Back to tools

03 / Tools

Grounding Query Classifier

Before you optimise an AI-visible query, ask whether the answer is likely coming from retrieval or model memory. Instant heuristic classification runs in your browser — no API key needed. Optionally confirm with a Gemini probe using your own key.

How to use this tool Privacy notice

Guide

How to use the Grounding Query Classifier

Most AI visibility work starts too late. Teams test prompts and passages before asking whether the query is even optimisation-sensitive. This tool answers that question first: is the model likely to retrieve external sources, or answer from memory? For the full strategic context, read Grounding vs. In-Model.

1. Paste your query

Enter the query you care about. Add an optional market (e.g. UK) and brand context if they matter for the intent. The heuristic classifier runs instantly in your browser — no API key needed.

2. Read the classification

You get three possible outcomes:

  • Likely Grounded — retrieval-sensitive; page-level SEO and passage work may influence the answer.
  • Likely In-Model — memory-led; entity authority and corroboration matter more than on-page retrieval optimisation.
  • Mixed / Model-Dependent — split the query and test sub-intents separately.

Grounding likelihood is a 0–100 score for retrieval sensitivity. Confidence reflects how clearly the signals align — not certainty across all models.

3. Expand the signal breakdown

Every result shows which signals fired: freshness, locality, comparison intent, evergreen phrasing, regulated topics, and more. Each signal shows direction, weight, and a plain-English explanation. This is a transparent classifier, not a black box.

4. Follow the next action

Grounded queries link forward to the Fan-Out Gap Analyser and SRO Snippet Tester. In-model queries route toward entity reinforcement and corroboration. Mixed queries include suggested subqueries you can classify with one click.

5. Optional: Gemini live probe

For advanced validation, add your Gemini key in Settings and run a live probe. The tool checks whether Google Search was invoked for your query and compares that with the heuristic result. Your key stays in your browser unless you opt in to remember it on this device.

What this tool does not do

It does not predict citations with certainty, replace live testing across models, or guarantee how Google AI Mode or ChatGPT will behave tomorrow. It gives you a fast, defensible first pass so you do not waste optimisation effort on memory-led queries.

Questions or edge cases? Get in touch.