While reviewing recent AI Search experiments from Indian practitioners, we came across a study by Growgence examining where ChatGPT and Perplexity source information when answering commercial recommendation questions — the kind of “what’s the best X” query that used to send someone to a review site.
The original finding was striking. Independent review media accounted for around 27% of citations in US-focused answers, compared with only 2% in India-focused answers. SEO and affiliate-style blogs, meanwhile, appeared at similar rates in both markets.
That raised a question worth exploring: if review publications were largely absent from the India-focused answers, which sources were AI systems using instead?
We reached out to Growgence for the methodology and the underlying breakdown. The team shared the results by market, category and AI engine, along with the limitations of the experiment. The additional data made the original finding more nuanced — and more useful for SEOs — than the headline number alone.
How the experiment worked
Growgence selected six commercial categories: skincare and personal care, eyewear, exam preparation, payments and payroll, supplements and protein, and wireless audio.
The team used buyer-style questions such as “What are the best…” and “Which are the best…”, keeping the India and US versions structurally similar. Exam preparation was the exception, since the underlying exams needed localisation — JEE and NEET for India, SAT and AP for the US.
- Tested 25 August 2026 — ChatGPT (gpt-4.1-mini) and Perplexity (sonar), web search enabled on both
- Each category tested once per market, per engine — 6 × 2 × 2 = 24 AI answers
- Those 24 answers produced 334 citations: 172 for India-focused questions, 162 for US-focused questions
- Domains classified into 8 buckets: independent review media, SEO/affiliate blogs, brand-owned sites, marketplaces & retailers, Reddit, LinkedIn, Indian earned media, and reference sources
The experiment therefore wasn’t measuring which brands ranked highest. It was examining the information sources sitting behind AI-generated buying answers — a different, and for SEOs a more useful, question.
The biggest difference wasn’t SEO blogs
The first useful finding appears when the entire citation mix is compared. Independent review media represented 27.2% of US citations and only 2.3% of India citations — a substantial gap. But SEO and affiliate blogs accounted for 44.8% of India citations and 41.4% of US citations: broadly similar. That’s not where the story is.
The source mix diverged far more strongly around brand and retail properties. Brand-owned sites accounted for 26.2% of India citations versus 18.5% in the US, while marketplaces accounted for 14.5% versus 3.7%. Together, brand-owned sites and marketplaces represented 40.7% of India citations, compared with 22.2% in the US sample.
Brand websites and marketplace listings weren’t just destinations after discovery. In this experiment, they were frequently part of the source material behind discovery itself.
The experiment doesn’t establish that brand and marketplace sources directly replaced review publications. It does show that the information mix supporting India-focused recommendations was materially different — and that difference, not the review-media number alone, is what an SEO team can actually act on.
The review-media gap was concentrated in two categories
The overall 27.2% US review-media share hides an important detail. Of the 44 US review-media citations in the whole study, 22 came from wireless audio and another 14 from eyewear. Those two categories alone accounted for 81.8% of every US review-media citation Growgence recorded.
Wireless audio
Eyewear
This suggests the aggregate India-versus-US comparison shouldn’t be read as a universal market rule. The gap was particularly strong in categories where the US has an established specialist review ecosystem — audio gear and eyewear both have decades of dedicated review publications behind them. A broad GEO benchmark might tell a team what happens across several industries while hiding the source behaviour of the one industry they actually work in.
There was no single India sourcing pattern
Looking only at the India data reinforces the point. SEO/blog content dominated payments, exam preparation and skincare. Brand-owned sources dominated eyewear and supplements. Wireless audio split almost evenly between marketplaces and SEO/blog content.
| Category | Dominant India source |
|---|---|
| Payments & payroll | SEO / blog |
| Exam preparation | SEO / blog |
| Skincare & personal care | SEO / blog |
| Eyewear | Brand-owned |
| Supplements & protein | Brand-owned |
| Wireless audio | Marketplace / SEO split |
The practical implication: “What sources does AI use in India?” is too broad a question for an SEO strategy. A better research question is which sources AI systems use when answering the commercial questions in your specific category. An eyewear brand following the overall India average might over-invest in external blog outreach, since blogs accounted for 44.8% of the full dataset — yet within eyewear specifically, 60% of citations came from brand-owned sites. Category-level evidence changes the strategy.
ChatGPT and Perplexity produced different source mixes
Growgence found another difference at the engine level. Perplexity produced 120 of the 172 India citations; ChatGPT produced 52. Perplexity generally returned close to 20 sources per answer, while ChatGPT returned fewer — so raw citation counts between the two shouldn’t be compared as though both systems expose sources the same way. The composition within each engine is still informative.
Marketplace, LinkedIn, earned media and reference citations that Growgence didn’t break out individually per engine are grouped under “Other sources.”
Among ChatGPT’s India citations, brand-owned sites, marketplaces and SEO/blog sources each appeared 16 times, with no Reddit or independent review-media citations at all. Perplexity’s citations leaned heavily on SEO/blog content (61) and brand-owned sites (29), and were the source of every single Reddit and review-media citation on the India side.
With six India-focused answers per engine, this isn’t evidence that ChatGPT avoids Reddit or review media. It’s evidence that the engine you choose for an AI-visibility study materially changes what you observe.
For SEO teams, combining ChatGPT, Perplexity and other AI systems into a single “AI visibility score” risks hiding exactly these differences in how each system retrieves and surfaces sources.
What SEOs should take from this
The most useful learning here is methodological, before it’s tactical. AI Search research should move past checking whether a brand appears in a response — the sources behind the response usually say more about where visibility is coming from, and where to investigate next.
-
Trace competitor recommendations back to their sources
If a competitor keeps getting recommended, check what’s supporting it. A recommendation built on the competitor’s own product pages is a different problem from one built on a marketplace listing, an independent review, an affiliate comparison, or a Reddit thread — and each calls for a different response.
-
Run the analysis at category level, not account level
The six categories in this study produced materially different source patterns. Generic GEO advice risks pointing a team toward source types that have little actual influence within their own commercial queries.
-
Treat brand-owned pages as retrieval surfaces, not just conversion pages
Brand sites made up 26.2% of the full India citation pool, and more than half of citations in eyewear and supplements. Clear product descriptions, specifications, comparisons, use cases and factual consistency aren’t only an on-site UX concern — they’re part of the information environment retrieval systems encounter.
-
Audit marketplace listings the same way
Marketplaces represented 14.5% of India citations overall, and 43.3% within wireless audio alone. Attributes, descriptions, specs, categorisation and consistency with the brand’s own site belong in an AI Search audit for any ecommerce brand.
-
Don’t treat Reddit as a shortcut
Reddit represented 5.8% of India citations and 5.6% of US citations — almost identical shares, and on the India side, entirely through Perplexity. The evidence here doesn’t support Reddit as a universal AI-visibility lever.
-
Treat this as a snapshot, not a benchmark
Growgence tested each market-category-engine combination once. That captures a snapshot of sourcing behaviour, not its stability. A stronger measurement programme repeats the important commercial prompts, logs every citation, classifies the sources, and tracks which domains and source types persist over time.
The conclusion
Growgence’s original experiment started as a simple India-versus-US comparison. The deeper breakdown leads to a more useful one: in this 25 August 2026 snapshot, India-focused buying answers contained far fewer independent review-media citations and substantially more brand-owned and marketplace citations than the US sample — but the difference wasn’t consistent across categories, and ChatGPT and Perplexity didn’t expose the same source patterns.
For SEOs, the lesson isn’t to chase one source type. It’s to map the information ecosystem around the commercial questions that matter to the business — which source types appear, which domains recur, where competitors get their support, how the pattern shifts by AI engine, and whether any of it survives a repeated test. Only then does the conversation move from generic GEO advice to evidence-led AI Search strategy.