WORKFLOW UPGRADES

🔌 Custom GPTs for faster surveys

I’m a big fan of custom GPTs, if you haven’t noticed. I think of them like the simple, entry level versions of the “AI agents” everyone’s been talking about (though it’s not quite the same - see the News section for a customer research agent example).

Just like my simple Bias Hunter custom GPT (and tutorial) I shared in the August edition, I’ve been using several “off the shelf” custom GPTs to streamline tedious parts of my work, like for surveys:

• Survey Crafter GPT
  It’s a pretty good tool for creating well-phrased, targeted questions, especially if you’re clear about your research goals up front. Helpful for the moments when good questions just don’t come easily, or - in the case of some of my clients - when no one in the team is a survey-writing expert.
  
  For more nuanced research, you may want to tweak the questions it generates slightly to fit your audience tone or context better. Overall, it’s a helpful time-saver for those looking to jumpstart surveys without starting from scratch.

• Survey Analyzer GPT 
  This one does a solid job at quickly summarizing large volumes of survey data, identifying common themes and trends efficiently. As with most things AI right now, this custom GPT can sometimes miss subtle insights or context-specific nuances that a human researcher might catch.
  
  It’s best used as an initial pass to help you quickly grasp big themes, followed by a closer look for deeper analysis if needed.

Here’s a sample of the kind of questions Survey Crafter GPT comes up with, based on very little input from me other than a 1-liner about my audience and objective. 👇

SURPRISE! 🐣

There’s one more thing I didn’t mention above - I’m still working on it, but I want you to have this while I build the collection.

I’m sick of wondering which tools out there do which tasks for customer research, and which tools I’ve missed. So I started collecting them all in one place, tagged with the tasks they do with AI, where in the world they store your data, and more.

Make your own copy or keep coming back to mine (I’ll update it whenever I find a new tool).

Get the 110+ AI Tools for Customer Research database here!

DATA PRIVACY

🙅 Data privacy considerations: How I anonymize data and remove PII before using AI

All the AI gurus out there hype up AI…without talking about all that data we’re sending into the void. But we have to address that.

Protecting personal customer data is paramount, especially when using AI tools like ChatGPT, Claude, and other non-corporate versions.

Failing to properly anonymize data can lead to breaches of privacy regulations. Here’s a guide on how I anonymize data effectively for myself and clients.

See my Anonymization Guide here.

AI FUNDAMENTALS

🖼️ 8 Ways AI combats bias in qualitative data analysis.

Even though they may also introduce some bias, most AI research tools are actually built to kill bias at every stage. Here’s the upside of using them in your customer research process, and how they can make insights more objective:

1 - Automatic privacy protection: Cuts Demographic & Observer Bias

Tools that hide personal details automatically keep responses and analysis neutral—no judgments based on identity. Check whether your existing user research tools are doing this, it’s a big plus.

2 - Instant feedback processing: Beats Recall Bias

Tools that analyze in real-time help you and the AI avoid memory reliance, keeping insights more accurate by analyzing and synthesizing them now.

3 - Sorting emotions fairly: Challenges Confirmation Bias

I’m not yet a fan of many sentiment functions in customer research tools, but sentiment analysis in AI is one way to “objectively” sort emotions, removing any influence from researcher expectations.

4 - Theme clustering and automated tagging: Avoids Selection Bias

Tools using AI to group related feedback can help ensure balanced representation across themes more than when humans do the thematic grouping alone. AI may at the very least be looking at the data through a different lens than we do, and offer a slightly different perspective in the resulting analysis and insights.

5 - Demographic cross-checking: Prevents Sampling Bias

AI tools that compare responses across demographics can potentially generate insights that better reflect everyone, and not just one group. Of course, this is a double-edged sword, and we know that bias is built in to the data sets used to train AI models.

6 - Balanced data representation: Reduces Recency and Frequency Bias

Many tools are starting to make sure they balance feedback from different sources, and show them on equal levels or with equal weighting, preventing recent or frequent responses from overshadowing others. Humans tend to remember what we just heard and emphasize that. AI can help us counter that tendency.

7 - Tracking feedback by segment: Fights Overgeneralization

Keeping track of group-specific feedback with AI can help us avoid making assumptions that certain anecdotes we’ve heard are a shared experience by many and not just a few customers (overgeneralizing about a limited or unique occurrence).

AI STUDIES

🧑‍🔬 Does ChatGPT give different answers to your colleagues?

The short answer? Sometimes!

Study: ChatGPT Doesn’t Trust Chargers Fans: Guardrail Sensitivity in Context

Researchers discovered that ChatGPT’s “guardrails”—rules to avoid harmful responses—can cause the AI to give different answers based on clues about a user’s identity, like their age, work role or interests. For instance, some users might get more cautious responses or the LLM may refuse to share information with them based on perceived traits.

What matters here:

• Inconsistent responses: AI may interpret the same question differently depending on who it thinks is asking!

• Hidden information: If AI thinks you’ll respond negatively to certain information, it seems to be able to intentionally withhold that information from you.

• Implications for teams: When team members use ChatGPT for research, varying answers could create gaps or inconsistencies in insights the LLM delivers, depending on who’s working with AI and what it thinks it knows about them.

So…

What if the AI thinks [because my colleague is male, or from Europe, or employed, and I’m not] that my colleague and I will be interested in completely different insights from our customer data? AI might even decide that I would be angry with it about a given insight, and decide not to share it with me (Social Desirability Bias at play?).

NEWS TO KNOW

⏩ AI Agent Updates and Why It Matters for Research

New updates this month: Claude is handlings tasks with complex instructions better than ever, plus can use a computer (things like moving a cursor or clicking a button - without you involved).

ChatGPT also now has memory, so it can “remember” what you discussed earlier.

Why this matters: With these new memory upgrades, AI agents like Claude and ChatGPT can handle ongoing, linked tasks in research, allowing researchers to set up “follow-up” workflows without starting from scratch each time.


Here’s a clear example of how this could work for us:

Let’s say you’re running a 6-month study to track customer responses toward a new product.

1. Initial survey: The AI agent guides each participant through a survey about their first impressions. It takes note of key first comments, concerns, or any frustrations they mention.

2. Monthly follow-Up: Each month, the agent sends a new survey, but it remembers what each participant mentioned before. If a participant previously noted a specific issue (e.g., “slow load time”), the agent can ask if this has improved or if it’s still a concern.

3. Personalized questions based on history: Rather than re-asking the same questions, the agent tailors each follow-up to their past responses, diving deeper into new areas or checking in on previous feedback. For example, if a participant loved a feature last month, the agent might ask if it’s still working well or if they’ve noticed new ways it could be improved.

4. Streamlined insights: With memory, the agent can summarize how each participant’s feedback has evolved over time. This helps identify trends, like improved satisfaction as updates are released or recurring issues that need attention.

These AI agents aren’t just taking notes—they’re capable of tracking and measuring real change over time.

Imagine knowing exactly how customer responses shift over time with no extra time spent by you, because you never have to start follow-up feedback requests from scratch ever again. It’s benchmarking with qualitative input on autopilot.

Something like this could be a truly dynamic conversation with customers that evolves with each response. I’m looking forward to that.

Plus, a few interesting tools I’m playing with -

• BrowserCopilot AI - They say AI performs poorly because it doesn’t understand your context - but this one does. It jumps in when you need it across any webpage, picking up on your work's context to help you draft, summarize, and generate content seamlessly without switching tools.
  

• Sendsheets - Sendsheets lets you send customized replies directly from Google Sheets, so if customers leave feedback or specific requests, you can respond to each person individually, referencing exactly what they asked about. I’m imagining a world (with tools like this) where we never leave a customer hanging after getting feedback again.
  
  Per usual, none of these are sponsored, I’m only hoping that these tools will help solve your problems, faster!
  

Thanks for being here, and best of luck in the month ahead.

See you next time!
Caitlin