Search Console Records AI Mode Follow-Up Queries
On 13 August 2026, Suganthan Mohanadasan published an analysis documenting conversational fragments appearing as queries in Google Search Console performance data. The analysis reports 1,127 such queries and 20,300 impressions across 16 months on one site, including AI Mode follow-ups such as "yes go on," and groups the fragments into seven categories using a prompt-based classifier.
An analysis published by Suganthan Mohanadasan documents conversational fragments appearing in Google Search Console's general performance report, including AI Mode follow-ups such as "yes go on" and "yes, pricing." The report attributes the behavior to Search Console treating an AI Mode follow-up as a new query and recording an impression when a page appears in the subsequent response.
Search Engine Journal's republication of the analysis reports that Google Search Advocate John Mueller confirmed the observation after SEO practitioner Anastasia Kourou raised the issue in early August. The retrieved reporting does not include Mueller's full statement or linked Google documentation, so the mechanism should be treated as reported rather than independently verified from primary documentation here.
Seven query categories
Mohanadasan reports extracting 1,127 queries and 20,300 impressions over 16 months from his own Search Console property, against millions of ordinary impressions. He characterizes the volume as small, but describes each observed row as a phrase submitted or generated during a real session.
The analysis groups these fragments into seven recognizable types, including:
- •Human conversational replies
- •Comparison-oriented questions
- •Tracker-bot activity
- •Agent prompts
It also describes a free, prompt-based classifier intended to label the query fragments. The retrieved source text does not provide the complete seven-category taxonomy or evaluation metrics for the classifier, so its precision and recall cannot be assessed from the available material.
Measurement remains incomplete
According to the analysis, the visible rows are only a floor because Google anonymizes many impressions. Mohanadasan also writes that combining the fragments with exports from Google's Generative AI reporting can produce a page-level view of AI-related search visibility.
For SEO and analytics teams, the immediate value is methodological rather than a complete measure of AI Mode traffic. Organizations analyzing comparable query logs often need to distinguish conversational continuations from conventional search intent, automated monitoring, and agent-generated requests before using the data for content or attribution decisions. The reported sample comes from a single site, so its frequencies should not be generalized across domains, markets, or query classes.
Mueller's reported comment adds an operational detail for large properties: the analysis states that larger sites may find more of these rows through Search Console's BigQuery bulk export. That makes export design, query normalization, and labeling rules relevant for teams attempting to monitor AI-assisted discovery alongside traditional search performance.
Key Points
- 1The analysis found 1,127 unusual queries and 20,300 impressions, providing a small but concrete view of AI Mode follow-up attribution.
- 2Search Console's standard report can record conversational follow-ups as new queries when a page appears in the subsequent AI response.
- 3For SEO data teams, query classification can separate likely human conversation fragments from automation, although anonymized impressions leave measurement incomplete.
Scoring Rationale
The report offers a practical observation about AI Mode attribution in a widely used analytics product, with potential relevance for search-data pipelines and visibility measurement. Its evidence is based on one site's data and lacks classifier evaluation details in the retrieved sources, limiting broader technical significance.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
