Loading Embedding to Text...

How to Match an Embedding Back to Text

Step 1

Paste Your Vector

Drop a vector into the top box — a JSON array of numbers like [0.12, -0.03, ...]. The easiest way to get one is our Text Embedding tool: embed a sentence there, copy the vector, and paste it here. Or just click Load Sample to try it instantly.

JSON array: the same format the Text Embedding tool gives you
Load Sample: fills in a real vector and a few candidates so you can see it work
Step 2

Add Candidate Texts

In the second box, list the texts you want to match against — one per line. These are the pool the vector gets compared to. You can add up to 50. The tool can only pick from what you give it; it can't invent text out of thin air.

One per line: each line is a separate candidate
Up to 50 candidates: plenty for a quick lookup
Step 3

Click Find Matching Text

Each candidate gets embedded with the same model, then compared to your vector using cosine similarity. Behind the scenes it's the same math you'd use for semantic search — closeness of meaning, not matching words.

Same model: candidates are embedded the same way your vector was
Cosine similarity: the standard way to measure how close two vectors are
Step 4

Read the Ranked Matches

The candidates come back sorted, best match on top, each with a similarity percentage. The top one is the text your vector most likely represents. Copy the ranking or download it as JSON.

Best match highlighted at the top of the list
Similarity score for every candidate, so you can judge confidence
Copy or download the ranking for your own use

What does "Embedding to Text" actually mean?

Let's be honest up front: you can't take a bare vector and rebuild the exact sentence that made it. An embedding is a one-way trip — the model keeps the meaning but throws away the words. There's no button that decodes numbers back into the original text.

So this tool does the thing that does work: matching. You give it a list of candidate texts, it embeds each one, and it finds which candidate's vector is closest to yours. If the original text is in your list, it'll come out on top. This is exactly how semantic search works — you compare a query against a set of known texts.

Closeness is measured with cosine similarity. Want to make a vector to test with? Head to the Text Embedding tool, embed a sentence, and paste the result here.

Frequently Asked Questions

Can you rebuild the original text from just a vector?

No. Embeddings are one-way — the model keeps the meaning but drops the exact words, and nothing reverses that. The most you can do is match the vector to texts you already have, which is exactly what this tool does.

Why do I have to provide candidate texts?

Because matching needs something to match against. The tool compares your vector to each candidate and ranks them. Without candidates there's nothing to compare to — a vector on its own is just numbers.

How is the closest match decided?

With cosine similarity. Each candidate is embedded and compared to your vector; the higher the score, the closer in meaning. Results are sorted with the best match on top.

Does the vector need to come from the Text Embedding tool?

It needs to be the same size (number of dimensions) as the vectors this tool produces, so the safest bet is a vector from our Text Embedding tool. A vector from a different model won't line up and the tool will tell you if the size doesn't match.

Is it free?

Yes, free and no signup. Paste a vector, add candidates, get the ranking. You might also like our Text Embedding generator and the JSON formatter.