Worksheet: compare AI training on user data language across two ToS versions

Photo by Monstera Production from Pexels

Every time a SaaS company updates its terms of service, the AI training clauses often shift, sometimes subtly, sometimes dramatically. You might notice new language about "model improvement," broader data retention, or looser opt-out rules. But spotting these changes manually is tedious and error-prone. This worksheet gives you a repeatable, step-by-step method to compare how two versions of a ToS differ on AI training, so you can decide whether the new terms protect your data better or worse.

TL;DR

• AI training clauses hide in privacy policies and ToS under different headings (look for "model improvement," "analytics," "product development").
• Side-by-side comparison reveals scope shifts: does the new version train on all user data or only anonymized input?
• Red flags include removal of opt-out language, expansion of "permitted purposes," or looser deletion timelines.
• Use this worksheet to document changes in plain language, then escalate to your legal or vendor-management team.
• Terms Doctor's A-F grading and change tracking can help you spot AI clauses faster and flag updates automatically.

---

Why AI training clauses matter

AI systems learn from data. The question is: whose data, and with what permission? Over the past two years, companies have rapidly added or expanded AI training clauses into their terms of service. Some are transparent ("we use your content to improve our models, and you can opt out"). Others are vague ("we may use your data for research and product development"). When a company updates its ToS, these clauses often change, sometimes expanding rights, sometimes adding limits.

If you rely on a tool for work, a small shift in AI training language can change your risk exposure. For example:

• A design tool might shift from "we don't train on your private projects" to "we train on all projects unless you toggle an option per file."
• A productivity app might expand its definition of "analytics" to include full conversation logs sent to external AI vendors.
• A freelance platform might remove a previous promise to delete contractor portfolios after account closure.

The problem: most people never notice these changes, because ToS updates are buried in notification emails and the documents themselves are dense. This worksheet helps you catch and document those shifts.

---

How to use this worksheet

This worksheet is designed for solo reviewers, small teams, and anyone managing vendor compliance. You'll need:

1. Two versions of the ToS (current and previous, or two different vendors)
2. 30–45 minutes of focused time
3. A text editor or spreadsheet to record findings
4. This worksheet printed or open in a second window

Work through each section in order. You don't need legal training, plain-language comparison is the goal. If you get stuck, ask your legal team or a compliance officer to review your findings.

---

Step 1: Locate and extract AI training language

AI training clauses don't always live in an obvious section. Search for these keywords in both documents:

• "AI," "machine learning," "model"
• "Train," "training," "learn," "learning"
• "Analytics," "product development," "research"
• "Improvement," "service improvement"
• "Data processing," "permitted purposes"
• "Aggregate," "anonymize," "pseudonym"
• "Vendor," "third-party AI," "service provider"

Use your browser's Find function (Ctrl+F or Cmd+F) to jump to each mention. Copy the full sentence and surrounding paragraph into a shared document. Don't paraphrase yet, keep the original wording exact.

Why this matters: AI clauses are scattered across privacy policies, terms of service, data processing addenda, and cookie notices. If you only skim the Privacy Policy, you'll miss the AI training clause buried in the "Product Development" section of the ToS.

---

Step 2: Create a side-by-side comparison table

Create a simple three-column table:

Aspect	Version 1 (Old/Vendor A)	Version 2 (New/Vendor B)
Data types used for training	Specify exactly what data: user content, logs, metadata, analytics?
Scope of training	Does it apply to all users or certain tiers? Opt-in or opt-out?
Retention timeline	How long is data kept before deletion?
Third-party sharing	Does the company share data with external AI vendors?
Opt-out mechanism	Can users disable training? Per-account or per-asset?
Transparency	Does the company disclose how training works?
Deletion rights	Can users request deletion of their training data?

Copy exact quotes from both versions into each cell. If a clause doesn't exist in one version, mark it as "Not mentioned" or "Removed."

""We're still seeing a majority of teams literally opening two PDF files on their desktop or looking at two pieces of paper side by side, and doing quite complex things with them."."

>, AI Document Comparison: How to Compare Documents at Scale With AI (5 Real

This quote captures the pain point: manual comparison is slow and error-prone. A structured table forces precision.

---

Step 3: Spot changes in scope and language

Now compare each row. Look for these red flags:

Scope expansion:

• Old: "We use anonymized analytics to improve search."
• New: "We use all user content, including files and conversations, to train AI models."
• Flag: Data definition broadened from analytics-only to full content.

Removal of opt-out language:

• Old: "Users can opt out of AI training via Settings > Privacy."
• New: "AI training is enabled by default. Contact support to disable." (or no mention of opt-out)
• Flag: Switching from opt-out to opt-in is a major shift in user control.

Vaguer permitted purposes:

• Old: "We train on user feedback to improve our sorting algorithm."
• New: "We may use your data for research, development, and other business purposes."
• Flag: "Other business purposes" is open-ended and harder to enforce.

Looser retention timelines:

• Old: "Training data is deleted 30 days after account closure."
• New: "Training data may be retained for up to 2 years for model refinement."
• Flag: Longer retention means your data is at risk longer.

New third-party sharing:

• Old: "We train models internally and do not share user data with AI vendors."
• New: "We may share training data with selected third-party AI vendors for model development."
• Flag: Your data is now leaving the company's control.

---

Step 4: Document your findings

Write a plain-language summary of what changed. You don't need legal jargon, focus on practical impact:

Example summary:

"Version 2 now trains AI models on all user documents (not just analytics). Previously, only anonymized metadata was used. Version 2 also removes the opt-out button and requires contacting support. Third-party AI vendors can now access our training data. Retention was extended from 30 days to 2 years post-deletion."

Key takeaway: Changes in AI training scope, removal of opt-out levers, and new third-party sharing are the biggest risks to your privacy.

---

Practical comparison checklist

---

How Terms Doctor can help

Terms Doctor runs 101 consumer-protection checks on every website you visit, including dedicated checks for AI training clauses. When you visit a SaaS tool's pricing page or sign-up form, the extension automatically pulls the ToS and flags AI training language, third-party data sharing, forced arbitration, and auto-renewal risks, all ranked A-F.

If a company updates its ToS, Terms Doctor's change-tracking feature alerts you to what shifted. You can then use this worksheet to dig deeper into the specific language changes and decide your next move.

Not legal advice: These comparisons and checklists are informational tools. If you're signing a contract worth significant money or exposure, have a lawyer review the final terms.

---

FAQ

---

Next steps

1. Download both ToS versions and save them with clear dates (e.g., vendor_tos_2024-09-01.pdf, vendor_tos_2025-01-15.pdf).
2. Work through the comparison using this worksheet. Expect 30–45 minutes for a thorough review.
3. Document your findings in a shared spreadsheet or document that your team can reference.
4. Escalate risks to your legal or compliance team if you spot red flags (scope expansion, removed opt-out, new third-party sharing).
5. Make a decision: Accept the new terms, request amendments, or evaluate alternative tools.
6. Use Terms Doctor to monitor for future updates and get automatic A-F grades on every vendor's terms.

Privacy protection starts with reading the fine print, and this worksheet makes that work faster and more repeatable for your whole team.

Additional Resources

• How to Compare Documents at Scale With AI (5 Real- ... - AI document comparison replaces slow, manual side-by-side review with scalable, semantic analysis, enabling teams to identify gaps, ...
• Insight | Comparing AI Language Models for Business - This article compares ChatGPT, Copilot, Gemini, and Claude, highlighting capabilities, privacy, and ethics to help organizations choose the right AI model.
• AI Document Comparison Tool: Find Gaps, Prove Every ... - AI-powered side-by-side comparison of documents, products, or any content. Auto-discovers dimensions, scores and ranks items, web research mode ...