AI training data value calculator
AI developers buy data by the token, and the price per token changes a lot depending on what the data is and how rare it is. Enter how much of each kind of material your company has, and the calculator converts it to tokens, applies a typical price range for that kind of data and adjusts for how unique and clean it is. You get a total value, the average price per million tokens and a breakdown that shows which material is worth the most.
Tokens and value by material
Amount times tokens per item gives the tokens. Tokens times the price range, times your multiplier, gives the value.
| Material | Amount | Tokens | Price per 1M tokens | Low | High | Share |
|---|---|---|---|---|---|---|
| Total |
How the value is worked out
Buyers compare datasets by price per million tokens, so the calculator converts everything into tokens first and prices each kind of material at its own rate.
The rates are typical ranges seen in licensing of private business data. Public web text sells for far less; rare expert material can sell for more.
Convert to tokens
Each amount is multiplied by the average token count for that kind of item, for example 500 tokens for a page or 12,000 for an hour of transcribed talk.
tokens = amount × tokens per item
Price by kind of material
Each kind has its own low and high price per million tokens. Labelled examples and solved support cases sit at the top, bulk email and chat at the bottom.
value = tokens ÷ 1,000,000 × price per 1M
Adjust for uniqueness and rights
Data that is already public is cut to 60 or 20 percent of its value. Unclear ownership cuts it further, because a buyer cannot use what you cannot license.
Adjust for industry, language and condition
Industry sets 0.9 to 1.4×. The English share scales value between 40 and 100 percent. Cleaned and labelled data earns 25 percent more, raw exports 30 percent less.
multiplier = public × industry × english × rights × condition
Token sizes and prices used
| Material | Tokens per item | Price per 1M tokens | Price per item |
|---|---|---|---|
| Documents, manuals and reports | 500 per page | $2,000 to $12,000 | $1 to $6 |
| Emails | 160 per email | $15 to $80 | $0.0024 to $0.0128 |
| Chat messages | 25 per message | $30 to $150 | $0.0008 to $0.0038 |
| Support tickets and Q&A | 600 per ticket | $300 to $1,800 | $0.18 to $1.08 |
| Source code | 10 per line | $300 to $1,500 | $0.003 to $0.015 |
| Call and meeting transcripts | 12,000 per hour | $500 to $3,000 | $6 to $36 |
| Expert-labelled examples | 300 per example | $2,000 to $10,000 | $0.6 to $3 |
Questions about AI training data value
How do AI companies price training data?
Most licensing deals are negotiated per dataset, but buyers compare them by working out the price per million tokens. A token is roughly three quarters of a word. Common text such as chat costs little per token, while rare expert material such as solved support cases, specialist manuals or labelled examples costs many times more.
What is a token and how are my files converted?
A token is the unit AI models read, about four characters or three quarters of an English word. The calculator uses typical sizes: a page of a document is about 500 tokens, an email about 160, a chat message about 25, a line of code about 10 and an hour of recorded conversation about 12,000 once transcribed.
Why is data that is already public worth less?
Large models have already been trained on the public web. If your manuals are on your website or your code is on GitHub, a buyer has probably seen it already, so it adds little. Private material that only exists inside your company is what buyers cannot get anywhere else.
Which of my data is worth the most to AI developers?
Per token, expert-labelled examples and complete problem and solution records, such as support tickets or transcribed calls with outcomes, are the most valuable. They teach a model how a task is done, not just what words look like. Bulk email and chat is cheap per token but often large enough to add up.
Do I lose my data if I license it for AI training?
No. A training license gives the buyer a cleaned copy to train on. You keep the originals and keep using them. The agreement sets what the buyer may do with the copy, how long they may keep it and whether you can license the same data to others.
How reliable is this figure?
It is a planning estimate based on typical price ranges, not a quote. Real prices depend on a buyer's current needs, how clean your data is and the deal terms. A buyer will always ask for a sample before making an offer.