What can I enter into an LLM when doing market research?
Honestly? Most non-PII data is fine, but only if you find and toggle one particular privacy setting.
Privacy. There’s so much out there about what to enter or not enter into an LLM. I’ve been guilty of giving the advice, “Don’t specify the company name when drafting the questionnaire. Say, ‘global CPG brand.’” But, let’s be real, with increasingly fast turnaround times expected of you, scrubbing every prompt is…less than realistic.
So, here are some quick tips on data privacy when it comes to using your LLM for market research.
Check your settings. Set it to “do not share my conversations to improve the model.”
Never ask the model to anonymize data for you. Instead, do it yourself by removing the name, email address, phone number, IP address at a minimum. Remove anything that counts as Personally Identifiable Information, or that, when viewed, could reasonably be used to identify the individual.
The real tip that makes the biggest difference? Use tools that let you opt out of training the AI models behind the tools and don’t enter PII into the tools. For Claude, ChatGPT, Copilot, and Gemini, that means going into your settings and specifically looking for the setting that lets you toggle off “train the model.” It’s annoying, yes, but any app you’re using to do market research is worth the annoyance of looking for the location where you opt out of sharing data to train AI models.
When using other tools, read the Terms and Conditions and Privacy Policies. Here’s what to look for:
Is it specific about whether or not data entered about projects is used to train the model? If it’s vague, walk away (if you can), or be careful what you’re entering into that tool about your customer.
Is there a separate privacy policy for you, the researcher, and the person responding to the questionnaires and interviews? Check each of them for the specific clause, “We do not use your prompts/your data to train our models.”
For those tools that use Claude or ChatGPT or some other LLM behind the scenes, a clause like, “We do not use your prompts/your data to train Anthropic or OpenAI models,” is what you want to see.
For those of you working in companies that have Enterprise agreements or Education agreements, your prompts and data aren’t being shared to train models by default. It’s part of the agreement. But it’s always worth a double-check for any tool, no matter the contract or subscription level.
A quick setting check and a little intentionality can go a long way towards keeping your work out of someone else’s AI answer.
If you found this helpful, share with a colleague and encourage them to sign up for the weekly tips! And tell me what questions you have about using AI for market research. I’ll do my best to address them in upcoming newsletters.
Learn more through my new classes starting in September! Sign up for individual classes or for a cohort at mrxplorer.com.