---
title: "Token waste calculator: how many tokens your team burns re-reading the same PDF | Brana"
canonical_url: "https://brana.one/token-waste-calculator"
last_updated: "2026-11-03T23:13:56.084Z"
locale: en
meta:
  description: "Free token waste calculator. Pick a PDF, slide deck or spreadsheet, or drop one in to measure it in your browser, and see how many tokens your team spends re-reading it with every question."
  "og:description": "It's not your employees. It's the PDFs. See how many million tokens your team spends re-reading the same document."
  "og:title": "Token waste calculator: how many tokens your team burns re-reading the same PDF"
---

Brana — home

Brana on LinkedIn

Token waste calculator

# It's not your employees.<br>It's the PDFs.

And the spreadsheets, slide decks and screenshots. The AI keeps no memory between messages, so every question re-reads the whole document and everything said so far. Then every colleague uploads the same file to their own chat and starts again. Pick a document and watch the tokens add up.

## 01**Ask it something. Then ask again.**

Two chats about the same 40-page PDF: on the left it's attached to the chat, on the right it's in Brana. Each click asks both chats the same question. Watch the usage budget.

~~Use your own document ↓~~

Without Brana

**PDF** 40-page PDF attached · 90K tokens

Question 1 · read 90.1K tokens

Question 2 · read 90.6K tokens

Question 3 · read 91.2K tokens

Usage budget27% used

<dl>

<dt>Document</dt>
<dd>270K</dd>

<dt>Conversation</dt>
<dd>1.9K</dd>

<dt>Answers</dt>
<dd>1.5K</dd>

</dl>

With Brana

**B** 40-page PDF is in Brana

Question 1 · read 3.6K tokens

Question 2 · read 7.1K tokens

Question 3 · read 10.7K tokens

Usage budget2% used

<dl>

<dt>Passages</dt>
<dd>18K</dd>

<dt>Conversation</dt>
<dd>3.4K</dd>

<dt>Answers</dt>
<dd>1.5K</dd>

</dl>

Ask another question about the 40-page PDF…

Illustrative budget of 1M tokens. Claude and ChatGPT don't publish their limits in tokens, but every plan has one, and this is how a chat about a file eats it.

## 02**Now multiply it by your team**

People who use it10Chats per person, per month8Each new chat starts with a fresh upload.

Every month

**21.9 million tokens to read the same 40-page PDF 240 times.**

10 people, 80 chats a month, 3 questions each. Most of it is the document again, not the answers.

Without Brana

**21.9M**

tokens a month

<dl>

<dt>One chat</dt>
<dd>273.4K</dd>

<dt>Whole document read</dt>
<dd>240 times a month</dd></dl>

With Brana

**1.8M**

tokens a month

<dl>

<dt>One chat</dt>
<dd>22.9K</dd>

<dt>Whole document read</dt>
<dd>Once, when it was added</dd></dl>

**92% fewer tokens.** 20 million tokens a month back in your team's limits.

Even at the low end of each estimate, it's 14.7 million tokens a month.

## 03**Try it with your docs**

Type the size of a real document, or let the page measure a PDF. The chats above switch to it.

PDF · Size (pages)

One copy is about 90,000 tokens.

Measure a PDF

↑Choose a PDF

Read in your browser. The file is never uploaded.

With your 40-page PDF, the attached chat runs out of budget at question 11.

## **Ask AI about token usage**

Check our maths with your own assistant. It opens a new chat with your document, your team and this page.

- [Ask ChatGPT](<https://chatgpt.com/?q=I'm%20checking%20how%20many%20tokens%20my%20team%20wastes%20by%20attaching%20documents%20to%20AI%20chats.%20Using%20the%20token%20waste%20calculator%20at%20https%3A%2F%2Fbrana.one%2Ftoken-waste-calculator%3A%20we%20ask%20about%20a%2040-page%20PDF%20(about%2090%2C000%20tokens)%2C%203%20questions%20per%20chat%2C%2010%20people%2C%208%20chats%20each%20a%20month.%20Explain%20why%20every%20question%20re-reads%20the%20whole%20document%2C%20check%20whether%20these%20assumptions%20and%20numbers%20are%20reasonable%2C%20and%20summarise%20how%20a%20shared%20knowledge%20base%20like%20Brana%20(https%3A%2F%2Fbrana.one)%20would%20reduce%20the%20usage%2C%20with%20bullet%20points%20and%20a%20short%20final%20summary.>)
- [Ask Claude](<https://claude.ai/new?q=I'm%20checking%20how%20many%20tokens%20my%20team%20wastes%20by%20attaching%20documents%20to%20AI%20chats.%20Using%20the%20token%20waste%20calculator%20at%20https%3A%2F%2Fbrana.one%2Ftoken-waste-calculator%3A%20we%20ask%20about%20a%2040-page%20PDF%20(about%2090%2C000%20tokens)%2C%203%20questions%20per%20chat%2C%2010%20people%2C%208%20chats%20each%20a%20month.%20Explain%20why%20every%20question%20re-reads%20the%20whole%20document%2C%20check%20whether%20these%20assumptions%20and%20numbers%20are%20reasonable%2C%20and%20summarise%20how%20a%20shared%20knowledge%20base%20like%20Brana%20(https%3A%2F%2Fbrana.one)%20would%20reduce%20the%20usage%2C%20with%20bullet%20points%20and%20a%20short%20final%20summary.>)
- [Ask Google AI Mode](<https://www.google.com/search?udm=50&q=I'm%20checking%20how%20many%20tokens%20my%20team%20wastes%20by%20attaching%20documents%20to%20AI%20chats.%20Using%20the%20token%20waste%20calculator%20at%20https%3A%2F%2Fbrana.one%2Ftoken-waste-calculator%3A%20we%20ask%20about%20a%2040-page%20PDF%20(about%2090%2C000%20tokens)%2C%203%20questions%20per%20chat%2C%2010%20people%2C%208%20chats%20each%20a%20month.%20Explain%20why%20every%20question%20re-reads%20the%20whole%20document%2C%20check%20whether%20these%20assumptions%20and%20numbers%20are%20reasonable%2C%20and%20summarise%20how%20a%20shared%20knowledge%20base%20like%20Brana%20(https%3A%2F%2Fbrana.one)%20would%20reduce%20the%20usage%2C%20with%20bullet%20points%20and%20a%20short%20final%20summary.>)
- [Ask Perplexity](<https://www.perplexity.ai/search?q=I'm%20checking%20how%20many%20tokens%20my%20team%20wastes%20by%20attaching%20documents%20to%20AI%20chats.%20Using%20the%20token%20waste%20calculator%20at%20https%3A%2F%2Fbrana.one%2Ftoken-waste-calculator%3A%20we%20ask%20about%20a%2040-page%20PDF%20(about%2090%2C000%20tokens)%2C%203%20questions%20per%20chat%2C%2010%20people%2C%208%20chats%20each%20a%20month.%20Explain%20why%20every%20question%20re-reads%20the%20whole%20document%2C%20check%20whether%20these%20assumptions%20and%20numbers%20are%20reasonable%2C%20and%20summarise%20how%20a%20shared%20knowledge%20base%20like%20Brana%20(https%3A%2F%2Fbrana.one)%20would%20reduce%20the%20usage%2C%20with%20bullet%20points%20and%20a%20short%20final%20summary.>)
- [Ask Grok](<https://grok.com/?q=I'm%20checking%20how%20many%20tokens%20my%20team%20wastes%20by%20attaching%20documents%20to%20AI%20chats.%20Using%20the%20token%20waste%20calculator%20at%20https%3A%2F%2Fbrana.one%2Ftoken-waste-calculator%3A%20we%20ask%20about%20a%2040-page%20PDF%20(about%2090%2C000%20tokens)%2C%203%20questions%20per%20chat%2C%2010%20people%2C%208%20chats%20each%20a%20month.%20Explain%20why%20every%20question%20re-reads%20the%20whole%20document%2C%20check%20whether%20these%20assumptions%20and%20numbers%20are%20reasonable%2C%20and%20summarise%20how%20a%20shared%20knowledge%20base%20like%20Brana%20(https%3A%2F%2Fbrana.one)%20would%20reduce%20the%20usage%2C%20with%20bullet%20points%20and%20a%20short%20final%20summary.>)

## **Why documents burn tokens**

Four facts behind the numbers, three of them from the AI vendors themselves.

01

### **Every message re-reads the whole chat**

Anthropic lists conversation length, attachment size and tool use as what drains a usage limit: each message sends everything above it again, files included.

Source: [~~Anthropic~~](https://support.claude.com/en/articles/9797557-usage-limit-best-practices)

02

### **A PDF page is 1,500 to 3,000 tokens**

That's Anthropic's figure for text alone, and each page is also sent as an image on top. A 40-page report is around 90,000 tokens before anyone asks anything.

Source: [~~Anthropic~~](https://platform.claude.com/docs/en/build-with-claude/pdf-support)

03

### **Spreadsheets are worse**

Pasted or attached as text, every cell is read: a number and its separator are a few tokens. A 5,000-row sales export with 12 columns is about 180,000 tokens, twice that report. This one is our estimate.

04

### **Every colleague starts from zero**

Microsoft's open-source TokenEconomizer found coding agents paying again for "the same facts rediscovered" on every turn. Your team does the same with every new chat and every fresh upload.

Source: [~~Microsoft TokenEconomizer~~](https://github.com/microsoft/TokenEconomizer)

> **“It's actually not our engineers that are driving the token consumption.” **

*Accenture's agentic AI strategy lead, in leaked internal audio. The example: non-engineers turning PDFs into slides. [~~Reported by 404 Media~~](https://www.404media.co/the-tokenpocalypse-is-here-companies-are-scrambling-to-stop-spending-so-much-on-ai/)*

## **Questions**

<details>

<summary>

### Why does asking about a PDF use so many tokens?

\+</summary>



Chat models keep no memory between messages. To answer your tenth question, the AI reads the PDF again, plus the nine questions and answers before it. A 40-page PDF is about 90,000 tokens of text, so ten questions read close to a million.

</details>

<details>

<summary>

### How many tokens is a PDF page?

\+</summary>



Anthropic puts it at 1,500 to 3,000 tokens of text per page, depending on how dense it is, plus the page's image. A scanned page with no text layer is read only as an image, around 1,500 tokens.

</details>

<details>

<summary>

### Why do Excel files use even more?

\+</summary>



When a spreadsheet is pasted or read as text, every cell counts, numbers and separators included. Thousands of rows add up to hundreds of thousands of tokens, usually to answer a question about a handful of them.

</details>

<details>

<summary>

### Why do we keep hitting Claude or ChatGPT limits?

\+</summary>



Usually because of repetition, not hard work: the same files uploaded in every new chat, long threads that re-read everything on each reply, and colleagues asking the same document the same questions separately.

</details>

<details>

<summary>

### How does Brana cut the waste?

\+</summary>



Your documents are added to Brana once. When anyone on the team asks Claude or ChatGPT a question, Brana sends only the passages that answer it, a few thousand tokens instead of the whole file, with a link back to the source.

</details>

<details>

<summary>

### Is my PDF uploaded when I measure it?

\+</summary>



No. The page count and the token estimate are worked out in your browser. The file is never sent to us or anyone else.

</details>

## **How we calculated this**

- 01Question k of a chat re-reads the document, the k−1 earlier questions and answers, and the new question. A chat's total is the sum over its questions, so it grows with the square of the questions asked.
- 02We assume a question is about 60 tokens and an answer about 500, and count answers as output too.
- 03With Brana, each question brings back about 3,000 tokens of passages instead of the document, and those passages stay in the chat like any reply. We add 500 tokens per question for Brana's tool description.
- 04Document sizes: PDF text from Anthropic (we use 2,250 tokens a page and leave out the page images); images from Anthropic's vision table; slides and spreadsheet cells are our estimates. A PDF you measure uses its own text, at about 4 characters per token.
- 05Sources checked November 1, 2026. These are estimates for comparison, not a quote: real use depends on the model, the app, and whether it caches.

2026-11-01.phase1-docs

**Stop paying to read the same documents.**

Brana stores your company's documents once and hands Claude and ChatGPT only what each question needs.

[Request an invite](https://brana.one/request-invite)