Guide
How to use the SRO Snippet Tester
AI search does not read your page. It compares short extracted snippets from a shortlist of sources and decides which one to quote when it builds an answer. How often your snippet wins that decision is your Selection Rate — the metric that replaces CTR for grounded AI queries.
This tool simulates that selection step. It is a proxy, not a native Google metric: a capable judge model stands in for Google's actual selection system. Treat the scores as directional, not definitive. For the full strategic context, read CTR Is Dead for AI Search.
1. Add your Gemini API key (Settings)
Why: The tool runs on a bring-your-own-key model so it can stay free and public without being abused. Your Google AI Studio key powers the judge and rewrite passes. It is stored in your browser's localStorage and sent to our API routes only when you run an action. It is never saved on our servers.
Leave the model strategy on Auto. The tool defaults to fast Gemini Flash and automatically escalates to Pro for complex or low-confidence contests, so you get realistic results without overpaying. On a shared machine, clear your key in Settings when you are done.
2. Define the grounded query
Why: SRO is query-specific and passage-specific, never a blanket page score. Enter the primary query, and if you know it, the specific fan-out sub-query the snippet actually competes for. Add the brand or entity you want cited so the tool can check entity salience — whether your passage names it explicitly.
3. Add the competing passages
Why: Models judge a chunk in isolation, not your whole page. Paste the exact extractable passage from your page, then the passages from the sources actually shortlisted for this query. Those are your real competitors — not your usual SERP rivals in general.
- Aim for 60–120 words per passage. That is the extraction sweet spot; the tool shows a live word count and flags passages that drift outside it.
- Fetch passages from a URL if you prefer. With your Gemini key set, the tool fetches the page then uses AI to pick the passages most likely to be selected for your query — quoted verbatim from the page. Without a key it falls back to keyword-based chunking. Always review before testing.
- Add up to five competitors. One target plus three to five rivals is the sweet spot.
4. Run the selection test
Why: The output is hybrid, not a black box. Every passage gets transparent deterministic SRO checks (answer-first directness, entity salience, self-containment, factual density, length fit, query match, clarity) plus a judge-model score that simulates the selection contest. The two are blended into a ranked comparison with an estimated selection rate.
You will see, for each passage:
- Its rank and estimated selection rate against the field.
- Why it wins or loses, in plain language, with specific strengths and weaknesses.
- A factor-by-factor breakdown so you know exactly what to fix.
5. Rewrite and re-test
Why: The value is in the fix, not just the score. Generate an SRO-optimised rewrite of your passage — it applies semantic compression, answer-first structure, entity salience and factual density while keeping your meaning. It never invents statistics; placeholders mark where you should drop in a real figure.
Click Use this & re-test to run the revised passage against the same competitors and see the before/after change in selection probability. Repeat until your snippet consistently wins, then publish and re-test in a few weeks once Google has re-crawled.
What this tool does not do
It does not observe a real Selection Rate (none exists natively), predict citations with certainty, scrape SERPs, or replace traditional ranking work — you still have to be indexed and ranking to be shortlisted at all. It gives you a fast, honest way to test and improve the passage-level selectability that determines whether an AI quotes your content or a competitor's.
Questions or edge cases? Get in touch. This tool is maintained in public and improves from real-world use.