A HubForge playbook

Designing Surveys and Forms with AIPlaybook

Design the forms your teams fill every week and the surveys that show whether a programme worked. Then do it faster with AI, without handing over your judgement.

12 chapters6 stages4 tools53 ideas40 terms explained

Rate this playbook

Where do you want to start?

Keep the working tools close

Read, plan, then build.

Take the playbook offline, work through the companion workbook, or start from a ready-to-adapt XLSForm.

Chapter 1 ·

What is wrong with these questions?

Eight questions from eight fields. Each hides one classic flaw.

  • Most data problems are designed in
  • The same eight flaws turn up in every sector
  • They are easy to miss in your own draft

Why this matters

Every report is only as good as the questions behind it.

Chapter 2 ·

Monitoring form or evaluation survey?

One steers the work while it happens. The other judges whether it worked. Decide before you write.

  • Different jobs, people and rhythm
  • Monitoring: short, routine, everyone
  • Evaluation: sampled, comparable, locked

You need

Answer two questions

Your answer appears here.

Side by side

Same phone, different job.

Monitoring formEvaluation survey

Different rhythms

Every month, or twice and compared.

Monitoring: every visit, every monthMonth 1Month 24Data every month, from every household. Used to steer the work.Evaluation: measured twice, compared with a group without the programmeBaselineEndlineProgramme villagesComparisonthe differenceSame questions, same way, both times

Meet the two examples

One programme, two instruments.

A fictional nutrition programme in 60 villages. You need no nutrition knowledge; the lessons work in any field.

From decision to question

Every question traces back to a decision.

If you cannot fill all four boxes, the question is not ready.

DecisionIndicatorQuestionTable it fillsMonitoringWhich children needa clinic referralthis week?Children whoseweight fell belowthe expected lineChild's weight today(kg)Children below theline, by village andworkerEvaluationDid the programmeimprove children'sdiets?Share of children 6to 23 months withminimum dietarydiversityFood groups thechild ate yesterday(standard module)Share meeting theindicator, programmevs comparison,baseline vs endline

Chapter 3 ·

Six stages. One question each.

The SURVEY method works for both kinds. Each stage asks one question of your draft.

  • Start from decisions, not questions
  • Write for the person answering
  • Test with real people before launch
Stage 1 of 6

Chapter 4 ·

Why it works.

Answers fail in predictable places. The method is built on decades of survey research about where.

  • Every answer passes through four mental steps
  • Error comes from who answers and what answers mean
  • People answer well when it feels worth it

Tourangeau, Rips and Rasinski, 2000

Every answer takes four steps.

Tap a step to see how it fails, and which stage of the method protects it.

Groves and Lyberg, 2010

Error comes from two sides.

Total survey error: tap any source to light up the stages that reduce it.

Four more findings

What research says about people and forms.

Chapter 5 ·

The field guide.

The rules to keep open while you write.

  • Six parts of a good question
  • Ten wording rules, each with a fix
  • A bank of tested questions to borrow
In the last 7 days, on how many daysdid your child eat any fruit?01 to 23 to 45 to 6Every day (7)Don't knowPrefer not to sayHint: count days, not the number of fruits.Asked of: the child's main carer1 A time periodeveryone answers aboutthe same days2 A count, not"often" or "rarely"3 Honest exitsso nobody has to guess4 One ideafruit only; vegetablesget their own question5 No overlapsevery number fitsexactly one option6 Clear who answersand a hint for thetricky part

Chapter 6 ·

How you ask, and who answers.

The same question gets different answers on a doorstep, a phone call or a screen. Design for the mode and for every respondent.

  • Each mode has its own rules
  • Design for low literacy and disability
  • Protect people on sensitive topics

Six modes

Pick a mode to see its rules.

Reaching everyone

Design for the people least likely to be heard.

Chapter 7 ·

Designing with AI.

AI drafts fast and checks well. It cannot tell you what your respondents understand.

  • Brief it like a colleague
  • Know its six typical mistakes
  • Keep the six promises

AI is good at

    AI is not good at

      Chapter 8 ·

      The workshop.

      Four tools to use on your own draft. Each one teaches the rule behind it.

      • Check a question
      • Feel skip logic as a user
      • Export an XLSForm

      Paste a question and its options. The doctor flags common wording problems. It catches patterns, not meaning.

      Try one:

      Then ask your AI

      Chapter 9 ·

      From paper to phone.

      Design on paper. Pick the tool. Let AI write the XLSForm, and let the tool check it.

      • A few questions pick the tool
      • Two ready XLSForms to download
      • The converter is your spell-checker

      Tick what is true for you

      Best fits

      Compare all eight tools
      ToolHow you buildOfflineLanguagesCostBest for

      Plans, prices and limits change; check each tool's own pages.

      Ready to use

      Two XLSForms, checked with ODK Validate.

      Upload either one to KoboToolbox, ODK Central or SurveyCTO, then change it to fit your programme.

      The recipe

      Let AI write the XLSForm. Let the tool check it.

        AI often invents functions or misspells a list name. The tool's converter is your spell-checker. Use it every time.

        Chapter 10 ·

        Ready to launch?

        Ten checks for your kind of instrument. Tick them off.

        • Separate lists for each kind
        • Most failures are caught here
        • Print it for the team
        0/10

        Start with the decisions.

        Chapter 11 ·

        Fieldwork, and after.

        In a survey the interviewer is part of the instrument. After collection, the codebook and the feedback loop make the data usable and fair.

        • Train interviewers question by question
        • Check data every day, not at the end
        • Give results back to the people who answered

        West and Blom, 2017

        The interviewer is part of the instrument.

        Answers vary by who asks. A manual and good training make every interviewer ask the same question.

        Training

        Four steps before anyone goes to the field.

          During fieldwork

          Check every day, by interviewer.

          After collection

          Seven habits that make data usable, and fair.

          Chapter 12 ·

          The library of ideas.

          The research and practice behind every stage, a glossary of every term, and a one-page cheat sheet.

          • 53 ideas in plain words
          • A glossary of 40 terms
          • A printable cheat sheet
          Shashidhar S A

          written by

          Shashidhar S A

          Founder, HubForge Global · MERL & AI Innovator

          • En-Intrapreneur
          • Driving evaluation ecosystems through People, Process & Product
          • Co-Chair, IOCE Comms & Storytelling
          • Co-Lead, EvalYouth Global Champion Engagement

          A note for the next reader

          Was this useful?

          Rate the playbook and, if you like, leave a short note about what worked.

          Loading community features…

          Your rating

          An anonymous browser ID remembers your rating. You can update your rating from this browser.