Client Data and AI Use

How we use AI, and what we never put into it.

Everyone uses these tools now. What matters is what happens to your data, and who checked the output.

Last updated 6 September 2026

We use AI tools in this work.

What matters is not whether a firm uses AI. Everyone does now. What matters is what happens to your data when they do, and who checked the output before it reached you.

1. Where AI is used

TaskAI usedWhat a person does
Crawling and technical scanningNo. Purpose-built crawling software Sets the crawl, checks the tool is not producing false errors before believing it
Reading large volumes of text, listings or reviewsYes Reads the output, checks claims against the source, discards what cannot be verified
Pattern-finding across order or traffic dataYes Frames the question, checks the arithmetic, decides what it means
Drafting reports and written materialYes, in part Every finding is checked against the live page before it ships. Every number is traced to its source
Judgments and recommendationsNo These are ours. A tool does not decide what you should do

2. What we do not put into an AI tool

  • Your customer records. Names, email addresses, order histories, anything identifying a person.
  • Your credentials. Logins, keys and tokens never go near a prompt.
  • Anything you have marked confidential, unless you tell us in writing that a specific tool is acceptable for it.
  • Financial records beyond the aggregate figures needed for the analysis you asked for.

Where a dataset must be analyzed and contains personal information, it is stripped of identifying fields first, or the analysis is done locally on our own machines. If neither is possible, we will tell you and let you decide.

3. Training

We use business-tier accounts whose terms exclude customer content from being used to train the provider's models. We do not opt in to training on client material, and we do not upload client work to consumer tiers where those protections do not apply.

4. Verification — the part that matters

AI produces confident text whether or not the underlying claim is true. That is not a defect we can prompt our way around; it is how the tools work. So the rule here is simple and it does not bend:

Nothing goes into a report that has not been observed on the live page, in the live data, or in a named published source.

If we could not verify it, it is either marked for verification or it is not in the document.

In our own research every finding is a specific, checkable claim about a live page. That discipline exists precisely because the tools make it so easy to produce something that reads well and is wrong.

5. Your material stays yours

Nothing you give us is reused for another client, fed into a shared model, or added to a library of examples. Where we describe an engagement publicly we do it anonymously, and only with detail that cannot identify you.

6. If you would rather we did not

Tell us and we will work without AI tools on your engagement. It will take longer and the quote will reflect that, quote will reflect that.

7. When AI writes something you will read

Where a substantial piece of written material was drafted with AI assistance, we say so in the document. We do not present machine-drafted work as though a person wrote every word, and we do not pretend a person did the reading when a tool did.

8. Corrections

If we get something wrong — a figure that does not hold, a finding that turns out to be stale — tell us and we will correct it in writing and explain how it got through.

Questions

Email info@marketinganalyticsconsultants.com