Solutions · Branding Understand which qualities people associate with your brand.
Organize open-ended associations by topic and compare them by target group, brand or survey wave – every assignment can be traced to the original wording.
What comes to mind when you think of our brand? Many answers, even more ways of putting it.
Open-ended associations show how people talk about your brand. That is exactly what makes them hard to count.
The challenge: Grouping keywords by hand
Reading every mention, assigning it to an image dimension, counting – and doing it all again in the next wave.
With deepsight: Topics, not keywords
Topic analysis groups different wordings into topics, using its own suggestions or your codebook with image dimensions.
Go to topic analysisThe challenge: Meant positively or critically?
“Expensive” can stand for premium quality or for rejection. If you only count mentions, you can’t see the difference.
With deepsight: Tone for each topic
“Pricey, but good” contains both. Sentiment analysis assesses the tone for each topic, not just for the response as a whole.
Go to sentiment analysisThe challenge: Key messages: getting through or not?
Whether a message is mentioned unprompted is hard to count cleanly in free-form keywords.
With deepsight: Messages as categories
Your key messages become categories in your codebook. You then see how often they are mentioned unprompted – by target group and across waves.
Go to group comparison
From association to topic
Open question: What comes to mind when you think of the Nordhof brand?
“It’s from around here, tastes like real milk.”
“Good quality, but at the supermarket it’s always one of the most expensive products.”
“The green packaging with the farm on it.”
“A farm in Lower Saxony, I think. A bit pricier, but organic.”
Select a topic to see the responses about it.
Fictitious brand and responses for illustration.
What you can compare associations by
Wave comparison is included in every paid plan – see the plan comparison.
What open-ended brand questions can tell you
- Which qualities do respondents associate with your brand unprompted?
- Which associations are meant positively, and which critically?
- How does your brand image differ from that of the competitor brands in your study?
- Are your key messages mentioned unprompted – and by whom?
- What would respondents trust your brand to offer in a new product category?
What you bring
Usually, the Excel or CSV export from your current survey tool is all you need – and you can keep using that tool.
From open mention to brand image
AI makes suggestions; you check them against the original and decide what to use.
- Check data
The check filters out empty and unintelligible responses, but also responses such as “no comment”. If “no comment” is itself a finding for your brand question, weigh up whether to use the check.
Data quality check - Build topics
Add your research question, target group and brand names to the study context. deepsight suggests topics or assigns mentions using your codebook.
Topic analysis - Assess sentiment
For each topic, you see whether a mention is meant positively, neutrally or negatively.
Sentiment analysis - Check against the original
You spot-check responses for each topic, merge similar topics and split topics that are too broad. Every change is saved as a version.
Topic refinement - Compare
Set topics side by side by brand or target group, with a base for each group.
Group comparison
What counts is what people mean, not the words they use.
“Premium” and “tastes like real milk” end up in the same topic: quality.
What you end up with
What matters with brand associations
Base and multiple mentions
Shares refer to the analyzed responses in a group; whatever the data quality check filters out doesn’t count toward the base. The base is shown for each group. Since one response can mention several topics, the shares often add up to more than 100%.
Waves as a time series
Add follow-up waves to a time series, using your codebook to keep the same categories. deepsight links topics with the same name and suggests matches for the rest, which you confirm. Significance indicators flag statistically noticeable changes compared with the previous wave and the first wave. A new AI run can produce different results; the aligned state of each wave stays fixed (see methodology).
Putting differences in context
If you wish, significance indicators show whether a target group differs noticeably from the rest of the sample in statistical terms; the test runs from 30 cases per group. The test assumes independent groups, which is not the case when the same people rate several brands. You can weight results by a weight column or to target distributions (rim weighting); weights do not automatically make a sample representative. A comparison alone does not prove causes such as a campaign.
Several brands in one study
Responses about several brands usually sit in one column per brand. There is no dedicated feature for this format: analyze these columns with the same categories, for example using your codebook. We will work out with you how to prepare the data for a comparison by brand.
Personal information in responses
Our own anonymization works without a language model and masks names in responses, such as those of store staff, before the analyses run. If translation is switched on, texts are translated before masking. The anonymization does not detect every detail – check results before sharing them.
Questions about analyzing brand studies
A word cloud counts words. Topic analysis groups wordings that address the same topic – even when they sound completely different.
Yes, including parent and child categories. You upload it as an Excel file (.xlsx) or CSV. deepsight assigns mentions strictly to your categories or also suggests additional topics. You do not need to train a model for this.
Set up the key messages as codebook categories and ask an unaided question. Your category definitions determine what counts as a mention – check the assignments against the original wording.
On request, deepsight checks the responses in the residual “other” category against your categories again. Select at least five of them and it also suggests a new category. Nothing is applied until you have reviewed it.
Yes, for example with responses from the French- and Italian-speaking parts of Switzerland. Integrated translation (AI-powered, or with DeepL or Google if you prefer) supports the analysis; the original remains available.
Related pages
Analyze your brand study
We will show you how open-ended associations are analyzed in a personal demo. Afterwards, you receive test access; we agree the scope with you – a completed brand wave, for example.