# AI Survey Creator for India: How Hercules Builds a Survey in Seven Steps

How an AI survey creator builds a survey in seven steps: brief, method, routing, questions, quality gates, deployment and analysis. With a worked example.

Published 2026-08-25 · Updated 2026-11-16
Canonical: https://hercules.works/blogs/hercules-survey-creation-process-india

Hercules Works builds a survey in seven steps: goal, methodology framework, flow and routing logic, questionnaire, a nine-dimension quality rubric with hard gates, deployment to the SuperJ panel, and handoff to the Poseidon analytics engine. You describe the study in plain language, then edit by chat. It covers 22 study types and as many Indian languages as the study needs.

## What an AI Survey Creator Should Do Besides Writing Questions

Most AI survey tools write a list of questions from a prompt. That is the easy part. The hard parts come before and after: deciding what you need to learn, choosing a method that can answer it, routing people so they only see questions that apply, and checking the draft before real respondents see it.

This guide walks through how Hercules Works, which we make, handles those parts. It is a vendor's account, so each step ends with something you can check for yourself. For a tool-neutral comparison of prompt-to-survey tools, see our [AI survey generator guide](/blogs/ai-survey-generator/). Here we stay with one pipeline and show a worked example.

## The Seven Steps

| # | Step | What it produces | What you do |
|---|---|---|---|
| 1 | Goal | A research brief: the business decision, objectives, key questions and target audience | Correct the brief. Everything downstream follows from it |
| 2 | Methodology framework | A method chosen for the objective, returned for you to approve or refine | Check the method fits the decision |
| 3 | Flow and routing logic | The section order and the skip, branch and carry-forward rules, written out before any question exists | Read the paths for who sees what |
| 4 | Questionnaire | Questions with answer options, scales and routing attached | Read every question aloud |
| 5 | Quality rubric | A score on nine dimensions, with hard gates | Fix anything it flags |
| 6 | Deployment | The survey is compiled and deployed to SuperJ | Preview it, pick the audience |
| 7 | Analytics handoff | When the study completes, responses go to Poseidon | Ask your questions, read the report |

Writing the flow in step 3 before the questions in step 4 is deliberate. A survey whose routing is designed afterwards tends to collect answers from the wrong people. Step 2 draws on a library of 22 study types, listed below.

You can also start without the full pipeline. Our How to Hercules page describes a Quick Question that "skips the survey builder entirely", a Guided Workflow that "walks you step by step from objective to launch", and a builder for your own questionnaire (the Plus plan lists an "Unlimited survey builder"). See [How to Hercules](/how-to-hercules).

## The Nine-Dimension Quality Rubric

Before a survey can be published, it is scored on nine dimensions. Five of them are hard gates: a zero on any of these blocks publishing.

| Dimension | Hard gate |
|---|---|
| Objective coverage | Yes |
| Measurement validity | No |
| Wording neutrality | Yes |
| Response design | No |
| Experimental integrity | Yes |
| Routing | Yes |
| Accessibility | No |
| Ethics | Yes |
| Reporting transparency | No |

Measurement validity means standard instruments and exact scales, such as NPS, CSAT and a 9-point hedonic scale. Routing is validated for reachability and correctness before deployment.

The rubric is a check, not a verdict on whether your question is worth asking. Use our own [methodology checklist](/blogs/survey-methodology-best-practices/) for the judgement a rubric cannot make.

## Conversational Editing

After the first draft you change the survey by chatting. The editor applies exact changes to wording, options, question type, routing and design, and then tells you what it changed and what it could not do. Requests that work well are specific:

- "Reword Q7 in simpler English."
- "Show Q28 only the options they picked in Q27."
- "Add a screener for two-wheeler owners and end the survey for everyone else."
- "Move the price question after the concept exposure."

Every survey version and logic version is stored in its version history. A survey moves through a status lifecycle: draft, review, approved, live, paused or rejected, and finally completed.

## Question Types and Routing Features

| Question types | Notes |
|---|---|
| Single and multi-select | Multi-select can carry a minimum and maximum |
| Rating scales and grids | Numeric and labelled, NPS 0-10 and 5-point satisfaction included |
| Ranking | Best for short lists |
| Open text | For verbatim answers |
| Yes/No | Screeners and simple checks |
| MaxDiff | Best-worst exercises for longer lists |
| Image, video and audio stimuli | As a question or as answer options |

| Routing feature | What it does |
|---|---|
| Skip logic | Moves people past questions that do not apply |
| Screen-out | Ends the survey for ineligible people |
| Complete early | Ends the interview for a segment |
| Carry-forward | Options picked in one question become the options in a later one |
| Option loops | Repeats a battery once per item picked, such as each brand used |
| Piping | Quotes an earlier answer in later question text |
| Display logic | Shows or hides a question on a condition |
| Consistency checks | Compares a later answer with an earlier one |
| Randomisation | Shuffles answer lists while anchored options stay put |
| Rotation | Monadic and sequential-monadic exposure of stimuli |

Detailed advice on skip logic is in our [skip logic guide](/blogs/skip-logic-survey-tool/). Age band, gender, city and NCCS are already known from SuperJ profiles, so you do not need questions for them.

## 22 Study Types and the Designs Behind Them

Each study type carries a playbook of KPIs and design choices. The 22 are: usage and attitude, brand health and tracking, concept testing, product testing, packaging, ad pre-test, ad tracking, pricing, feature and claim prioritisation, CSAT, customer experience, churn, segmentation, market sizing, shopper path, customer profiling, B2B decision-maker, public opinion, employee engagement and eNPS, usability, mixed methods, and exploratory.

Designs include single, pure, sequential, proto and reverse sequential monadic, paired comparison, side-by-side, MaxDiff, Kano, TURF, Van Westendorp, Gabor-Granger and monadic price cells. It picks by counting stimuli: none means a descriptive survey, one means a single monadic read, two to four means the monadic family, and five or more points to MaxDiff or a screening round.

Not available: choice-based conjoint, constant-sum allocation and timers. For Gabor-Granger and Van Westendorp the builder designs the survey, but Poseidon does not run the analysis. See [pricing research](/blogs/pricing-research-platform-india/) and [Kano](/blogs/kano-model-survey-tool/).

## Worked Example: One Prompt and the Survey It Yields

The prompt and structure below are illustrative. They show the shape of a draft, not a real study or its output.

> Prompt: "I need to choose which of three masala chai flavours to launch as a ready-to-drink. Test them with tea drinkers aged 18 to 45 in Tier 1 and Tier 2 cities. Keep it under 10 minutes. Field in Hindi and English."

| Part | What the draft contains |
|---|---|
| Brief | Decision: pick one flavour to launch. Audience: tea drinkers, 18-45, Tier 1 and 2 |
| Method | Concept test, sequential monadic: each respondent sees all three concepts one at a time, in rotated order |
| Screener | Which of these drinks have you bought in the last month? Tea sits among decoys. People who did not buy tea are screened out |
| Warm-up | How often they drink tea, and in which occasions |
| Concept block, three times | Concept shown as an image. Purchase intent (5-point), uniqueness, relevance, believability, one open-ended "what do you like or dislike?" |
| Display logic | A follow-up "why not?" shown only to those who say they would definitely not buy |
| Piping | The concept's name is quoted back in each question |
| Closing | Preferred of the three seen, and usual brand of tea |
| Quality | Attention checks about one in every seven questions |
| Handoff | Poseidon reports top-2-box purchase intent by concept and city tier |

Notice what is missing. It does not ask age, city or gender, because the profile already holds them. Rotating the order of the three concepts stops the first one shown from always getting the freshest attention. You would check, as in step 5, that the "why not?" question really is reachable.

## Running This Study on Hercules

If your job is a consumer survey in India, the pipeline above takes you from a prompt to a live study. Rehearse the routing on the Free plan first, which deploys to the general population and lets you send SuperJ survey links to friends and family. Then target the real audience on a paid plan.

Pricing is per response, and the free plan covers a 30-response pilot: see [pricing](/pricing). How respondents are verified and sampled: [panel methodology](/panel-methodology).

SuperJ has 2 crore Indian respondents. The verified, survey-eligible panel is a subset, and availability depends on your targeting by city, tier, NCCS, age and gender (male or female). Aadhaar-based KYC is done once, and each study gets a zero-knowledge proof. Surveys run in as many Indian languages as the study needs. See the [panel methodology page](/panel-methodology). The limit: the panel is app-based, so smartphone users only. Offline households need face-to-face fieldwork.

## Frequently asked questions

### Can an AI survey creator write my whole questionnaire?

Yes, as a draft. Hercules drafts the brief, method, routing and questions from your prompt, then scores the result on a nine-dimension rubric. You still need to read every question and test every routing path before launch.

### What is the difference between an AI survey creator and an AI questionnaire generator?

A generator writes questions. A creator, as we use the term, also sets the goal and method, builds routing, checks quality and deploys to respondents. Our AI survey generator guide covers the question-writing side.

### Can I edit the AI draft by chatting?

Yes. You can ask to reword a question, change options or type, add a screener or change routing. The editor applies the change and says what it did and what it could not do. Each version is kept in version history.

### What survey logic is supported?

Skip logic, screen-outs, complete-early, carry-forward, option loops, piping, display logic, consistency checks, randomisation, and monadic and sequential-monadic rotation.

### Which languages can surveys run in?

as many Indian languages as the study needs. Ask for the specific languages you need, and have a native speaker read each translation, including scale labels and brand names.

### Can I target an audience on the Free plan?

No. The credits and pricing help page says the Free plan has no audience filters and deploys to the general population only. Targeting by city, NCCS or age needs Plus or Pro.

### Can I bring my own questionnaire?

Yes. You can paste or upload an existing questionnaire, or a PDF, and the platform still builds the study design, business flow and routing logic around it. You can also build your own in the builder, which the Plus plan lists as unlimited.

### What does the quality rubric block?

A survey scoring zero on a hard-gate dimension cannot be published. The gates are objective coverage, wording neutrality, experimental integrity, routing and ethics.
