Does this make our AI use "GDPR compliant"?

No tool makes you compliant; compliance is about your whole processing. This workflow addresses the obligations that AI use touches most directly: minimisation, safeguards and accountability. Document it and you have an answer you can explain and defend.

What about documents under NDA rather than GDPR?

Same workflow, but check what the NDA protects. Placeholders replace personal data; contract terms, trade secrets or other confidential content stay readable. In the review step you can select extra text and have it replaced too. Also check whether your NDA defines third-party tooling explicitly.

How do we handle AI output in client deliverables?

After you put the real names back with the recovery key, treat the output like any draft you wrote yourself: reviewed by a person before it leaves the firm.

Can we standardise this across the team?

Yes. ShareSafe Free works per file and builds the habit per person; ShareSafe Workspace makes it a team workflow with shared vaults, roles and an activity log.

All guides

Using AI on client documents under GDPR: a 5-step workflow for consultants

The short answer. A defensible AI workflow for client documents has five steps: classify the document, replace the personal data, review the result, work with the AI on the protected version, and put the real names back with a recovery key that you hold. It supports GDPR's data minimisation principle, keeps personal data the task doesn't need away from AI vendors, and is much faster than redacting by hand. It is also simple enough to document once and follow daily. It does not make you compliant on its own; it gives you a way of working you can explain.

→ Run a client document through the workflow

Why consultants specifically

Advisory work is the worst case for casual AI use: high document volume, other people's confidential data, professional duty of confidentiality, and clients who will ask how you work with AI. Sooner or later an RFP or a contract negotiation will contain the question "describe how you use AI with our data". A written workflow turns that question from a threat into a differentiator.

The 5-step workflow

Step 1 — Classify first. Does the document contain personal data or client-confidential information? If genuinely not (public reports, your own templates), use AI directly. Everything else goes through steps 2 to 5. When in doubt, it is not in doubt.

Step 2 — Replace. Drop the document into ShareSafe.ai and choose reversible replacement. Names, companies, IBANs, addresses and identifiers become consistent placeholders. Replace data the task doesn't need. Leave amounts the model has to compare or calculate with. Processed in the EU, no account, the document is not kept as a file.

Step 3 — Review. Check the before and after view. Add what detection missed, especially indirect identifiers: roles in small companies, dates plus places, project names a sector insider would recognise. Placeholders replace personal data; contract terms, trade secrets or medical content stay readable. In the review step you can select extra text and have it replaced too. A short check that carries much of the safety.

Step 4 — Work with the AI. Summarise, compare, rewrite, analyse on the protected file. Consistent placeholders keep the document workable: [NAME_01] stays [NAME_01] throughout. Check the answer for your task.

Step 5 — Put the names back. The AI's output contains placeholders. Your recovery key (Identity Key) puts the real names back. The key is given to you and ShareSafe Free keeps no copy; don't paste it into the AI.

Writing it into your processing register

GDPR Article 30 asks you to record processing activities. This workflow is easy to record because it is fixed:

Purpose: document analysis and drafting support using AI assistants.
Safeguard: pseudonymisation (reversible replacement) before any transfer to AI vendors (Art. 4(5), Art. 32), with human review; recovery key held by the firm.
Minimisation: personal data the task doesn't need is replaced before transfer (Art. 5(1)(c)).
Retention: protected files and recovery keys per your existing client-file retention policy; ShareSafe Free does not keep the uploaded document as a file and keeps no copy of the key.

Four lines. Compare that to documenting ad hoc raw uploads, which you cannot.

Team habits that make it stick

Make the protected version the default: store it next to the original so colleagues reuse it instead of re-uploading the raw file.
One rule for juniors: no client document enters any AI tool without a processing receipt. The receipt shows the workflow ran.
Put the workflow in your proposal boilerplate. Clients increasingly ask; answering before they ask builds trust.
When document volume grows, the per-file routine becomes the bottleneck. That is the moment for ShareSafe Workspace: vaults, team access, an activity log and sharing via links, around the same principle.

FAQ

Does this make our AI use "GDPR compliant"?: No tool makes you compliant; compliance is about your whole processing. This workflow addresses the obligations that AI use touches most directly: minimisation, safeguards and accountability. Document it and you have an answer you can explain and defend.
What about documents under NDA rather than GDPR?: Same workflow, but check what the NDA protects. Placeholders replace personal data; contract terms, trade secrets or other confidential content stay readable. In the review step you can select extra text and have it replaced too. Also check whether your NDA defines third-party tooling explicitly.
How do we handle AI output in client deliverables?: After you put the real names back with the recovery key, treat the output like any draft you wrote yourself: reviewed by a person before it leaves the firm.
Can we standardise this across the team?: Yes. ShareSafe Free works per file and builds the habit per person; ShareSafe Workspace makes it a team workflow with shared vaults, roles and an activity log.

ShareSafe.ai keeps raw files in the EU. Minimal retention. You hold the key. Try it with your own document →