How to Track Google AI Mode Traffic

Google has introduced dedicated reporting for generative AI features in Search Console, giving website owners more insight into how their pages appear in features such as AI Overviews and AI Mode. However, there is still a gap when it comes to seeing the exact search queries behind that activity.

Google’s Search Console reporting can show AI-related performance, but the underlying query data is not fully available through the Search Console API or BigQuery exports. This has led SEOs to develop different ways of identifying conversational and AI-related searches within their existing Search Console data.

Why AI Queries Are Difficult to Track

AI Mode searches can look very different from traditional Google searches. Instead of short phrases such as “best mortgage rates UK”, users may enter complete questions, detailed instructions or follow-up messages.

Some examples can even look like parts of an AI conversation, such as:

“Yes, go ahead”
“Can you give me more options?”
“What about the cheaper one?”
“Write this as an email”
“Explain this in simple terms”

These searches can appear in Search Console, but identifying which ones came from AI Mode is not always straightforward.

Google also anonymises some rare queries, meaning that any method used to identify AI-related searches will only provide part of the overall picture.

Four Ways to Find AI Mode Queries

As SEOs have investigated the data, several methods have emerged for finding possible AI Mode and AI Overview queries.

1. Export Your Full Query Data

One approach is to move beyond the standard Search Console interface and export a larger set of query data.

The Search Console interface limits the number of rows that can be viewed at once, which can make it difficult for larger websites to analyse their complete query profile. Exporting the data through tools such as Excel, the Search Analytics API or other reporting platforms allows SEOs to work with a much larger dataset.

Once exported, the queries can be reviewed and sorted to identify searches that appear conversational or AI-generated.

2. Use a Custom Regex Filter

Another option is to use a custom regular expression directly within the Search Console query filter.

A regex can be designed to identify common conversational phrases and instructions, including words such as “write”, “generate”, “explain”, “summarise” and “draft”. It can also look for conversational responses such as “yes”, “okay”, “continue” and “show me more”.

This approach is quick and free, making it useful for SEOs who want to investigate AI-related queries without using additional software.

However, regex has limitations. It relies on predefined patterns, so unusual queries, non-English searches, mixed-language phrases and other AI-related strings may not be detected.

3. Use Search Console Visualisation Tools

Some third-party tools can make Search Console data easier to explore by adding charts, filters and other visualisation features.

These tools can help SEOs identify unusual changes in search behaviour and investigate large amounts of query data more efficiently.

The downside is that they may not have a dedicated AI Mode filter. Users may still need to identify AI-related queries themselves using their own rules or filters.

4. Use AI or Machine Learning Classification

A more advanced option is to use machine learning to classify search queries automatically.

Instead of relying solely on specific words or phrases, a trained model can assess the wider meaning and structure of a query. This can make it easier to identify unusual conversational searches, follow-up prompts, copied text and other patterns associated with AI search.

A classification system could separate queries into categories such as:

Full conversational searches
Short conversational replies
Follow-up questions
Rank-tracking queries
AI agent prompts
Pasted text
Ordinary searches

This approach can be particularly useful for websites with tens of thousands of queries, where manually checking each search would be impractical.

Why Traditional Filters Are Not Always Enough

Regex and keyword-based approaches are useful starting points, but they cannot identify every type of AI-related search.

For example, an AI-related query may not contain obvious words such as “write”, “generate” or “explain”. Users may also paste long pieces of text into Google or use a mixture of languages.

This makes classification more complicated, particularly for websites receiving traffic from different countries and languages.

Machine learning can help by looking at the overall structure and context of a query rather than simply searching for individual words.

What This Means for SEOs

The growing availability of AI search data gives SEOs another way to understand how people are interacting with Google.

AI Mode is changing the way users search, with longer and more conversational queries becoming increasingly common. Monitoring these searches could help website owners understand what people are asking, what information they are looking for and how their content is being surfaced in Google’s AI features.

However, the data should be treated as an indication rather than a complete picture. Google’s anonymisation of rare searches means some AI-related queries will remain hidden.

For now, the most practical approach is to combine Search Console’s generative AI reporting with query exports, filters and classification methods. This can provide a clearer view of AI-related search activity while Google continues to develop its reporting tools.

The Future of AI Search Reporting

Google’s introduction of generative AI reporting is an important step towards making AI search performance easier to measure. However, SEOs still need more detailed query-level information to understand exactly how users are discovering their websites through AI Mode and AI Overviews.

As Google’s AI search products continue to develop, more detailed reporting could eventually make third-party workarounds less necessary.

Until then, website owners can use a combination of Search Console exports, custom filters and automated classification to uncover useful patterns within their existing data.