Skip to content
Open Screeb

Analyze your data with your own AI tool

Export your Screeb responses and use ready-made prompts to analyze them in ChatGPT, Claude, Gemini or your internal assistant.

How-to5 min read, updated Nov 16, 2026

Get your survey responses out of Screeb and analyze them with the AI tool of your choice, with prompts ready to copy.

  • Check your company policy and your data processing agreements before you send data to an external AI tool.
  • Remove or mask the personal data you do not need for the analysis (names, emails, phone numbers).

An AI tool only knows what you give it. Pick the source that fits your workflow:

Source Best for
Export survey responses A one-off analysis of one survey: upload the file to your AI tool.
Stitch A continuous flow into your data warehouse, then into your AI tool.
Webhooks Each response pushed to your own systems as it arrives.

Good prompts share the same structure:

  1. Role: who the AI acts as.
  2. Context: what the survey is, who answered, when, how many responses.
  3. Data: what you upload, and what each column means.
  4. Task: one precise question.
  5. Output format: table, bullet list, top 5, length limit.
  6. Guardrails: ask for quotes as evidence, and to say “not enough data” rather than guess.

In the prompts below, replace the [placeholders] with your own values.

Data needed: an export of the open-text answers of one survey (one answer per row, with the date).

You are a product researcher. The attached file contains open-text answers
to the question "[question]" from [number] users of [product], collected
between [start date] and [end date].
Group the answers into 5 to 8 themes. For each theme give:
- a short name,
- the number and share of answers,
- two verbatim quotes as evidence,
- a one-sentence summary.
Put answers that match no theme under "Other". Do not invent quotes.

Data needed: an export of an NPS survey with the score, the follow-up comment and the date.

You are a customer experience analyst. The attached file contains NPS
responses (score 0-10) with an optional comment, from [start date] to
[end date].
1. Compute the NPS (% promoters 9-10 minus % detractors 0-6) and give the
counts for promoters, passives and detractors.
2. List the 3 main reasons mentioned by detractors and the 3 main reasons
mentioned by promoters, with one quote each.
3. Tell me which topic, if fixed, would most likely improve the score.
Show the calculation. If a figure cannot be computed from the file, say so.

Data needed: responses from several periods (for example the last 6 months), with the date of each response. Stitch is the most convenient source.

You are a data analyst. The attached data contains [metric: NPS, CSAT, CES]
responses for [product] from [start date] to [end date].
Show the evolution month by month in a table (responses, average score,
share of positive answers). Then point out:
- the largest increase and the largest decrease,
- the themes in the comments that appear or disappear over time,
- any period with too few responses to conclude.
Do not explain a variation without evidence in the data.

Data needed: responses together with user properties (plan, country, role, seniority…). Properties are included in the exports and in Stitch when you send them to Screeb.

You are a product analyst. The attached data contains survey responses
and these user properties: [plan, country, role].
Compare the segments by [property]. For each segment give the number of
responses, the average score and the top 3 themes in the comments.
Highlight the segment that differs most from the others and propose two
hypotheses to explain it. Mark clearly what is a fact from the data and
what is a hypothesis.

5. Turn feedback into a prioritized list of actions

Section titled “5. Turn feedback into a prioritized list of actions”

Data needed: an export of a feedback or feature request survey.

You are a product manager. The attached file contains feedback sent by
users of [product].
Produce a backlog of the 10 most important requests or problems. For each:
- a title and a one-sentence description,
- how many users mention it,
- the impact on users (high/medium/low) and why,
- one representative quote.
Sort by number of users first, then by impact. Merge duplicates.

Data needed: responses from a cancellation, exit or low-score survey.

You are a customer success analyst. The attached file contains answers
from users who [cancelled / gave a score under 7 / left the product].
Identify the recurring reasons, ranked by frequency. Separate what we can
fix (product, pricing, support) from what we cannot (change of need,
company closure). For each reason, suggest one action and the team that
should own it. Quote the answers that show the strongest frustration.

Data needed: the output of the previous analyses, or the raw export.

You are writing for the executive team. From the attached analysis,
write a one-page summary with: 3 key findings, 3 risks, 3 recommended
actions, and the number of responses behind each statement. Use plain
language, no jargon, and keep it under 250 words.
  • One question per prompt. Chain several short prompts instead of one long one.
  • Ask for evidence. Quotes and counts let you check the AI’s claims against the data.
  • Check the numbers. AI tools can miscount on large files: ask them to show the calculation, and compare a few figures with the results shown in Screeb.
  • Mind the volume. If your file is too large for the tool, split it by period or by survey, then ask for a synthesis of the partial results.
  • Name your columns. Tell the AI what each column means (for example “score is the answer to the NPS question, from 0 to 10”).
  • Keep the language consistent. If answers are in several languages, ask for themes in a single language and keep the quotes in their original language.