Who does Gemini cite for your query, and why not you?
Run a search query through Gemini's live grounding several times. See which pages and sentences it keeps using, compare your page against them claim by claim, and test a rewrite before you publish.
Grounding report
Test a rewrite against the same grounding
Edit your page text below and run the comparison again. It reuses the grounding above, so this step costs no new Gemini call and changes nothing on your website.
Add facts only after verifying them. A better score isn't proof that Gemini will cite you. Publish it, then confirm with a fresh check once Google has recrawled the page, because Gemini reads Google's index rather than your live page.
Keep this report
Every export is built in your browser from this run. Nothing is stored on the server, so save what you want to keep. The HTML report is the page above as one file; Markdown suits a ticket or a note; JSON and CSV carry every judgement.
If this tool saves you time, you can sponsor the work on GitHub.
How this works
Observe
The query runs through Gemini with Google Search grounding, the same grounding system behind its AI answers, once per run. The report counts the runs each source, claim and search appeared in, merges claims that state the same fact across runs, and shows the sentence on each cited page that carries each claim.
Compare
TypeSafe's Jev judges closed questions over your page: which sentence states each grounded claim, whether your page answers each search Gemini ran, and whether your best sentences stand alone and state a concrete fact.
Fix and retest
Each gap becomes one edit, quoting the claim the answer used. Test a rewrite against the same grounding in seconds, then confirm with a fresh check once Google has recrawled the published page, because Gemini reads Google's index rather than your live page.
Gemini's API grounding is a close proxy for AI Mode and AI Overviews, not the same system, and grounding varies from run to run, which is why a check is several runs. Runs happen within a minute of each other, so they measure that variation, not how grounding drifts over weeks. Sentence scores describe the judge's probability for a call, not measured accuracy.
Method and limitations
“Missing” means no sentence in the text checked states the claim. Claims come from Gemini's answer and may be wrong; verify a fact before adding it. Competitor sentences are located on their pages by the same judge and can be imperfect.
Your page is read from its server HTML without running JavaScript, the way the crawlers behind ChatGPT, Claude and Perplexity read it. Gemini works from Google's index, and Google renders JavaScript before indexing, so a page built by JavaScript is flagged and pasting its text compares the full page. Content behind a login or inside images isn't seen. Up to 120,000 characters of page text are compared. Checks are limited per connection and per day to keep the tool free.
Check these next
- AI Mode and AI Overview Query Classifier
Sorts the AI Mode conversations out of your Search Console queries and labels every one.
- Keyword Relevance Filter
Filters your Search Console queries by business relevance and own brand, with editable labels and CSV or JSON exports.
- AI Crawler Access Checker
Probes your site as GPTBot, ClaudeBot and PerplexityBot and finds the blocks that keep you out of AI answers.