A lot of business knowledge sits in designed PDFs, locked Word files and messy shared docs. When the AI reads a designed PDF, columns, page numbers and footers can get mixed into the text. This lesson shows three simple tools that avoid that: plain text files, a log, and dated snapshots.
The mistake most people make
They upload a designed brochure and expect the AI to pick out small figures from tables and footnotes.
“A business owner uploaded an 18-page brochure with three-column layouts and asked for a summary of the warranty terms.”
“The AI mixed up the warranty periods for the domestic range and the export range.”
How it works: the pilot's logbook
An IndiGo pilot at Bengaluru airport doesn't read a glossy magazine before take-off. They open a logbook with the same columns every time: date, aircraft, route, fuel, defects, signature. It is short and leaves no room for doubt. Plain text with simple headings does the same job for the AI.
Three tools
1. Plain text with simple marks
Use # for a heading, - for a bullet and | to make a table. This style is called Markdown. You and the AI both read it easily.
2. A log of bad answers
A running file where you paste each bad answer and the fix you made. After a month it shows where your requests need more detail.
3. Dated snapshots
One file that records the state of things on a given day, such as 'stock-2026-09.md'. The AI gets exact starting figures and knows how old they are.
Habits
Turn my notes from this week into a short snapshot I can give the AI later. Here are the notes, messages and updates: """ [Paste this week's notes, chat messages or status emails here] """ Use these sections, in plain text with simple headings: # Snapshot · [today's date] ## 1. Key figures (as a table) ## 2. Work in progress and where it stands ## 3. What's stuck, and who has to decide ## 4. Things the AI should assume Use only what is in the notes. If a figure or owner is missing, write [CHECK]. No opening or closing chat.
Scattered notes become one dated page you can paste into any later request.
Working File Templates
Copy or save these plain text files to use directly in your workspace.
A plain-text project checkpoint capturing goals, progress, locked decisions, and next actions.
# Project Snapshot: ERP Billing Migration Date: 2026-09-20 · Lead: Ananya Sen ## 1. Goal Migrate 450 customer billing ledgers from legacy desktop software to the cloud portal by month-end. ## 2. Where Things Stand - Exported: All 450 customer master records extracted to CSV. - Reconciled: 380 ledgers verified with zero balance discrepancies. - Pending: 70 ledgers flagged for manual review due to unmapped credit notes. ## 3. Locked Decisions (Do Not Change) - Cut-off date is fixed for September 30. No invoices will be generated in the old system after 5:00 PM. - Rounding off: Amounts rounded to nearest rupee on final invoices. - Customer communication: Sent via email and SMS only after reconciliations are fully complete. ## 4. Current Blockers - Transporter billing mismatches on 18 accounts from the Surat warehouse. - Awaiting bank confirmation for ₹1.2 lakh unallocated deposit. ## 5. Next Immediate Action Review the 18 Surat transporter accounts with accounts team by Wednesday 2:00 PM.
A lightweight timestamped audit log to track what the AI generated and who signed off on it.
[2026-09-21 14:15:00 IST] RUN-0104 · DISPATCH SUMMARY - Task: Summarise weekend lorry dispatch manifests for depot managers. - Input File: manifests-sep-19-20.csv (SHA256: 7f83b2...a1) - Prompt Version: dispatch-v2.1 - AI Output: 4 depot summary tables generated. - Sign-off: Reviewed and confirmed by Suresh Nair (Depot Supervisor). Zero discrepancies found. [2026-09-22 09:30:00 IST] RUN-0105 · CUSTOMER CREDIT AUDIT - Task: Check overdue balances against 30-day payment limits. - Input File: receivables-aging-sep21.csv - Prompt Version: credit-check-v1.4 - Flagged: 6 accounts flagged for credit freeze. - Sign-off: Approved by Priya Mehta (Finance Lead). Action memo forwarded to collections.
Additional Prompts for This Lesson
Tailored prompts for specific workflows and departments.
Re-seeds the AI with your exact project state and decisions so you resume work without missing context.
You are my project collaborator. We are continuing an ongoing initiative. Here is our latest project snapshot: """ [Paste your session-snapshot.md content here] """ Here is the task for our current session: """ [Describe what you want to achieve in this session] """ Rules: 1. Ground your response entirely on the decisions and constraints recorded in the snapshot. 2. Do not re-open decisions listed under "Locked Decisions" unless explicitly requested. 3. If an open blocker prevents the task from completing, identify it first and suggest two actionable paths. 4. Keep the output formatted in plain text with clear headings.
Pasting the snapshot gives the AI your exact starting line, avoiding repetitive re-briefing.
Why does plain text usually work better than a designed PDF when you give the AI a document?
Show the answer
Plain text has clear headings and tables and nothing else. A designed PDF has columns, footers and layout that can get mixed into the text when it is read.