Tokens, explained: how to get the most out of SOMM DIGI AI (and any AI)

The short answer: a token is a small piece of text that an AI reads or writes: a short word, part of a longer word, or a punctuation mark. Your question and the AI's answer both count. OpenAI's rule of thumb for English is that one token is about four characters, or three-quarters of a word, so 1,000 tokens is roughly 750 words. Every token costs the app real computing power, which is why AI apps set limits.

What's a token?

• Short, common words are often a single token.
• Long or unusual words are split into pieces. Gewürztraminer or Châteauneuf-du-Pape may take several tokens each.
• Other languages often need more tokens for the same meaning than English does. The four-characters rule is for English.
• Both directions count: the words you send and the words you get back.
• Conversations add up. In most AI chat apps, the model rereads the conversation so far each time you send a message. A very long thread costs more per question than a fresh one.

Why AI apps limit tokens

Every answer runs on powerful computers, and an AI app pays for the tokens its model reads and writes. SOMM DIGI AI runs on today's best AI models, so every token has a real cost. Limits by plan keep the app fast for everyone, fair between users and affordable to run, instead of raising prices for all or cutting the quality of the answers.

SOMM DIGI AI is free to start, with no card. Plans and allowances can change, so I don't list them here: the current ones are on sommdigi.ai.

Seven habits for better answers with fewer tokens

• Be specific. Instead of “Tell me about Burgundy”, ask “Flashcards on the red Grand Crus of the Côte de Nuits, Certified level.” Vague questions get long, general answers.
• Say the format and the length. “In five bullet points.” “As a table.” “Under 100 words.” You get what you need and nothing more.
• Group related topics. “Flashcards on Barolo, Brunello di Montalcino and Chianti Classico, Certified level” is one request instead of three.
• New topic, new chat. Moving from Burgundy to Sherry? Start fresh, so the model isn't rereading a long Burgundy thread.
• Refine instead of re-asking. “Shorter.” “Only the Grand Crus.” “Now as questions.” A short follow-up costs less than repeating the whole prompt.
• Paste only what matters. One page of your notes, not the whole file.
• Save the good answers. Copy them into your notes or flashcards. Rereading your own notes costs nothing, and repetition is what makes it stick. Here is how to build your own wine data.

If you run out

• Review what you saved. Your notes and flashcards are the best revision anyway.
• Use the free Prompt Guide in another AI. It gets sommelier-level wine answers out of ChatGPT or Claude too.
• Need more? See the current plans at sommdigi.ai, or write to me at info@sommdigi.com.

Limits aren't there to stop you studying. They're there so the tool stays good, fast and affordable for everyone who uses it.

Questions people ask