# Stop hitting Claude AI usage Limits | We Call Shotgun > Why document work burns through usage, and how to build a private converter that gives Claude fewer tokens to read. Source: https://wecallshotgun.com/blog/claude-keeps-asking-for-more-money Language: en Author: Toni Dos Santos Published: 2026-10-28 Updated: 2026-10-28 --- Ai # Stop hitting Claude AI usage Limits By [Toni Dos Santos](https://wecallshotgun.com/toni-dos-santos) • Oct 28, 2026 • 8 min read • **Summary:** Why document work burns through usage, and how to build a private converter that gives Claude fewer tokens to read. Most of our clients who’ve switched to Claude in the past 6 months now get asked by employees to pay for extra usage almost daily. The IT department or HR need to manage frustration and over-budget spending on top of AI adoption. They upload heavy PDFs and PowerPoint decks. Then the work gets interrupted. After all the effort to get a team using its AI licences, that’s a frustrating place to end up. Buying more usage keeps things moving. My first question is what we’re asking Claude to read. A whole deck to rewrite the copy. A PDF full of page layouts when the job is to summarise the wording. The attachment can ask far more of Claude than the prompt suggests. That’s why I built Plainly. It turns documents into Markdown on my Mac, ready to check and upload. ## A short prompt can carry a heavy file Claude breaks text into tokens: words, parts of words and punctuation. Images use tokens too. The material it receives is the input. What it writes back is the output. Both matter to usage. “Summarise this report” is a short request. Claude still has to process the report. Your model and effort setting affect consumption too. There isn’t a fixed exchange rate between tokens and the percentage in Settings. Once you hit a limit, you can wait for a reset or, on eligible paid plans, buy more usage. [Anthropic explains the limits here.](https://support.claude.com/en/articles/11647753-how-do-usage-and-length-limits-work) With new models, token consumption is supposed to go down. But with recent announcements of pricing increase and hard limits, token frugality remains a central concern. And follow-ups carry earlier context. Asking for a shorter answer doesn’t remove the attached report. Caching can reduce the cost of reused material, though, so counting every follow-up as a fresh, full-price upload would overstate it. [Anthropic’s usage guidance covers both.](https://support.claude.com/en/articles/9797557-usage-limit-best-practices) Rolling Claude out to a team? Our [Claude training for enterprise teams](https://wecallshotgun.com/claude-training) starts from the work each role already does. ## Claude can see every page you send When Claude processes a PDF visually, it gets the extracted text and an image of each page. Useful for a chart. Extra input when all you need is the wording. [That’s how Anthropic describes PDF processing.](https://platform.claude.com/docs/en/build-with-claude/pdf-support) ![Diagram: a PDF report becomes extracted words (text tokens) plus an image of each page (visual tokens) before Claude reads the input together with your question and the conversation context.](https://substack-post-media.s3.amazonaws.com/public/images/d554c2d3-7068-4012-85e8-ef333ab50041_1470x900.png)A PowerPoint exported to PDF or rendered as slide images can create the same overhead. A deck read through text extraction is a lighter job. Markdown is text with simple markers for headings and lists. The file ends in `.md`. It keeps the words and their basic structure without sending a picture of every page. These worked examples make the difference easier to see. They’re illustrative assumptions, not measurements from client files. ![Bar chart of worked estimates: Markdown and Word at 4,100 tokens for a 5-page document, the same PDF read visually at 11,600 tokens, and a 30-slide PowerPoint exported to PDF at 60,000 tokens.](https://substack-post-media.s3.amazonaws.com/public/images/027cd55a-6b6e-450e-b9cb-2e00a2ffb699_1470x900.png)The assumptions behind the chart: - The document example assumes 4,100 text tokens. The deck assumes 15,000. - Both allow 1,500 visual tokens per rendered page or slide. - For the deck: 15,000 + (30 × 1,500) = 60,000 tokens. Actual image costs depend on the model and resolution. [Anthropic documents image processing here.](https://platform.claude.com/docs/en/build-with-claude/vision) In this example, removing the slide images cuts 45,000 input tokens. It doesn’t promise 4 times your subscription allowance. The file is only part of what Claude processes. A clean Word document is already a sensible choice for text work. Markdown earns its place when conversion removes visual material the task doesn’t need. ## The chart still matters For rewriting slide copy, the extracted text gives Claude something to work with. For a design review, it needs the slides. Sending the 6 slides under discussion keeps the rest of the deck out of that request. Conversion needs checking, too. Tables can lose their reading order. A chart’s labels can survive while the relationship between them disappears. I keep the original nearby and add back the page that carries information the text has lost. A smaller file with a broken table creates another job. ## Why I built a local converter Free online converters exist. Some upload your document to another provider before the result ever reaches Claude. For a public report, that may be acceptable. For a salary grid or client contract, I want to know who gets the file and what they keep. Plainly opens in a browser, but the conversion runs on my Mac. I drop in a document, check the extracted text and download the Markdown. It uses [Microsoft’s MarkItDown](https://github.com/microsoft/markitdown), a free, open-source conversion tool for PDF and Office files. I added the interface around it. The fuller version also reads scanned pages locally. That’s my reason for using it with company work: conversion doesn’t require another cloud provider. A free converter that processes files on your computer can meet the same requirement. Uploading the finished Markdown to Claude still sends its contents to Claude. The company’s rules about what belongs there still apply. ![Diagram of the Plainly workflow: a PDF or PPTX is converted locally into a Markdown file, the text is checked on your computer, and only the approved file is uploaded to Claude.](https://substack-post-media.s3.amazonaws.com/public/images/07b550e6-eabb-4c5b-bc42-dae43fa190db_1470x900.png) ## Build the basic version You can create the app from scratch in Claude Code. I share the prompt below. Our paid subscribers also get access to: - **The full prompt** and a detailed guide on Notion - **The compiled .dmg app for Mac** (download it and add it to your Applications folder) - **The full app folder as a zip** (download it to your computer and launch it locally) The starter version extracts text from ordinary PDFs and PowerPoint files. You’ll need Claude Code in the desktop app and permission to install software on your Mac. Building it uses your Claude allowance. Running the finished local conversion doesn’t require an AI call. - Create an empty folder called `Plainly`. - In Claude Desktop’s Code tab, choose **Local** and select that folder. - Choose **Manual** permissions to review changes as Claude makes them. [Anthropic documents those controls here.](https://code.claude.com/docs/en/desktop) Paste this: ``` Build a small local document converter called Plainly using Microsoft MarkItDown: https://github.com/microsoft/markitdown I'm a non-technical Mac user. Explain the steps and permission requests in ... ``` **Want to go deeper?** At [We Call Shotgun](https://wecallshotgun.com/enterprise), we help startups and scale-ups integrate AI into their product and GTM processes. Explore our [AI adoption programs](https://wecallshotgun.com/enterprise) for hands-on workshops and deployment support. ## Keep Reading [Ai Aug 27, 2026 ChatGPT finally makes AI images that are useful for our brand OpenAI's image 2.0 model is the best for marketing. This week it added transparent backgrounds and stickers. 4 months in, here's what works 14 min read Read →](https://wecallshotgun.com/blog/chatgpt-finally-makes-ai-images-that) [Ai May 8, 2026 Your ChatGPT "Mega Prompts" now make your outputs worse. The 6-part brief replaces your 400-word prompt in ChatGPT 5.5. Stop training AI as an intern. Start briefing it like an agency. 8 min read Read →](https://wecallshotgun.com/blog/your-chatgpt-mega-prompts-now-make) [Ai May 2, 2026 Stop writing AI slop. Start preventing it. LLMs all sound the same because they’re trained on the average. AI slop writing needs forced constraints and writing frameworks to remove 80% of the slop. The remaining 20% writing is your key added value. 11 min read Read →](https://wecallshotgun.com/blog/ai-writes-like-ai-slop) ## Related AI Training Solutions Programs tailored to your role, industry, or function. [Financial Services](https://wecallshotgun.com/ai-training-financial-services) [Healthcare](https://wecallshotgun.com/ai-training-healthcare) [Legal Teams](https://wecallshotgun.com/ai-training-legal) [Operations](https://wecallshotgun.com/ai-training-operations) [Product Teams](https://wecallshotgun.com/ai-training-product) [Marketing](https://wecallshotgun.com/ai-training-marketing) [Sales Teams](https://wecallshotgun.com/ai-training-sales) [Customer Support](https://wecallshotgun.com/ai-training-customer-support) [HR Teams](https://wecallshotgun.com/ai-training-hr) [Data & Analytics](https://wecallshotgun.com/ai-training-data-analytics) [Executives](https://wecallshotgun.com/ai-training-executives) [C-Level](https://wecallshotgun.com/ai-training-c-level) [UK Hub](https://wecallshotgun.com/ai-training-uk) ## Need help with AI Adoption? We take the passenger seat and lead your teams to adopt AI for real work. [Book a Call](https://wecallshotgun.com/#contact)