An organic-traffic drop requires comparing demand, impressions, rankings, clicks and the search experience before attributing it to AI Overview. The generated block may matter, but other hypotheses remain: seasonality, ranking changes, technical problems or measurement changes.
Identify which stage changed. Losing clicks, disappearing as a source and losing recommendations are different events and may require different interventions.
Start with affected pages and queries
Compare equivalent periods, accounting for calendar, device, country and query groups. Separate branded from unbranded searches, informational from commercial pages, and concentrated from widespread losses.
Check website changes, redirects, indexing, consent and analytics collection. A session-measurement issue can coexist with stable Search Console clicks. Documenting the difference avoids changing content to fix an instrumentation problem.
Combine signals before concluding
| Signal | Hypotheses to examine | Next check |
|---|---|---|
| Impressions and rankings fall | Ranking, indexing or demand change | Queries, pages and technical status |
| Stable impressions with fewer clicks | CTR, result-page composition or intent | Captures and result comparison |
| Site cited, brand not endorsed | Source participation without recommendation | Content, offer and selection criteria |
| Brand recommended, few visits | The answer resolves part of the query or offers no click | Journey, attribution and conversion |
| Decline only in analytics | Collection, consent or channel classification | Configuration and dataset reconciliation |
The table organizes hypotheses, not automatic diagnoses. Average position and query mix may change together; compare equivalent groups.
Observe the answer buyers see
Select important queries and capture result pages, including whether AI Overview appears. Repeat the sample to assess variation. One screenshot does not represent every user session.
Distinguish source, mention and recommendation inside the answer. In Fly Vet’s June 2026 Brazilian measurement of 100 Portuguese questions, Google AI Overview had 81 source classifications and no recommendations. That historical snapshot shows why citation counts alone cannot assess supplier selection.
Google AI Overview and the Gemini app are different surfaces; the cited measurement excluded Gemini. This sample does not estimate the traffic loss experienced by all Brazilian websites.
Complementary instruments
Search Console: tracks search impressions, clicks, queries and positions. Those indicators alone do not classify the brand’s role in generated text.
Answer captures: document how brands and sources appear within the examined universe. They do not measure every conversation or replace analytics.
Analytics and CRM: track sessions, actions, opportunities and revenue under available attribution. No visit does not prove no influence.
Server logs: help observe access. User-agent alone does not authenticate a crawler; a request is not a citation.
Match action to the problem
For technical losses, fix access and indexing. If a summarized answer satisfies a simple question, consider additional page utility: a tool, comparison, documentation, firsthand experience or a decision step worth visiting.
If the brand is a source but not a solution, improve offer fit and verifiable evidence. If recommendations are lost in commercial questions, inspect the suggested alternatives and selection criteria.
Maintain a stable question core for SOV alongside traffic and business indicators. The review arXiv 2607.14035 reinforces that effects depend on the observed design; timing alone does not isolate the generated block’s impact.
Frequently asked questions
How can I confirm AI Overview caused the drop?
Combine comparable queries and periods with result-page observation and other changes. Without isolating factors, describe a hypothesis or association rather than proven causality.
Does being cited resolve traffic loss?
Not necessarily. Citation, visits and conversion are different outcomes. Specify which the commercial objective requires.
What percentage drop is meaningful?
It depends on volume, seasonality, normal variation and business impact. No universal threshold fits all websites.
Should I stop investing in SEO?
Follow the diagnosis. SEO may support discovery and conversion while GEO adds analysis of participation in answers.
How murmur works
murmur works through a SWAS model, combining proprietary measurement software with specialists who diagnose, plan and execute GEO to grow Share of Voice (SOV). Eligible contract models may include an SOV-growth guarantee, with the metric, baseline, period, scope and conditions agreed in writing. This does not guarantee an individual answer, a fixed position or sales.
Author: Mateus Gomes, murmur operator. This article explains our approach and discloses our commercial interest.