Most organisations measure culture with numerical ratings on engagement, leadership, or company culture. While these have a place in understanding sentiment, they don’t drive action, and it is likely missing a big part of the picture. Culture lives in everyday stories — how decisions are made, how conflicts are handled, how failures are treated. Doing this at a large scale has traditionally been too time consuming, but by automating analysis with AI analytics, and giving end-users an ability to ‘talk’ to the results, is opening new doors in how we listen to, and understand, employee experience.
We outline a simple three step framework on how to apply new tech and how this changes how we understand culture and experience.
1. Gathering: Ask for real, long stories
Start by treating your survey like mini‑interviews rather than a form. Focus on specific moments (“Think of the last time you had an idea to challenge the status quo…”) and ask people to describe what happened, what helped, and what got in the way. Encourage long responses.
Because large language models (LLMs) can analyse huge volumes of free text, you can confidently invite long‑form responses from thousands of employees. This turns a traditional survey into the equivalent of virtual focus groups with 1,000+ participants in an instant, and lets you understand people’s lived experience far more deeply than check‑boxes alone.
You can also, with strong governance and transparency, complement this with different survey formats (chat-bot front-end), or collecting data from passive data sources (e.g., work forums, patterns in collaboration tools).
In practice:
Frame questions around real decisions, trade‑offs, and interactions, push for long answers
These experience questions take time for participants, so limit questions to 4-5
Communicate clearly how the data will be used and protected
2. Analysing: From raw text to structure and insights
Traditionally, this is where you would need a team of analysts to trawl through the comments. Now, LLM tools can rapidly cluster them into topics (what are they talking about), sentiment (how happy or unhappy they are). What used to take teams of analysts’ weeks can now be done in hours by a single person.
Research is beginning to show that, with good prompt design and human oversight, AI‑based coding can match or exceed human consistency for many qualitative tasks, but validation is essential. In these types of free-text lead surveys you can find great, new insights you might not have expected, where people’s responses get onto a side-topic half-way through, be prepared to challenge your expectations of what topics to focus on. But be careful about which AI tools you use for this, not all are made equal, this needs a very high degree of accuracy.
In practice:
Find the right GenAI tool, more generic, basic models may hallucinate
Pre-define a set of behavioural themes you want to track, but be open to add more
Check for bias and misclassification, and refine prompts or models as needed
3. Sharing: Give leaders ownership by “talking” to the responses
Most surveys fail when insights don’t translate into action. Here, new tech can again help. Instead of a static report, provide leaders with a conversational interface; a chatbot connected to the anonymised research data. They can ask, in plain language, questions like “What do people suggest we should do differently to improve decision-making”.
This “talking to your participants” model changes how leaders interact with behavioural insights. It gives them ownership over what they explore, surfaces relevant examples on demand, and encourages more frequent, curiosity‑driven engagement. Also, it often uncovers topics you didn’t even intend to ask about. This creates accountability in actioning the outputs, as they will be more relevant, and more actionable as you can ask what people suggest should be improved.
In practice:
Build a secure chatbot on top of the analysed data, and test it thoroughly for any chance of hallucinations. Some survey platforms now provide this capability out-of-the-box
Summaries in itself can be vague, bring forward anonymised quotes; actual examples are more likely to drive empathy and actions
Curate example questions and prompts for HR and business leaders to use
What this means for the future of culture work
For organisations, this approach enables a new kind of listening: deep, story‑driven, regular, and scalable. Behavioural science provides the lens; AI provides the speed and reach; humans provide the judgment and ethics. Companies who embrace this; better gathering, smarter analysis, and interactive sharing, will be able to understand their culture in real time, hear diverse voices more clearly, and design interventions grounded in how work actually feels on the ground.
Remember, insights is just the beginning, the real impact is actioning change after this is done. Using this approach for quick pulse checks on how people are feeling throughout the year can be powerful. When you implement a change, apply this approach as a way to see how sentiment is improving with new initiatives.
