121surveys
Sign in
Menu

AI features

AI surveys: what the AI features of 121surveys do – and where their limits are

121surveys uses AI for seven tasks: categorising, summarising and counting open answers, reviewing the whole questionnaire, improving individual questions, translating surveys and drafting questionnaires based on SERVQUAL or HR models. The AI only runs with your account’s own API key at OpenAI, Anthropic (Claude), Mistral or Google Gemini, and every result is labelled as AI-generated.

Building and testing is free. To publish, you choose a paid plan.

The seven AI features at a glance

  • Categorise open answers

    The AI assigns every text answer to a question a theme and a sentiment: positive, neutral or negative. The run works through the answers in batches in the background, over all answers or only new ones, and shows its progress. The number of categorisations running at once is limited per person.

  • Summary of open answers

    At the click of a button you get a short summary with the key findings, trends and recommended actions. You choose whether all answers or a sample go in; before the run, a cost estimate shows the expected usage. The result is saved – as a new version or in place of the last one.

  • Count and normalise mentions

    The frequency analysis groups similar mentions under one label – for example “IHK”, “ihk” and “IHK Köln” become “IHK” –, counts each group and shows original examples. For questions with an “Other” field, the AI matches synonyms to the existing answer options.

  • AI text analysis of the questionnaire

    Before launch, the AI checks the whole questionnaire including its filter logic for methodological pitfalls, among them leading questions, double negatives, recall bias, order effects and switches in the form of address. Each finding has a severity – error, warning or note – and, where the problem lies in the question text, new wordings you apply with one click. Every run is kept, and you mark findings as done or ignored.

  • Optimise individual questions

    For a single question, the AI checks common questionnaire-design errors – from double-barrelled questions and vague terms such as “often” to overlapping or unbalanced answer options. You get back an improved wording in the survey’s language, the problems it found and concrete suggestions, which you accept or discard.

  • Translate a survey

    The AI translates all texts of a survey into a target language and only fills fields that are still empty; anything you translated by hand stays as it is. Every AI text counts as machine-translated until a person edits it in the translation check or marks it as reviewed. Until then, respondents see a notice on the page concerned.

  • SERVQUAL and HR generator

    From a description of your organisation, the AI drafts a questionnaire: SERVQUAL with its dimensions and one expectation and one perception item each, or an HR model – engagement, psychosocial workload based on COPSOQ, training evaluation after Kirkpatrick, onboarding, exit, diversity, wellbeing, leadership or stay interview. The questions go onto their own pages in the survey’s language.

Providers, keys and data flow

Which providers, and whose key?

Supported providers are OpenAI, Anthropic (Claude), Mistral and Google Gemini. A person with the Administrator role stores the API key for the whole account under Manage Workspace → AI Settings; 121surveys has the provider confirm it before saving, stores it encrypted and only ever shows it masked. For each task you pick provider and model from the list your key returns live – at Anthropic, the current Claude models. 121surveys does not use an AI key of its own for customer data.

Where does the data go?

Only when someone starts an AI feature does the 121surveys server – hosted in Germany – send the required texts directly to the chosen provider’s API, using your account’s key. The log records provider, model, task, token count and duration, not the content. A monthly budget in US dollars with a warning threshold and per-person limits stops further calls once it is reached.

What is not sent to the AI?

Before open answers and question texts leave the server, 121surveys replaces e-mail addresses, phone numbers, URLs, IP addresses and IBANs with placeholders such as [EMAIL]. Test submissions, abandoned responses and answers still waiting for credits are left out. Questionnaire review, optimising and translating only send questionnaire texts, and the generators only your company description – no answers.

How are AI results labelled?

In line with Art. 50 of the EU AI Act, every AI result and every AI suggestion in the interface carries the label “AI-generated”: categories, analyses, optimisations, translation suggestions, generators and text analysis findings. Once you accept a suggestion, it becomes your text and loses the label. The questionnaire shows respondents unreviewed AI translations as “Machine-translated”.

What the AI cannot do

No complete anonymisation

The clean-up works with patterns: names, postal addresses or descriptions that make a person identifiable remain in the text, and long numbers are sometimes wrongly replaced as phone numbers. If you expect sensitive free text, switch on “PII Protection” on the question, which replaces recognised details with placeholders as soon as an answer is saved – or don’t ask for it openly.

Suggestions, not statistics

The AI only counts frequencies within the sample it is given, and themes can differ between batches. Language models make mistakes. Crosstabs, significance tests, NPS and conjoint estimates are calculated by 121surveys without AI; the AI results add to these figures but do not replace them.

Nothing happens without a person

The AI never changes the questionnaire on its own: a rewrite only replaces the question text after your click, with version history and undo. An AI translation only becomes reviewed text through a person. The AI features are not available via the MCP server, and without a stored key all AI features stay off.

Frequently asked questions

Do I need my own AI key?

Yes. The AI features only work with an API key of your account at OpenAI, Anthropic, Mistral or Google. A person with the Administrator role enters it under Manage Workspace → AI Settings and assigns a provider and model to each task; from then on, everyone with access in the account uses the same key, including in sub-accounts. Without a key, you get a prompt to set up AI instead of a result.

What does AI analysis cost?

121surveys charges no credits for AI features; credits are only used by completed and imported responses. Your provider bills the cost of the AI calls through your key. Before an analysis, an estimate shows the expected usage, and a monthly budget with a warning threshold caps spending for the whole account or for individual people.

Who can use the AI features?

Categorising, summarising, counting, questionnaire review, optimising, translating and the generators are available from the Editor role upwards. Only people with the Administrator role change keys, model assignments and the budget, and each of these changes is recorded in the audit log. The model recommendations per task are visible from the Viewer role.

Is the use of AI compatible with the GDPR?

121surveys signs a data processing agreement with customers under Art. 28 GDPR and hosts in Germany. Only question texts and open answers go to the AI provider, stripped of recognisable personal data beforehand, and only when someone uses an AI feature. You choose the provider yourself; whether processing there is permitted is your decision as the controller.

In which language does the AI respond?

What you read – themes, summaries, findings, explanations – the AI writes in the language of your interface. What goes into the questionnaire – optimised questions, rewrites, generated questions – is written in the survey’s language. Quoted question texts and answer options stay in the original, and the sentiment codes positive, neutral and negative are fixed values.

Can I find earlier AI results again?

Yes. Saved analyses stay with the question, together with model and sample size. Every run of the AI text analysis is stored with provider, model, person and time; a new run carries over ignored findings but not resolved ones – if the AI finds a problem again, it has not been fixed. Runs can be deleted one by one, and any question texts you applied stay in place.

Try the AI features

Create an account, add your AI provider’s key and run the AI text analysis on a draft questionnaire – before the first answer comes in. Building and testing a questionnaire is free.

Start for free