AI in Market Research: Analyzing Open-Ends Faster and Better

Open-ended survey responses – known as open-ends – are the most valuable yet most difficult-to-analyze part of any survey. They contain the unfiltered voice of respondents: reasons, nuances, ideas, and criticism that would never surface in closed questions.

But the reality is: in many market research institutes, open-ends are either manually coded – with enormous time investment and quality fluctuations – or simply ignored because analysis would be too expensive or too slow. AI fundamentally changes this equation.

A typical survey with 2,000 participants and three open questions generates 6,000 free-text responses. For larger studies – brand tracking, customer satisfaction, employee surveys – that quickly reaches 20,000 to 100,000 texts.

The traditional evaluation process looks like this:

1. A coding team reads all responses
2. A code frame is created (20-50 categories)
3. Each response is assigned to one or more categories
4. Results are aggregated and incorporated into the report

This sounds structured – but in practice comes with significant problems:

• Time: Manually coding 6,000 texts takes 3-5 working days – at a cost of 3,000-8,000 euros
• Inter-coder reliability: Different coders assign the same text differently. Studies show agreement rates of only 70-80%
• Code frame bias: The code frame is often created based on the first 100-200 responses. Topics that emerge later are squeezed into residual categories
• Scaling problem: With growing sample sizes, costs and time increase linearly
• Time pressure: In agile research cycles, manual coding is the bottleneck

Modern AI systems for text analysis combine multiple NLP methods to evaluate open-ends faster, more consistently, and more deeply than manual coding:

1. Automatic Topic Extraction

Instead of a predefined code frame, the AI recognizes autonomously which topics appear in the responses. This is particularly valuable because:

• Surprising topics don't disappear into residual categories
• The topic structure reflects the data, not the researcher's assumptions
• Emerging trends are detected early

2. Sentiment Analysis per Topic

The AI recognizes not only what is being discussed but also how – positively, negatively, or neutrally. This transforms open-ends from a qualitative to a quantitative data source. Illustrative result (fictional): "78% of mentions about customer service are positive, but 65% about wait times are negative."

3. Automatic Coding

When a code frame already exists, AI can take over the assignment – with uniform criteria. This is particularly relevant for tracking studies where comparability across waves is critical.

4. Summaries and Quotes

AI can summarize the key statements per topic and identify representative quotes – material that flows directly into presentations and reports.

The question market researchers rightly ask: Is AI evaluation good enough? The answer depends on the task:

For topic identification:

• AI often finds additional topics – especially niche topics and emerging trends
• Assignment quality can be checked on a sample and compared with the inter-coder reliability of manual coding
• AI works consistently: all texts are assigned according to the same criteria

For sentiment detection:

• AI detects irony and sarcasm better than early models, but doesn't yet match human intuition in borderline cases
• For aggregated evaluation (percentage positive/negative per topic), the assignment is usually good enough if you check samples
• For aspect-based sentiment, AI surpasses manual evaluation because it consistently analyzes every single text

Not every market research study benefits equally from AI-powered text analysis. Here are the areas with the highest impact:

Brand Tracking

Open questions like "What do you associate with brand X?" generate massive text volumes across multiple waves. AI enables consistent coding across waves and detects shifts in brand perception that get lost in the manual process.

Customer Satisfaction Studies

NPS comments, open satisfaction questions, improvement suggestions – this is where the greatest treasure of unstructured feedback lies. AI shows automatically, at the aspect level, which topics are associated with satisfaction and dissatisfaction.

Product Tests & Concept Tests

Verbal reactions to product concepts contain nuances that scales don't capture. AI recognizes emotional reactions, concerns, and spontaneous associations – and quantifies them.

Ad Testing

Open questions about advertising effectiveness generate particularly diverse responses. AI can analyze recall elements, emotional reactions, and message comprehension simultaneously.

Employee Surveys

Internal surveys often contain the most honest and detailed free-text responses. AI enables systematic evaluation; handling personal information carefully is critical, especially with employee feedback.

AI text analysis doesn't replace the entire research process – it integrates into existing workflows:

1. Questionnaire design: As before, with open questions at strategically relevant points
2. Fieldwork & data collection: Unchanged (online panel, CATI, CAWI, etc.)
3. Data preparation: Export of free-text data (e.g., CSV or Excel)
4. AI analysis: Upload to the text analysis platform, automatic topic and sentiment analysis
5. Review & refinement: Researchers review AI results, rename topics, add context
6. Integration: Results flow back as structured variables into the dataset or directly into the report

The decisive advantage: Step 4 is much faster than manual coding. The time saved can be invested in deeper interpretation and better consulting – the part where human expertise is irreplaceable.

deepsight was developed from the ground up with market research requirements. The deepsight Cloud platform offers:

• Analyze responses in other languages with the integrated translation (AI-powered, or DeepL or Google if you prefer)
• Runs on Microsoft Azure in the EU; AI features via Azure OpenAI, also in the EU; DPA under Art. 28 GDPR
• Our own anonymization without a language model: personal information is detected and masked before the analyses run. If translation is switched on, texts are translated before masking. The anonymization does not detect every piece of information – check results before sharing them.
• Topic extraction and aspect-based sentiment analysis in one workflow
• Export as Excel or CSV (original data with code columns), plus reports with PowerPoint export
• Special system integrations are implemented as custom projects
• Your own codebook – including parent and child categories – for consistent coding across waves

Learn more on our page for market research institutes – with concrete use cases.

Open-ends have long been the stepchild of market research – too laborious to evaluate, too expensive for large sample sizes, too inconsistent for tracking. AI solves all three problems: It analyzes faster, cheaper, and more consistently (the same criteria across waves).

For market research institutes, this means: open-ends evolve from a cost factor to a differentiator. Those who can offer their clients fast, deep, and reliable open-end evaluations have a genuine competitive advantage.

Get to know deepsight in a personal demo. Afterwards, you receive test access. We agree the scope of the test with you, based on your project.