Spotibo Monitoring
Sign upSign in

Semantic Relevance Checker

About, how to use

You can paste formatted content copied from a web page. It becomes editable Markdown here, directly in your browser.

0/20,000 characters
AI provider
Analysis mode
Saved keys are encrypted in an HTTP-only cookie for 30 days and shown only in this masked field.

Why use this tool?

For AI SEO, paste formatted page content directly from your browser, keep its headings and structure as editable Markdown, and see each section's semantic relevance  in a color-coded view. You can sort the sections by score and inspect whether a preceding heading improves a continuation's relevance. This helps you find passages to clarify, expand, or reorganize when reviewing a page.

Metehan Yeşilyurt's September 2026 report on leaked ChatGPT retrieval data described fetched pages being split into passages and scored separately. Smart sections mode offers an approximate, transparent version of this reported chunking approach.

Separately, the 2024 Google Content Warehouse API documentation leak lists pageEmbedding, siteEmbedding, and siteRadius. These fields suggest that Google uses vector embeddings to represent content at both the page and site level; siteRadius describes how far page embeddings deviate from a site embedding.

How AI search may process content

A common retrieval-augmented workflow roughly follows these steps:

  1. Extract readable content from relevant sources.
  2. Split the selected text into smaller chunks while keeping context such as headings.
  3. Find relevant chunks using semantic similarity and other retrieval methods.
  4. Select or rerank useful chunks and provide them to a language model to help compose an answer.

How to use the checker

  1. Paste your source content and enter a query or text to compare. Rich content copied from a web page is converted to Markdown locally in your browser, so you can review and edit it before analysis. Plain text stays as pasted.
  2. Choose Gemini or OpenAI, enter your own API key, and select an analysis mode. Then click Analyze.
  3. Check the overall similarity and the color-coded passages. Higher percentages indicate a closer semantic match. You can sort section results by relevance.

Analysis modes

  • Smart sections (default): starts sections at Markdown headings and splits long sections near your character target (1,100 by default), favoring paragraph or sentence boundaries. We use character counts instead of token counts for simplicity. This mode was inspired by Metehan Yeşilyurt's report on a ChatGPT UI retrieval leak, which described fetched pages split into passages and scored individually. This checker uses its own heading and length rules with Gemini or OpenAI embeddings; it does not reproduce ChatGPT's retrieval system.
  • Sections: treats text separated by empty lines as separate sections.
  • Whole text only: scores the two complete texts without passage results.
  • Whole text + sentences: scores the complete texts and highlights each sentence separately.

The scores are estimates based on the selected provider's embeddings. Compare results within the same provider; they are not search rankings or ChatGPT's internal relevance scores.

Your API key and content

Your API key is not saved by default or stored in our database. If you choose “Remember this API key on this device,” the key is stored for 30 days in an encrypted, HTTP-only browser cookie. Your source text and comparison text are sent to the selected provider to generate embeddings.