In August, SEO specialist Anastasia Kuru noticed queries in the Search Console report that didn’t look like search queries: “Yes,” “Yes, continue,” “Yes, price.” John Mueller confirmed that Search Console includes data from AI Overviews and AI Mode in the overall report, and subsequent questions within AI Mode are recorded as new queries. This opens up the opportunity to analyze fragments of AI dialogues that appear in your statistics.
In this article, we’ll break down how to recognize such fragments in Search Console data, how they differ from regular queries, and how to use this information for optimizing content for AI search.
Seven types of AI fragments in Search Console data
I analyzed 16 months of data from my site and identified seven categories of such queries. Each has its own origin and characteristics.
Short answers
These are single words or phrases like “yes,” “of course,” “show me.” They occur when a user responds to AI during a dialogue, and the response is treated as a search query. Positions in such cases are taken from the answer block, not from the search results.
Comparisons in dialogue
Phrases like “what about Resend?” or “what if we try Gemini?” — the user asks AI to compare an alternative with the already received answer. This signals a specific interest in your content.
Questions addressed to a conversational partner
The grammar of such queries implies the presence of a listener: “Can Meta Ray-Bans be hacked?”, “How do I sell this?”. They differ from regular search phrases.
Synthetic prompts from SEO tools
Regular queries, for example, “evaluate [company] by [criterion]” or “. my location is usa.” They repeat daily for months, indicating automation.
Full instructions for agents
Machines sometimes log entire prompts, for example, “find on the web… return 3 most relevant results… don’t invent URLs.” Such strings end up in the report as queries.
Errors and table headers
People and pipelines search as-is. In my data, there was an entire row from a ranking tracker CSV file that received 146 impressions.
Long queries without markers
Phrases of 10+ words without obvious signs of dialogue. The classifier sends them for manual review, as they could be either AI fragments or regular long-tail queries.
How to distinguish AI fragments from regular queries
The key difference is the addressee. A regular long query is addressed to no one, while a conversational one is directed at a conversational partner. Four signals reveal the conversational nature:
- presence of an address;
- interrogative construction;
- incomplete sentences;
- conversational particles.
Length itself is a weak signal, so queries of 10+ words with interrogative intonation are classified as conversational, while those without markers are sent to quarantine.
Practical conclusions from the analysis
On my site, AI fragments began appearing in March 2026, steadily 20–30 impressions per month. For example, the query “yes” gave 110 impressions and 6 clicks at an average position of 4.5 — this is impossible in regular search results, but typical for an AI answer block.
Comparative queries perfectly correlate with my articles: “what about Claude?” led to the WebMCP guide, “what about Xcode?” — to the post about Xcode. This shows that users actively compare alternatives, and your content can be mentioned in AI answers.
Frequently asked questions
Can AI Overviews be separated from AI Mode in the data?
No, Google combines them in the report as web search. The fragment indicates the fact of a dialogue, but not the specific type of surface.
Why does Google hide queries and clicks in the Generative AI report?
The report shows impressions, pages, countries, devices, and dates, but not queries and clicks. The API does not support this data, and the BigQuery export does not contain an AI column. The only way to get data is the export button in the interface.
How to set up a classifier for your site?
The classifier genai_conversation_queries is available in the free Search Console MCP v2.4.0. Install MCP with one command, select your project, and run the analysis. For large sites, use the BigQuery version v4.1.0.
Conclusion
Analyzing AI fragments in Search Console provides unique insights into how users interact with AI and find your content. Use the classifier to uncover hidden opportunities and optimize content for AI dialogues.
Start with the free tool and get data that most SEO specialists miss.
Leave a Reply