Google Search Console now gives ecommerce SEO teams a partial view of a new kind of discovery behavior: prompts submitted through Google’s AI Mode can appear in Performance reporting as queries. Practical Ecommerce reports that Google’s John Mueller confirmed this inclusion and demonstrated a regular-expression filter for finding likely prompts.
The important word is likely. Search Console does not provide a dedicated AI Mode dimension, and prompt-like wording is not proof that a query came from AI Mode. The practical opportunity is therefore not precise channel attribution. It is to create a repeatable segment of conversational, question-led queries, inspect the needs they reveal, and compare those needs with the store’s existing category, product, buying-guide, and support content.
What Google confirmed—and what remains unavailable
The confirmed capability is narrow but useful: AI Mode prompts are included in Search Console query data. Teams can consequently encounter those prompts in the same reporting environment used for conventional Google Search queries. Mueller also showed that a custom regex can help isolate queries beginning with common question words.
This does not mean Search Console labels individual rows as AI Mode traffic. A filter based on language patterns identifies candidates, not a verified source. Someone can enter “how,” “what,” or “which” in conventional Search, while an AI Mode prompt may begin with none of those words. Query filtering will therefore produce both false positives and false negatives.
Nor does the reported capability establish complete prompt visibility or AI Mode conversion attribution. Search Console applies its normal reporting constraints, including privacy-related query omissions, and its data should not be presented as a complete transcript of customer activity. Ecommerce teams should describe the segment internally as “likely conversational or AI Mode-style queries,” unless Google introduces an explicit traffic classification.
Build a repeatable candidate-query filter
Start in Search Console’s Performance report and select the relevant search property, date range, and search type. Open the Queries view, add a query filter, choose the custom regex option, and use a pattern based on prompt-like openings. The source gives this example:
^(what|where|why|how|when|which|who|can|are|does|do|is|should|could|would|won't|weren't|wasn't|shouldn't|couldn't|cannot|can't|did|didn't|will|wouldn't|isn't|aren't|ain't|was|were|won't|shouldn't|couldn't|wouldn't|haven't|hasn't|hadn't|mustn't|needn't|mightn't|shan't|oughtn't|may|might|must|shall|ought|need|have|has|had|won't)\b
This is a discovery filter rather than a universal standard. Language, market, spelling, punctuation, and customer vocabulary will affect what it captures. International stores should build separate patterns for each reporting language instead of translating one expression mechanically. Teams should also retain the exact regex, property, date range, country, device, and search type used so later exports remain comparable.
Export the filtered rows into a spreadsheet or reporting workspace. Preserve the original query and its Search Console measures, but add human-review fields such as intent category, relevant page, content gap, commercial relevance, and confidence that the wording is genuinely conversational. Reviewers should be able to distinguish observed Search Console data from their own classification.
Classify prompts by ecommerce decision, not wording alone
A long list of question words is not yet an optimization plan. Categorize each candidate according to the decision the shopper appears to be making:
- Product selection: “which” and “what” queries comparing types, features, sizes, or use cases.
- Compatibility and fit: questions about whether an item works with another product, body measurement, space, system, or environment.
- Usage and care: “how” prompts concerning setup, maintenance, cleaning, storage, or troubleshooting.
- Policy and reassurance: questions involving delivery, returns, warranties, authenticity, safety, or availability.
- Exploratory discovery: broader requests in which shoppers describe a problem, recipient, occasion, budget, or desired outcome rather than naming a product.
Then map each query cluster to the page type that should answer it. Selection questions may belong in a buying guide or category introduction. Fit and specification questions usually require accurate product-page content. Repeated care questions may justify a guide linked from relevant products. Policy questions should point toward authoritative service information rather than being duplicated inconsistently across the site.
Do not assume every query needs a new page. Several variants may express one underlying need, and creating near-duplicate articles can fragment useful information. A stronger response may be to improve a category page’s decision support, clarify a specification table, add a comparison module, or connect an existing guide more visibly to products.
Turn the review into prioritized optimization work
Prioritization should combine the Search Console evidence with business context. Begin with clusters that have a clear relationship to products or categories, expose an obvious information gap, and can be answered responsibly. Then inspect the pages already receiving impressions for those queries. A mismatched landing page can indicate that Google lacks a more suitable answer or that the intended page does not communicate its relevance clearly.
For product pages, verify that essential claims—dimensions, materials, compatibility, included components, care, and restrictions—are explicit and supported by the product record. Do not add speculative FAQ answers merely because a prompt appeared. For category pages, consider whether shoppers can understand differences among product types and move from a described need to an appropriate selection. Editorial guides can address multi-step decisions, but they should link to current categories and avoid recommendations that become inaccurate when inventory changes.
FAQ additions are appropriate when a question is recurrent, product-specific, and answerable in a concise way. They are less useful when they merely restate marketing copy or conceal critical specifications below the main content. The objective is better decision support, not manufacturing question-shaped text.
In SEO reporting, maintain this candidate segment alongside—not in place of—standard query groupings such as branded, non-branded, category, product, and informational searches. Report changes as movements in the filtered dataset, not as definitive AI Mode growth. Annotate regex changes because expanding the pattern can create an apparent increase that reflects methodology rather than customer behavior.
Guard against overreading an imperfect segment
The chief risk is false precision. The filter cannot prove that a particular prompt originated in AI Mode, reveal the full generated response, show which products were mentioned within that response, or establish that an AI Mode interaction caused a sale. Search Console query performance and ecommerce analytics may support separate observations, but the source does not establish prompt-level revenue attribution.
Short reporting periods can also encourage teams to chase isolated wording. Look for recurring needs and coherent clusters rather than treating every row as a content brief. Query omissions mean absence is not evidence that customers never ask a question. Conversely, an impression does not automatically justify content production; the request may be irrelevant, too distant from the assortment, or unsafe to answer without specialist review.
Use the workflow as a monthly or quarterly discovery exercise: apply a documented filter, export candidates, remove obvious conventional-query noise, cluster by shopper decision, map clusters to existing pages, and assign only the highest-value gaps. Record whether the response is a product-data correction, category enhancement, guide update, FAQ addition, internal-linking change, or no action.
The defensible takeaway is modest but actionable. Search Console can surface AI Mode prompts among query data, while a language-based filter can help ecommerce teams investigate them. Treating that filter as a research lens—not an attribution system—allows operators to learn from emerging discovery patterns without confusing inference with confirmed reporting capability.
