Vector Similarity
Want to know if your page is semantically aligned with what your target keyword actually means - not just whether the keyword appears on it? This tool measures it the same way search engines do, using real vector embeddings.
Why I built this
Modern search isn't pure keyword matching anymore. Google's models read your content the way a person would: they turn every page into a vector that captures its meaning, then compare that vector to the meaning of the search query. If your page is semantically close to the query, you have a shot at ranking. If it's drifting off-topic, you don't, even when the keyword appears fifty times.
I wanted a way to measure this directly, so I built a small tool that takes any page or text snippet, turns it into an embedding vector, and computes the cosine similarity against your target keyword. You get a score from 0 to 1 - simple, but useful.
The tool is heavily inspired by iPullRank's Orbitwise, which pioneered this kind of “relevance engineering” - moving beyond keyword density to make sure content is semantically aligned with what search engines actually look for. The big difference: this tool uses Google's Vertex AIas the default embedding model. It's the closest publicly available proxy to the algorithm Google itself uses in Search, which means the scores you get here track much closer to the signal that actually ranks pages.
What it helps you do
Vector similarity scoring
Get a cosine similarity score (0 to 1) between your page and your target keyword. Higher means more semantically aligned. Track the score over time as you optimize.
Compare against the SERP
In multi-site mode, the tool pulls the top 10 ranking URLs for your keyword and runs the same check on each one. See exactly where you stand and which competitor is closest to the semantic core of the query.
Choose your embedding model
Vertex AI by default - the closest match to what Google uses in Search. MixedBread available as a faster, free alternative when you just need a quick sanity check.
Built for relevance engineering
Use the score to map keywords to the right pages, fix cannibalization (two pages with similar scores for the same keyword is usually a sign), or pick semantically relevant pages for internal linking.
How to use it
- 1
Enter your target keyword
Type the query you're trying to rank for. The tool turns it into an embedding vector, which becomes the semantic anchor for everything that follows.
- 2
Pick an embedding model
Vertex AI is the default and what I recommend - it's closest to what Google uses in Search. MixedBread is the lighter, faster option for quick checks.
- 3
Single-page mode: paste a URL
The tool fetches the page content and shows you the similarity score immediately. Useful for evaluating one page or a draft you're about to publish.
- 4
Multi-site mode: let the tool pull the SERP
Switch modes and the tool fetches the top 10 ranking results for your keyword via Google Custom Search, runs the similarity check on each, and highlights where your page sits in the chart.
- 5
Read the score and act on it
Below 0.7 probably off-topic. Between 0.7 and 0.85 relevant but improvable. Above 0.85 well aligned. Past comparisons are saved in the History tab so you can re-run after a rewrite and see whether your changes moved the needle.
Think this tool can be improved? Let me know how.
I use this tool every day. If you spot a bug, miss a feature, or have ideas for what to build next - I want to hear about it. Pick whichever way is easiest for you.