click-through rate. Add those expectations across the same matched rows. Compare the expected total first with their previous clicks, then with their actual current clicks. This is an accounting bridge, not Google's estimate of causes or a traffic forecast.

Use this constructed two-row record. One repair-service query/page has 200 impressions and twenty clicks before: 10% click-through. A maintenance-guide query/page has 800 impressions and sixteen clicks: 2%. Together they supply 36 clicks from 1,000 impressions, or 3.6%. Afterwards, the service row has 100 impressions and five clicks, while the guide row has 900 impressions and eighteen clicks. There are still 1,000 impressions, but only 23 clicks, or 2.3%.

At the earlier rates, current exposure would supply ten service clicks plus eighteen guide clicks: 28. The bridge is therefore 36 previous clicks → 28 exposure-adjusted clicks → 23 actual clicks. Eight fewer clicks arise from changing the amount and mix of exposure at the old rates; a further five arise from the service row's observed rate falling from 10% to 5%. The guide's rate stays 2%. Both components matter. Neither is an estimate of clicks lost to AI.

This specifies two different investigations. The eight-click exposure component leads to checking why valuable service impressions fell and whether guide gains address comparable buying needs: demand, ranking, availability and eligibility remain candidates. The five-click rate component identifies the matched service result for inspection of its displayed offer, competitors and search features. A blanket snippet rewrite cannot address the exposure component, while a demand-only explanation leaves the within-row deterioration unexamined. Confirmed access faults still warrant prompt repair rather than waiting for this accounting.

Fix the bridge's order before comparing periods. This version changes exposure first and rates second; reversing that order changes the component allocation. Do not sell either split as causal attribution. Even an exact query/page/country/device cell can contain changing people and result contexts. Small counts, average-position changes and unobserved features still limit the diagnosis.

What should the analyst's deliverable make visible?

Request one reconciliation showing each matched key, both periods' impressions and clicks, the exposure-adjusted expectation, each difference, commercial priority and the supporting next check. Preserve the raw exports, filters and dates. A reviewer should be able to reproduce the arithmetic and see which proposed job each row supports. Do not require a new subscription before testing a small priority set with available records.

Coverage belongs in that deliverable. Google's query tables omit anonymised queries and can truncate other rows; canonical and property/page aggregation choices also affect the available totals. Query filters can change which activity is included. Use compatible aggregation when comparing coverage, and keep any unmatched or unreconciled activity outside the matched-row conclusion. Never force visible query rows to explain the entire property's decline by assigning missing activity a zero rate. Google's dimensions and limitations, Google's explanation of filtering and limits

Where coverage is poor, report that limitation and investigate useful page-level patterns rather than presenting a precise whole-site explanation. A director can reject a recovery proposal that explains the shown rows but silently assigns the unobserved remainder to the same cause. The next commission should follow the evidence and commercial priority, not whichever component makes the largest percentage.

It is a credible hypothesis for some queries. Pew's analysis of March 2025 browsing by 900 US adults observed traditional-result clicks on 8% of visits with AI summaries versus 15% without. Searches differed, and result pages were reconstructed later. The comparison was observational, Google-specific and older; it is not a causal loss estimate for your site. Pew's original analysis and method

Use Google's Generative AI performance report and Bing's AI Performance report alongside a dated sample of the actual lost queries. These exposure reports and sampled results cover different observations. Save market, device and result context. Your own repeated sample can establish what was observed there; it cannot reconstruct every customer's earlier search page or isolate causality by itself. Google's report definition, Bing's report definition

Avoid hiding useful answers simply to force a visit. Give relevant people a reason to continue: a meaningful comparison, original evidence, a service-specific answer, availability or an action they need to complete. The visit must offer something useful beyond a repeated summary.

Which response fits the evidence?

A broken redirect or accidental exclusion calls for a technical repair and repeated inspection. A page that no longer answers the buying situation calls for research and a substantive update. If consolidation is justified, preserve useful customer routes and map retired URLs deliberately; a traffic threshold alone is a poor deletion rule. A weaker result against a better competitor calls for understanding that difference, rather than changing dates or adding keywords.

If click loss persists on answerable informational queries, assess whether the content still earns relevant consideration and whether the business depends too heavily on those visits. Test a small alternative route to customers through an appropriate existing relationship, email, partnership, event or other channel. Compare qualified outcomes and full cost, not just replacement visit counts.

Diversification can reduce dependence on one source. A broken sales page or missing tracking still needs technical recovery, and the cause of the search decline needs its own evidence. Assign separate owners for that recovery and the commercial channel decision.

How do you know the response worked?

Save the original finding, changed pages, release date, expected mechanism and acceptance measure. Repeat the relevant technical check immediately, then review search and commercial evidence over a period appropriate to crawling and buying. Annotate other changes.

Google can take time to recrawl and reassess pages, so allow for that delay in the recovery review. Retain a stable comparison where feasible and disclose uncertainty. Assess whether demand is commercially useful and its cost defensible, with enough explanation to guide the next decision. Avoid making several further changes before the first can be assessed.