If Your Respondents Can't Type, Your Form Is Broken: Surveys for Low-Literacy Audiences
Nearly 30% of rural Indian women aged 7+ can't read a simple message. Compare CAPI, IVR, audio self-interviews and voice web forms for low-literacy surveys.
The short version
- A typed form screens out readers before it asks a question: India's 2023-24 labour force survey put rural women's literacy at 70.4% (age 7+), against 84.7% for rural men and 88.9% for urban areas.
- The phone gap stacks on the reading gap. In GSMA's 2024 survey, 71% of Indian adult women owned a mobile and only 36% owned a smartphone, so a web link alone misses most rural women.
- IVR reaches feature phones but few people pick up: in a 2017 randomised trial, IVR response was 0.8% in Bangladesh and 1.1% in Tanzania, against 5.4% and 8.6% for live calls.
- Enumerator-led CAPI on KoBoToolbox or ODK is still the safest default for non-readers. Voice web forms fit when people have a smartphone and you want their own words, without an interviewer reading every question.
A text survey sent to rural India filters out nearly 3 in 10 women before they see the first question. India's own statistics office puts literacy among rural women at 70.4%, and literacy there means reading one simple message. If you need a survey for illiterate respondents, the input method is the first design decision, not the last.
The respondent is not the problem. The text box is. We compared the four methods field teams actually use to survey people who don't read: interviewers with tablets, automated phone calls, audio self-interviews and voice web forms. Each has a cost, and the evidence says the obvious winner is not always the right one.
To survey people who cannot read, remove reading and typing from the task. Use an interviewer with a tablet app such as KoBoToolbox or ODK, an automated phone call (IVR), an audio self-interview on headphones, or a voice web form. Choose by device access: interviewers for no phone, IVR for feature phones, voice forms for smartphones.
Who a typed form leaves out
Start with the official definition. India's Periodic Labour Force Survey counts a person as literate if they can read and write a simple message with understanding in at least one language. By that low bar, rural literacy was 77.5%, with women at 70.4% and men at 84.7%. In urban India the figure for all persons was 88.9%.
The state numbers matter more than the national ones if you work in a specific district. Rural women's literacy in the same table was 61.8% in Rajasthan, 62.6% in Madhya Pradesh and 65.0% in Bihar. In those places a typed survey loses well over a third of the women it was meant to reach.
| Group (PLFS 2023-24, age 7+) | Men | Women | All persons |
|---|---|---|---|
| Rural India | 84.7% | 70.4% | 77.5% |
| Urban India | 92.9% | 84.9% | 88.9% |
| Rural Rajasthan | 83.6% | 61.8% | 72.5% |
| Rural Bihar | 81.5% | 65.0% | 73.3% |
Literacy figures flatter the people who can read
Being counted as literate is not the same as being comfortable with a 20-question form. Writing a simple message once is a long way from typing a paragraph on a phone keyboard in a second script. We think the real share of people who will struggle with a typed survey is higher than the literacy gap suggests, though no official statistic measures it directly.
The global picture runs the same way. UNESCO's literacy page says 739 million adults still cannot read and write, and two-thirds of them are women. A survey that depends on reading is a survey that undercounts women, almost everywhere.
The phone gap stacks on the reading gap
Literacy is only the first filter. The second is the device. GSMA's 2025 report, an industry study based on its own 2024 consumer survey, found that 71% of Indian adult women owned a mobile phone, but only 36% owned a smartphone. For men, the figures were 84% and 58%.
- Own a smartphone36%
- Own a basic or feature phone only35%
- Own no mobile phone29%
GSMA also reports that, among people who know about mobile internet but don't use it, reading and writing difficulties are among the top barriers. That is the same barrier your form puts in front of them. Design around it, or accept a sample of people who already read well.
Four ways to survey people who don't read
Every method below replaces reading with listening and writing with speaking or tapping. They differ in who holds the device, who pays for the connection, and whether a stranger hears the answer. Here is the trade-off in one table.
| Interviewer with tablet (CAPI) | Automated phone call (IVR) | Audio self-interview (ACASI) | Voice web form | |
|---|---|---|---|---|
| Who holds the device | Trained enumerator | Respondent | Enumerator brings it, respondent uses it | Respondent |
| Respondent needs | Nothing but time | Any mobile, even a basic one | Headphones and a short demo | Smartphone, data, one tap on a link |
| Main strength | Reaches everyone, handles consent and confusion | Cheap per call, no data needed | Privacy on sensitive topics | Own words, own time, no interviewer cost |
| Main weakness | Travel cost; a stranger hears every answer | Very low response; keypad answers only | Still needs a visit; closed answers only | Excludes people without smartphones; needs a signal |
| Where it beats a voice web form | Coverage of people with no phone | Reaches basic phones | Private answers on sensitive yes or no questions |
We would not pick a method from this table alone. Decide first who owns a working device in your target group, then which questions are sensitive. The rest of the post works through each method on that basis.
Interviewer-led CAPI is the default for a reason
Computer-assisted personal interviewing means an enumerator visits, reads each question aloud and taps the answers into a tablet or phone. The two tools you will meet most often are KoBoToolbox and ODK. Both are built for offline fieldwork, and both let you add an audio question that records from the device microphone.
Cost is rarely the blocker. As of October 2026, KoBoToolbox's pricing page lists a free Community plan for nonprofits, government agencies and educational institutions, with 5,000 submissions a month and 1 GB of file storage. ODK's server, Central, is open source and free to self-host if you can run a Linux server. We compare the field tools in more depth in our guide to field research forms.
Where an interviewer helps
An enumerator explains consent, rephrases a confusing question and notices when someone is tired. For a respondent who has never used a form, that human bridge is the whole method. No remote tool replaces it on a first visit.
Where an interviewer hurts
The same person who helps also listens. On questions about money, health, violence or caste, a respondent may give the answer they think the visitor wants. Interviewers also summarise. A 2-minute answer becomes a 6-word note in a text field, and the respondent's own phrasing is gone. Recording the answer as audio, which both KoBoToolbox and ODK support, keeps the original.
IVR reaches basic phones, but few people answer
Interactive voice response calls a number, plays recorded questions and takes answers through the keypad. It needs no smartphone, no data and no reading. On paper it is the perfect tool for the middle segment in the chart above.
The response data is less kind. Pariyo and colleagues randomised mobile numbers in Bangladesh and Tanzania to either a live interviewer call (CATI) or an IVR call, in 2017. IVR response was 0.8% in Bangladesh and 1.1% in Tanzania, against 5.4% and 8.6% for live calls. The paper also found IVR respondents were more often men, in both countries.
- Live interviewer (CATI)
- Automated (IVR)
When IVR is still the right call
Our view is that IVR earns its place for short, repeated check-ins with people who already know you. A weekly 3-question call to farmers enrolled in a programme is very different from a cold call to a random number. Keypad answers also limit you to numbers and choices. If you need reasons in someone's own words, IVR is the wrong tool.
Audio self-interviews protect privacy on sensitive questions
Audio computer-assisted self-interviewing, or ACASI, splits the difference. An enumerator brings a tablet and headphones. The respondent hears each question privately and answers by tapping coloured or picture buttons. The enumerator never hears the question or sees the answer.
It works with people who have never touched a computer. Bhatnagar and colleagues tested a colour-coded version with truck drivers and their wives in semi-rural Tamil Nadu. Only about 10% had ever used a computer, yet roughly 75% reported no problem using the system. In the same study, 63% of the wives were illiterate or had only primary education.
The cost is that ACASI still needs a visit, and answers are closed. You get a tap on a green circle, not a sentence. For sensitive yes or no questions, that is exactly right. For "why", it is not.
Voice web forms let respondents hear and speak
A voice web form is a link that opens in the phone browser. The question can be a recording in the creator's own voice. The answer is a recording too. The respondent taps play, listens, taps the microphone and speaks, in whatever language they think in.
This is the method we build. Would You Rather shows one question per screen, plays a recorded prompt where you add one, and accepts a spoken answer with a time limit you set. An optional "type instead" box exists, and it is off by default, which suits audiences who don't write. You can open the form without sign-in and share it as a QR code. We cover the mechanics in adding audio questions to a form.
What a voice form does not fix
Be honest about the gaps. The interface still has written buttons, such as "Next" and "Submit". Choice options are text labels, so a non-reader needs them read aloud in the prompt, or needs the question turned into a spoken answer. The form also needs mobile data to load and to upload each recording.
Would You Rather also does not transcribe answers. You get audio files and links in your results or Google Sheet, and you listen or run your own transcription. For a rural study, that is often an advantage: a local team member who speaks the dialect will hear more than any speech engine. We explain the trade-off between spoken and typed answers in why voice answers run longer.
How to build a form for someone who cannot read it
The method matters less than the question design. These steps apply whether you run CAPI, ACASI or a voice form, and they come straight from the barriers above.
- 1Map device access before you pick a toolAsk a local partner who in the group owns a smartphone, a basic phone or none. That answer chooses between voice forms, IVR and enumerators.
- 2Write questions in the spoken language, not the official oneRecord each question as a native speaker would say it at home. Formal translations confuse listeners as much as readers.
- 3Read the options aloud inside the questionIf a question has choices, say them in the recording: 'Is it one, two, or more than two?' Keep lists to 4 options or fewer.
- 4Turn reasons into spoken answersReplace long checklists with one open question answered by voice. 'Tell us how you sold your last harvest' beats 12 tick boxes.
- 5Plan the first contact in personEven with a self-completed form, someone should explain consent and show the microphone button once. A field visit or community meeting works.
- 6Pilot with 5 to 10 people and watch themSit beside them, say nothing, and note every hesitation. Rewrite or cut those questions before launch.
If your audience speaks several languages, we wrote separately about running one form in Hindi, Tamil or any language. For in-person launches, a printed QR code at a community meeting is often the simplest bridge. See QR code forms at events.
When a voice web form is the wrong tool
We build voice forms, so take this section seriously. There are clear cases where you should not use one.
No smartphone, no data, no form
If most of your respondents fall in the second or third segment of the phone-access chart, a web link cannot reach them alone. Use enumerators, or let an enumerator hand over a phone running the voice form. That hybrid keeps the respondent's own words while the visit handles access.
Representative samples need a sampling frame
A link shared in a WhatsApp group reaches whoever is in that group. It will skew towards younger, more connected people, the same skew the IVR trial found. If your study must estimate a rate for a district, you need a proper household sample and probably interviewers. A voice form is better suited to qualitative depth, follow-ups and groups you already know.
Our evidence is thinner than we would like
The strongest numbers in this post describe the problem: literacy and phone access. The evidence on methods is thinner. The IVR trial used random dialling, the hardest case. The ACASI study measured acceptability, not answer quality. We know of no published randomised comparison of voice web forms against CAPI with low-literacy respondents, and we have not run one. Until someone does, treat voice forms as promising for this group, not proven.
Where to start this week
Pick one question from your current survey that matters most and produces the thinnest answers. Record it in the local language, in your own voice. Then put it in front of 5 people from your target group, in person, and let them answer by speaking.
Listen to the 5 recordings. If they say more than your typed responses usually do, extend the pilot. If they stall at the screen, you have learned your group needs an enumerator, and that is a useful answer too. Either way, you have tested the method for the cost of an afternoon. When you are ready to analyse, send the answers to Google Sheets with audio links.
References
- National Statistical Office, India, Periodic Labour Force Survey Annual Report, July 2023 to June 2024, September 2024. Used for literacy rates by sector, sex and state (Table 7), the definition of literacy and the sample size.
- UNESCO, Literacy, citing UNESCO Institute for Statistics 2025 estimates, accessed 8 October 2026. Used for the global count of adults who cannot read and write.
- GSMA, The Mobile Gender Gap Report 2025, industry research based on the GSMA Consumer Survey 2024. Used for mobile and smartphone ownership in India and literacy as a barrier to mobile internet.
- Pariyo, G. et al., Effect of the Data Collection Method on Mobile Phone Survey Participation in Bangladesh and Tanzania, JMIR Formative Research, 2023. Used for IVR and CATI response rates and respondent profile.
- Bhatnagar, T. et al., Color-Coded Audio Computer-Assisted Self-Interviews for Poorly Educated Men and Women in a Semi-rural Area of South India, AIDS and Behavior, 2013. Used for ACASI feasibility with low computer experience.
- KoBoToolbox, Pricing, accessed 8 October 2026. Used for the Community plan's eligibility and limits.
The weakest point in this source base: the method comparison rests on one phone trial in two countries and one acceptability study, and the phone-ownership figures come from industry research rather than official statistics.
Questions people ask
How do you survey people who cannot read?
Replace reading and writing with listening and speaking. The four common methods are an interviewer reading questions aloud and entering answers on a tablet (CAPI), automated phone calls with keypad answers (IVR), audio self-interviews where respondents hear questions through headphones, and voice web forms where they hear a recorded question and answer out loud. Most large studies use interviewers, because they also handle consent and confusion.
What is the best survey tool for illiterate respondents?
For interviewer-led household surveys, KoBoToolbox and ODK are the standard choices. Both are free to start, work offline on Android, and record audio. For automated phone surveys on feature phones, use an IVR platform. For respondents with smartphones who can tap a link, a voice web form lets them hear the question and answer by speaking. Pick by who holds the device, not by feature list.
Do IVR surveys work in rural areas?
They reach people without smartphones or data, which is their real strength. But response is low. In a 2017 randomised trial in Bangladesh and Tanzania, about 1% of numbers called by IVR produced a response, against 5% to 9% for calls by a live interviewer. IVR respondents were also more often men. IVR works best for short, repeated check-ins with people who already expect the call.
What is ACASI in survey research?
ACASI stands for audio computer-assisted self-interviewing. The respondent wears headphones, hears each question from a recording, and answers by tapping the screen, often on colour-coded or picture buttons. No interviewer sees the answer. It is used for sensitive topics such as sexual health or violence. A 2013 study in semi-rural Tamil Nadu found most participants managed it even though only about 10% had used a computer.
Can a voice survey replace face-to-face interviews?
Sometimes, not always. A voice web form removes the need for the respondent to read or type, and the answer arrives in their own words. But it still needs a smartphone, mobile data and someone to explain consent. For first-time contact with people who have never used a form, an enumerator is safer. Voice forms work well for follow-ups, diaries and groups already used to WhatsApp voice notes.
How do you design survey questions for low literacy respondents?
Ask one question at a time, in the respondent's spoken language, with a recorded audio prompt. Keep answer options to a small number and read them aloud in the prompt. Use pictures where an option can be drawn. Prefer open spoken answers over long lists. Pilot the form with 5 to 10 people from the target group, watch where they hesitate, and cut or rewrite those questions before launch.
Drafted with AI from published research, then edited and approved by Shivam Singh.
Would you rather hear the answer?
Make a voice form in a minute. It’s free.
Create a form