Most small and mid-sized businesses are already using AI, whether officially or through staff pasting text into free tools. That makes AI data privacy a present-day concern rather than a future one. You do not need a large compliance team to manage it well, but you do need a clear list of what to check. Use the checklist below as a starting point, and involve a qualified legal advisor for decisions specific to your jurisdiction and sector.
1. Map what data goes where
You cannot protect data you have not identified. For every AI tool or feature in use, record:
- The tool or provider name and who in the company uses it.
- What data is sent: customer names, emails, financial details, health information, internal documents, source code.
- Where it is processed and stored, including the country or region.
- Whether the data leaves your systems at all, or stays in your own environment.
Include unofficial use. A short, non-judgmental survey of staff often reveals tools nobody knew about.
2. Classify your data
Not everything needs the same level of care. A simple classification makes later decisions easier:
| Class | Examples | Default rule for AI tools |
|---|---|---|
| Public | Published website content, marketing copy | Any approved tool |
| Internal | Process documents, meeting notes without personal data | Approved business accounts only |
| Confidential | Customer records, contracts, financial data | Approved tools with suitable contractual terms |
| Restricted | Health data, ID documents, payment card data, passwords | Do not send to AI tools without a specific, reviewed case |
3. Check vendor terms carefully
Consumer and business versions of the same AI product often have very different data terms. For each provider, confirm:
- Whether your inputs and outputs are used to train their models, and how to opt out.
- How long data is retained, and whether you can delete it.
- Where data is processed, and whether that meets your obligations.
- Whether a data processing agreement is available and signed.
- Which subprocessors they rely on.
- What security certifications or audit reports they can share.
4. Minimize what you send
The safest data is data that never leaves your systems. Practical techniques:
- Send only the fields a task needs, not entire records.
- Replace names and identifiers with placeholders before sending, and restore them afterward where needed.
- Summarize or filter documents locally before passing them to a model.
- For highly sensitive workflows, consider models hosted in your own environment.
5. Control access inside AI systems
AI assistants that search company documents must respect existing permissions. A person who cannot open a file should not be able to get its contents through a chat question. Check that:
- Users sign in with individual accounts, not a shared login.
- Retrieval is filtered by each user’s permissions.
- API keys are stored securely, rotated and never placed in front-end code.
- Admin access is limited and reviewed regularly.
6. Mind your automations
Workflow tools that connect AI to your CRM, inbox and accounting system move data between many services. Each connection is a place where data can be copied, logged or retained. Review what each workflow step stores in its execution history and how long. Some platforms, such as n8n, can be self-hosted to keep data in your own infrastructure; we compare the options in n8n vs Make vs Zapier.
7. Set retention and deletion rules
- Decide how long prompts, outputs, transcripts and logs are kept.
- Make sure deletion requests from customers cover AI systems and their logs too.
- Avoid keeping full conversation histories by default if you do not need them.
8. Be transparent with customers
- Update your privacy notice to describe AI processing where relevant.
- Tell people when they are interacting with an AI system.
- Check whether any processing requires consent or a specific legal basis under the laws that apply to you.
- Be careful with automated decisions that significantly affect individuals, such as credit or hiring outcomes; many jurisdictions have specific rules here.
9. Write a one-page staff policy
Long policies go unread. A one-page guide should cover which tools are approved, what data may never be pasted into them, how to request a new tool and who to ask when unsure. Pair it with a short training session and real examples from your business.
10. Review regularly
AI tools change quickly: providers update terms, add features and launch new products. Put a short review on the calendar, for example twice a year, to recheck your tool inventory, vendor terms and staff policy. Also review whenever you add a new AI feature that touches customer data.
11. Plan for incidents
Decide in advance what happens if sensitive data is sent to the wrong tool or exposed through an AI feature: who is notified, how the data is deleted, how affected people are informed if required and how the cause is fixed.
AI data privacy self-assessment
- We know every AI tool in use and what data it receives.
- Our data is classified, and staff know the rules for each class.
- Vendor terms on training, retention and location are checked and documented.
- AI assistants respect user permissions.
- Retention periods are set for prompts, outputs and logs.
- Customers are told about AI use where relevant.
- Staff have a short, clear policy.
- There is an incident plan.
Privacy-first AI with SiteWright Technologies
When we build AI features and automations, data handling is part of the design: minimal data, the right vendor settings, permission-aware retrieval and sensible logs. Learn more about our AI and automation services or contact us to review your current setup.


