The dollar problem: what AI recommends to Indian small businesses buying software

01 / 07Summary

An Indian shop owner who asks an AI assistant which software to buy gets two different answers, depending on whether the question says India. We asked 36 software questions across 12 categories on four answer engines, from an Indian connection, twice each, and repeated the questions without an India cue from the United States: 336 answers naming 214 products, each traced to its company’s headquarters.

The location of the person asking made little difference. The words of the question made almost all of it.

02 / 07Why this matters

A small-business owner in India can now ask ChatGPT, Gemini, Perplexity or Google’s AI Mode which software to buy. Not everyone will think to add “in India”. They type the question the way they would ask a friend, and the engine already knows, from the connection, where they are.

For buyers, a wrong-market answer costs time, a migration or a tool that cannot do the job. For Indian software companies, it is a recommendation they never get. We wanted to measure how often it happens and what changes it.

Asked from India for “the best POS system for a small retail store”, the engines named a product 29 times. Not once was it Indian.

Study 003, Fig. 1

03 / 07Method

We wrote three questions for each of 12 software categories, in the words a small-business owner uses.¹ One names India. One carries an Indian cue without the word India: a city, a rupee budget, GST, PF and ESI, IndiaMART leads, Diwali, a kirana store. One has no India cue at all. Every question was asked twice on each of four engines with the request sent from India. The 12 no-cue questions were also asked once from the United States, as a control.²

Answers

336288 from India, 48 from the US

Questions

3612 categories × 3 phrasings

Products

2141,589 mentions; HQ traced for 211

Engines

ChatGPT (search on), Perplexity, Gemini, Google AI Mode

Runs

2 per question from India; 1 per no-cue question from the US

Fieldwork

5 October 2026

What counted

A mention is a product’s name in the answer text, counted once per answer. Citation labels and source links were removed first, so a source titled “Zoho” is not a mention. A product counted only in a category where it competes: “syncs with Shopify” in an answer about WhatsApp tools is not a recommendation of Shopify.³ We report mentions, not endorsements: a product named in order to be ruled out still counts. Citations, the links presented as sources, were logged (5,210 in all) and are published but not analysed here.

A product is Indian if the company that sells it, or the company that owns it, is headquartered in India.⁴ Freshworks, founded in Chennai and headquartered in San Mateo, California, counts as foreign. Counting by where companies were founded instead lifts the no-cue share from 16% to 21% of 444 mentions; the gap between cued and uncued questions remains. Three products whose headquarters we could not verify are left out of every share.

A price is any amount written with a currency: ₹, Rs, INR or rupees; $, USD or dollars.⁵ Each answer was classed as rupees only, dollars only, both, or no price.

Disclosure: some brands that appear in this study are clients of Citation Research. Their results were produced by the same questions, engines and rules as every other brand’s.

04 / 07Findings

1. The question decides the answer

With no India cue, 69 of 444 product mentions in India-sent answers were Indian: 16%. Half of those answers, 48 of 96, named no Indian product at all. When the question named India, the share was 65% (316 of 483). An indirect cue worked as well as the word itself: 65% (274 of 423). The first product named followed the same pattern, Indian in 76 of 96 India-named answers and in 17 of 96 no-cue answers.

2. Compliance categories turn Indian; general tools stay foreign

Where Indian law shapes the product, a cue changed everything. For POS, all 77 product mentions in answers to cued questions were Indian; for the no-cue question, none of 29 were. Payroll went from 26% (9 of 34) to 100% (69 of 69). Where the product is the same everywhere, a cue did little: project management was 10% Indian even when the question named India (8 of 79), email marketing 15% (11 of 71).

The most-named product in each category tells the same story. Zoho Books appeared in all 24 accounting answers sent from India and Zoho Desk in all 24 helpdesk answers. Canva was in all 24 design answers, WATI, a Hong Kong company, in all 24 WhatsApp answers, and HubSpot in 21 of 24 CRM answers.

3. Prices in dollars

Only 111 of the 288 India-sent answers quoted any price. Among those answering a no-cue question, 25 of 32 quoted dollars only and 3 more quoted both currencies. Among cued answers, 60 of 79 quoted rupees only and 5 quoted dollars only. From the United States, all 14 priced answers were in dollars.

The engines differed in how often they priced anything. Google AI Mode quoted a price in 44 of 72 India-sent answers, Perplexity in 10 of 72. GST, the tax every Indian business invoices, appeared in 8 of 96 no-cue answers and in none of the 48 sent from the United States.

4. Software that is not sold in India

Four products named to Indian questions cannot be bought, or used as intended, by an Indian business. Intuit withdrew QuickBooks from India in 2023. Square and Clover do not process payments in India. Gusto runs payroll only for businesses registered in the United States.⁶ In the five categories where these products compete, 33 of the 40 no-cue answers sent from India named at least one of them. QuickBooks also appeared in 2 answers to questions that named India.

Thirty-five answers sent from India named software that is not sold in India. None said so.

Study 003, Finding 4

5. Location moved the answer only slightly

Every request in this comparison carried an Indian network location, yet the uncued answers looked much like the American ones. For the same 12 no-cue questions, one run from each country, the Indian share was 15% from India (33 of 217 mentions) and 9% from the United States (22 of 236). All four engines followed the pattern. Google AI Mode gave the largest Indian share to no-cue questions, 26% (30 of 117); Gemini the smallest, 8% (10 of 120).

6. Which products were named

India-sent answers named 205 different products, 4.7 per answer. The most named were Indian accounting and billing tools, almost all of them only when the question was cued: Vyapar appeared in 49 answers, none of them to a no-cue question. Zoho carried much of the Indian share. Products sold under the Zoho name made up 43 of the 69 Indian mentions in no-cue answers; without them, the Indian share of no-cue mentions falls to 6% (26 of 401).

05 / 07What it means

For Indian buyers the lesson is short: say where you are. For software companies, the data point to a few practical steps.

1. Indian vendors win on cued questions and lose the rest.In 48 of 96 uncued answers sent from India, no Indian product appeared. Pages that answer the generic question, not only “best X in India”, are where the gap is.
2. Compliance is the strongest cue.GST, PF and ESI, e-way bills and kirana billing turned the answers Indian. Vendors whose product handles these should say so in plain words on pages engines can read.
3. Publish rupee prices.When answers quoted prices to cued questions, they were mostly in rupees. A visible rupee price page gives an engine something to quote.
4. Foreign vendors selling in India face a different contest.Foreign products made up 84% of mentions in uncued answers and 35% in cued ones. Once India is in the question, they compete with Indian tools built for GST and Indian payroll.

06 / 07Limitations

• One day, one sample. All answers were collected on 5 October 2026, two runs per question from India and one from the US. Answers change between runs and over time.
• Three phrasings per category. Each cell of Fig. 1 rests on 8 or 16 answers. Category differences are large, but a different phrasing could move any single category.
• Cues were not uniform. Some cues were stronger than others (a rupee budget, GST, a city). The “Indian cue” row averages them; it does not rank them.
• Named is not recommended. We count mentions. Some products were named to be compared or ruled out, and the first product named is not always the engine’s pick.
• Headquarters is one definition of Indian. Freshworks, JustCall and others were founded in India and are counted as foreign. The founding-country count is in the data files.
• Availability was checked for four products only. Others named may also be hard to buy or use in India; we listed only those with a primary source.
• We observe the answers, not the reasons. We cannot say why an engine did not use the request’s location.

07 / 07Data and reproducibility

The data are published under CC BY 4.0. Version 1.0 corresponds to this publication; corrections will be issued as new versions with a changelog. The analysis script re-creates every number on this page from the answer log.

• questions.csv36 rows: category, cue type, wordingCSV
• answers.csv336 rows, with full answer text and codingCSV
• mentions.csv1,589 rows: one per product per answerCSV
• products-hq.csv214 products: HQ, owner, founding country, source URLCSV
• availability.csv4 products not sold in India, with sourcesCSV
• source-domains.csv831 cited domains, with countsCSV

For questions about the data, or a reanalysis you would like us to check, write to research@citationresearch.com.

@techreport{citationresearch2026study003,
  author      = {{Citation Research}},
  title       = {The dollar problem: what AI recommends to Indian small businesses buying software},
  institution = {Citation Research},
  number      = {Study 003},
  year        = {2026},
  url         = {https://citationresearch.com/research/ai-software-recommendations-indian-small-businesses}
}