How AI tokens are counted

Where pages meter your output, AI tokens meter your use of the AI Assistant. Every time the assistant reads your template and writes a change, that work is measured in tokens and subtracted from your monthly token balance.

The token balance

Tokens work exactly like pages — a running balance, not a percentage:

1. At the start of each billing period, a Monthly fee entry adds your plan’s included tokens.
2. Every AI request subtracts the tokens it consumed.
3. Any tokens you buy on top are added to the balance too — and unlike your monthly allowance, those don’t expire at the reset date. Purchased tokens are held separately from your subscription and roll over until you use them.

On the Free plan the monthly entry adds 1,000,000 tokens. Settings → Usage → Overview shows Used tokens, Available tokens, and the date the allowance resets.

What consumes tokens

Action	Consumes tokens?
Asking the AI Assistant to generate a template from a prompt	Yes
Asking the assistant to change an existing template (“make the totals bold”)	Yes
Generating a template from a file	Yes — logged separately as Document analysis
Editing the template, Sample Data, or Translations by hand	No
Previewing, test rendering, or production rendering	No — renders consume pages, never tokens
Undoing an AI change	No — but the tokens already spent are not refunded

The distinction that matters: the assistant costs tokens, rendering costs pages. They are separate allowances and running out of one does not block the other.

Why the cost per request varies so much

Token cost is not per message — it scales with how much the assistant has to read and write. In practice a single request can range from under 3,000 tokens to almost 200,000:

Kind of request	Typical cost
A small, targeted tweak on a simple template	~2,500 – 6,000
Analysing an uploaded document (Document analysis)	~9,000 – 10,000
Generating a substantial template from a prompt	~25,000 – 50,000
A change against a large, complex template	~150,000 – 200,000

Two things drive the number up:

• The size of the template. The assistant has to read the whole template definition before it can change anything, so the same instruction costs far more on a long, complex template than on a short one.
• The size of the change. Rewriting a whole layout writes far more output than renaming a field.

Note

These figures are indicative, taken from a real ledger — they are not a price list. Use the Activity ledger for your own numbers rather than budgeting from the table above.

Reading the token ledger

Settings → Usage → Activity → AI tokens records every token movement.

Each row shows:

Column	Meaning
Type	Monthly fee (allowance added), AI usage (an assistant request), or Document analysis (an uploaded file being read)
Template	The template the request applied to — blank for requests not tied to one
Date	Timestamp, to the millisecond
Tokens used / Tokens added	The movement itself
Balance	Your remaining tokens after this entry
Cost / M tokens	What you were charged per million tokens, for tokens billed beyond your included allowance

Like the Pages ledger it is newest-first, so read bottom-up to follow the balance forward.

Tip

If your token balance dropped faster than expected, sort through the AI usage rows and look at the Template column. A handful of requests against one large template will usually account for most of it.

Keeping token use down

• Be specific. “Make the invoice totals bold” costs less and works better than “improve this”. See Writing good AI prompts.
• Edit small things by hand. Changing a colour, a label, or a margin in the editor or the Page Properties form costs nothing.
• Batch related instructions into one request instead of a back-and-forth of small ones — each round trip re-reads the whole template.
• Start from an example or a file rather than describing a complex document from scratch.

Running out of tokens

When the balance reaches zero the AI Assistant stops until you top up or the period resets. Manual editing and rendering keep working — only the assistant is blocked. To carry on:

• Wait for the reset date shown on the Usage page.
• Buy extra tokens to add to the current balance.
• Upgrade your plan for a higher monthly allowance.

Related articles

• How pages are counted — the other thing Pagr meters.
• Buy extra credits — top up tokens mid-period.
• Writing good AI prompts — get more out of each request.
• Generate a template from a file — what Document analysis entries come from.
• Billing — your plan’s included tokens.