Which Sources Does AI Actually Cite for Best-in-India Queries?

Which Sources Actually Get Cited by AI When Indians Ask "What's the Best X in India", and Why the Business Press Is Almost Invisible in the Answer

Melivana | PR Intelligence Series 2026, Report 7 of 8

A primary-data audit of AI answer-engine citations for India-context commercial discovery queries, built on the same query round used across this trilogy (Report 6, "State of AEO in India," and Report 8, "AI Answer Share"). Prepared by the Melivana PR Intelligence unit.

Executive Summary

We built this report expecting to confirm a comfortable thesis: that when AI systems answer questions about Indian companies, they lean on a small, predictable cast of business mastheads, The Economic Times, Moneycontrol, Mint, Business Standard, and that the job of a modern Indian communications team is therefore to win placements in those ten or so outlets. That was the working hypothesis of this series. It is the hypothesis almost every AEO deck in the market repeats.

Then we ran the audit. And the data overturned it.

On 8 July 2026 we posed 20 India-context commercial discovery questions, "best neobanks in India," "top HR SaaS platforms in India," "leading D2C skincare brands in India," and the like, spread evenly across the five sectors this trilogy tracks (Fintech, SaaS, D2C, Healthcare, Manufacturing), to three answer engines with live web search enabled: ChatGPT (gpt-4o), Gemini (2.5-flash) and Perplexity (sonar-pro). Gemini and Perplexity returned cited answers on 100% of the questions. ChatGPT's gpt-4o answered closed-book, surfacing no citations at all. The 40 sourced answers produced 334 citation instances spread across 237 distinct domains.

Here is what those 334 citations actually looked like, and it is not what the industry keeps telling itself.

The headline: the top 10 domains captured just ~18% of all AI citations. The citation landscape for commercial "best X in India" queries is radically fragmented, a genuine long tail of 237 domains, and it is led not by the business press but by social and user-generated content (YouTube, Reddit, LinkedIn, Instagram), by Wikipedia, by brands' own websites, and by SEO listicles and market-research pages. The single most-cited source in the entire study was youtube.com, and it accounted for roughly 3.3% of citations. Reddit tied it. The most-cited news outlet a communications team would normally chase did not meaningfully appear at all.

Traditional Indian business dailies contributed effectively 0% of the citations in this round. The startup-tech press, YourStory, Inc42, Entrackr, contributed effectively 0%. Regional-language outlets contributed effectively 0%. For this specific and enormously common class of query, the commercial recommendation question, the "which one should I pick" question that increasingly starts a buying journey, the outlets that PR spends most of its budget chasing were, in AI's answer, nearly invisible.

This is a harder finding than the one we set out to write, and a far more useful one. It says the concentration story is wrong for commercial discovery, that the target list is not short but sprawling, and that a press-only PR strategy is close to invisible to AI in exactly the moment a consumer is deciding what to buy. Five findings define the 2026 index:

1. Radical fragmentation: the top 10 domains capture only ~18% of citations, 237 distinct domains across 334 citations. There is no oligopoly of outlets; there is a long tail.
2. The single most-cited source is youtube.com, at ~3.3%, tied with reddit.com. No source exceeds a low-single-digit share. The distribution is flat at the top, not spiked.
3. Traditional business and financial press captured ~0% of citations in this round, because commercial "best X" queries pull listicles, community threads and brand pages, not news articles.
4. Regional-language outlets captured ~0%, the vernacular citation gap is total for these queries, a visibility blind spot for Bharat-first brands.
5. Essentially every sector had a different leading domain, there is no stable cross-category leader; the top source in Fintech is not the top source in Healthcare or Manufacturing. Fragmentation shows up across sectors as well as within them.

The strategic payoff is the reverse of the conventional playbook. If AI's answer to a commercial query is assembled from Reddit threads, YouTube reviews, LinkedIn posts, Wikipedia entries, brand-owned pages, app-store listings and SEO listicles, then a strategy that invests only in earned media placements is nearly invisible to the machine at the point of purchase. The playbook has to widen, into Wikipedia presence, structured brand-owned content, review-and-listicle placement, a YouTube and community footprint, and app-store optimisation, while still valuing earned media for the news, crisis and reputational queries where the business press very likely does dominate (a distinction we draw carefully throughout, because our audit only measured commercial discovery).

This report is the middle panel of a trilogy built on one shared query round. Report 6 measures whether Indian brands are answer-ready at all. Report 8 measures the share of the answer each brand wins. Report 7, this one, measures the ground beneath both: which sources AI actually reaches for when it builds its picture of "the best in India." The ground turned out to be very different from the map.

What This Report Measures, and What It Does Not

Before the findings, a precise framing, because precision is what separates an intelligence product from marketing, and because the honesty of this report's scope is exactly what makes its contrarian conclusion trustworthy.

This is the Media Citation Index: an audit of which source domains AI answer engines cite when responding to India-relevant commercial discovery questions. It is not a ranking of brands (that is Report 8), and it is not a readiness diagnostic (that is Report 6). The unit of analysis is the cited domain, youtube.com, reddit.com, pharmeasy.in, ibef.org, not the company being written about.

A "citation," for our purposes, is any instance in which one of the tested engines attributed a claim to, linked, or footnoted a source domain in its answer to one of our prompts. We counted the linked source chips that Gemini and Perplexity expose in their responses. We counted every instance, then also tracked how many distinct domains those instances resolved to, which is how we can speak both to raw citation volume (334) and to breadth (237 domains).

Three scope decisions matter enormously to how you read the results:

We tested commercial discovery queries specifically, the "best/top X in India" archetype. This is deliberate. It is one of the highest-commercial-intent, fastest-growing query types in AI usage: the question a real buyer asks when they are choosing a neobank, an HR platform, a skincare brand, a diagnostics provider or a contract manufacturer. But it is only one archetype. News queries, event queries, crisis and reputational queries almost certainly pull a different, and far more press-heavy, source mix. Nothing in this report should be read as a claim that the business press is irrelevant to AI in general. It is a claim about the specific, load-bearing case of commercial discovery, where our data shows the press is close to absent. We flag this repeatedly and on purpose.

We tested three named engines with search on, on a single date. ChatGPT (gpt-4o), Gemini (2.5-flash) and Perplexity (sonar-pro), all with web search enabled, on 8 July 2026. Crucially, ChatGPT's gpt-4o answered closed-book in this configuration, it produced answers but surfaced no citations, so every one of the 334 citations in this study comes from Gemini and Perplexity, both of which cited on 100% of their answers. That is an important limitation and we treat it as one.

These are observed, tallied counts, not modelled estimates. Unlike a projection, the figures here are direct counts from a real, reproducible round: 334 citation instances, 237 distinct domains, specific per-domain tallies. Where we write "~18%," it is 60 of 334 rounded, not a triangulated guess. The judgment in the exercise is in classification, deciding that pharmeasy.in is a brand-owned domain and mordorintelligence.com is a market-research aggregator, not in the counts themselves. Where classification is arguable, we say so.

That combination, real counts, honestly bounded scope, is what lets this report say something genuinely useful rather than something reassuring.

The Headline in Full: ~18% to the Top 10, and a Flat Peak

Across the 40 sourced answers, the ten most-cited domains, in observed rank order, were:

1. youtube.com, 11 citations (~3.3%)
2. reddit.com, 11 citations (~3.3%)
3. linkedin.com, 8 citations (~2.4%)
4. instagram.com, 5 citations (~1.5%)
5. wikipedia.org, 5 citations (~1.5%)
6. ibef.org, 4 citations (~1.2%)
7. builtin.com, 4 citations (~1.2%)
8. customfit.ai, 4 citations (~1.2%)
9. pharmeasy.in, 4 citations (~1.2%)
10. mordorintelligence.com, 4 citations (~1.2%)

Just below the top ten sit play.google.com (3), mysa.io (3), greythr.com (3), 1mg.com (3) and onsurity.com (3), a cluster of brand-owned and app-store domains that underlines the pattern rather than breaking it.

Summed, the top ten domains account for 60 of 334 citations, roughly 18%. The other ~82% is divided across 227 further domains, the overwhelming majority of which appear once or twice. This is the opposite of the concentrated, oligopolistic picture the industry (and, candidly, the first draft of this report) assumed. It is a long-tail distribution with a strikingly flat peak: no domain breaks 3.5%, the top two are tied, and the drop from first to tenth is a gentle slope from 11 citations to 4, not a cliff.

Why does this matter so much? Because the entire strategic logic of "AEO for PR" has been built on the assumption of concentration, the idea that because AI cites a handful of authoritative outlets, you win AI visibility by placing stories in that handful. Our data says that for commercial discovery, there is no handful. There is a crowd of 237, led by platforms and pages that no press-relations program targets.

Three observations make the shape concrete.

First, the leaders are not publishers, they are platforms and communities. The top four domains are YouTube, Reddit, LinkedIn and Instagram: social and user-generated content, not journalism. When Gemini or Perplexity is asked to name the best neobanks or the best HR software in India, it frequently grounds its answer in a YouTube review, a Reddit thread where users compare options, a LinkedIn post, or an Instagram presence, the places where real people actually discuss and rank products.

Second, the encyclopedic and the commercial sit right alongside them. Wikipedia (5) provides the neutral reference layer. Brand-owned domains, pharmeasy.in, 1mg.com, greythr.com, onsurity.com, mysa.io, provide the primary-source product information. SEO aggregators and market-research pages, builtin.com, customfit.ai, mordorintelligence.com, ibef.org, provide the ready-made "top 10" lists that map perfectly onto a "best X" query. App stores (play.google.com) provide install-level validation.

Third, the business press is simply not on the board. Not in the top ten, not in the cluster just below it, not in any position that would matter to a media list. For this query type, on this date, across these engines, the outlets a PR team lives and dies by did not surface.

That is the headline, and the rest of the report is its anatomy and its implications.

Key Finding 1, Radical Fragmentation: The Top 10 Own Only ~18%

The first and most important finding is structural, and it inverts the assumption this series began with: AI does not concentrate its trust onto a few outlets for commercial discovery queries, it disperses it across a very long tail. 237 distinct domains carried 334 citations. The top ten took ~18%. The median domain in the study was cited exactly once.

For communicators, this is a paradigm break in the opposite direction from the one usually preached. The standard AEO sermon says: the universe has collapsed onto ten citable outlets, so narrow your focus to them. Our data says: for the commercial query, the universe has exploded, the answer can be sourced from any of hundreds of pages, and the ones that win are platforms, communities, brand pages and listicles, not a tidy press list. You cannot buy your way into an 18%-to-the-top-ten distribution with three placements in three newspapers, because the newspapers are not in the distribution.

This fragmentation is rational from the engine's perspective. A commercial recommendation query does not have a single authoritative "news of record" answer the way an earnings report or a policy change does. "What is the best HR SaaS in India" is a matter of comparison, opinion, feature fit and user experience, and the densest, most decision-relevant text for that lives in Reddit threads, YouTube walkthroughs, comparison listicles, and the vendors' own sites. The engine reaches for the pages that actually contain product-level comparison, and those pages are spread across the whole web, not concentrated in a masthead tier.

The practical consequence is stark and specific: the addressable surface for AI visibility on commercial queries is not ten outlets but hundreds of pages of many different types. A modern Indian AEO strategy that models itself on a short press list is, for this query class, aiming at the wrong target entirely. The winnable surface is a portfolio, community presence, review coverage, listicle inclusion, brand-owned depth, encyclopedic presence, not a media list.

There is a corollary that ties directly to Report 8: because the distribution is so flat and so wide, no single placement moves your answer share much, but the portfolio does. Concentration would have made AI visibility a matter of winning a few big placements. Fragmentation makes it a matter of showing up, credibly, across many small surfaces. That is a different discipline, and most PR functions are not yet built for it.

Key Finding 2, The Single Most-Cited Source Is youtube.com, at ~3.3%

The most-cited individual domain in the 2026 round was youtube.com, with 11 citations, roughly 3.3% of the total, tied with reddit.com. Let that sit next to the assumption it replaces. The old thesis had a flagship business daily commanding something like one in seven citations. The real data has a video platform commanding roughly one in thirty, tied with a discussion forum, and no source doing appreciably better.

Two things are remarkable here, and both are strategically load-bearing.

The peak is low. A ~3.3% ceiling means there is no dominant source to win. In a concentrated market, you would identify the 14%-share leader and go earn it. Here, the "leader" holds 3.3%, and the second, third and fourth places (reddit 3.3%, linkedin 2.4%, instagram 1.5%) are all social platforms too. The top of the curve is not a target you capture; it is a texture you participate in. You do not "place a story" on YouTube or Reddit the way you place one in a newspaper, you build a presence, seed genuine content, and earn community mentions over time.

The peak is social and user-generated. The four most-cited domains are all platforms where the content is made by users, creators and communities rather than by newsrooms: YouTube reviews and explainers, Reddit comparison threads, LinkedIn commentary, Instagram brand and creator presence. For a commercial "best X" query, the engine is effectively crowdsourcing, grounding its recommendation in what actual users and creators say, because that is the most decision-relevant material available. Wikipedia (fifth, 5 citations) then supplies the neutral factual scaffold.

The implication for PR is uncomfortable and clarifying at once. The single highest-value citation surface for commercial discovery is not a masthead you can pitch; it is a creator-and-community ecosystem you have to earn your way into. A brand with a rich, credible YouTube footprint (independent reviews, founder explainers, comparison videos), an authentic and non-astroturfed Reddit presence, an active LinkedIn narrative, and a well-maintained Wikipedia entry is far better positioned for AI commercial visibility than a brand with a thick press clip-book and none of the above. That is a reallocation most Indian communications budgets have not made.

Key Finding 3, Traditional Business and Financial Press: ~0% of Citations

The finding that most directly overturns this series' original thesis is this: the traditional Indian business and financial press captured effectively 0% of citations in this round. The Economic Times, Business Standard, Mint, Financial Express, Moneycontrol, the outlets the first draft of this report crowned as the citable core, did not meaningfully surface in the answers to these 20 commercial discovery questions. The startup-tech press (YourStory, Inc42, Entrackr) that we expected to be "category-decisive" for Fintech, SaaS and D2C likewise registered at ~0%.

This is not a knock on those outlets' quality or reach. It is a statement about query to source fit. The reason is mechanical, and understanding it is the key to the whole report:

Commercial "best X" queries pull comparison and recommendation content, and the business press does not primarily produce that. A newspaper writes that a fintech raised a round, that a regulator issued a circular, that a founder resigned. It rarely publishes an evergreen, structured "the 10 best neobanks in India, ranked, with pros and cons" page, and when it does, that page competes against dozens of purpose-built listicles, comparison sites and community threads that are more on-target for the query. The engine, asked for "the best," reaches for the pages that are literally about "the best." Those are SEO listicles (builtin.com, customfit.ai), market-research summaries (mordorintelligence.com, ibef.org), community rankings (reddit.com), creator reviews (youtube.com) and the vendors' own comparison-friendly pages, not news articles.

The business press's core strengths are aimed at a different query. Factual density about events, freshness on breaking developments, authority on markets and policy, these are exactly the properties that make the business press dominant for news and reputational queries. "What happened to Company X," "is Company Y in trouble," "explain the RBI's latest move", for those, we would fully expect the business dailies to reappear near the top, and communicators should not abandon them. But for "which one is best," the strengths do not match the question, and the citations go elsewhere.

The honest, bounded reading is therefore twofold. First: for commercial discovery, a press-only PR strategy is nearly invisible to AI. If your brand's entire AI-visibility bet is earned coverage in the business press, and your buyers are asking AI "what's the best in India," you are close to absent from the answer. Second: this near-zero share is specific to this query type. It is emphatically not evidence that earned media is dead. It is evidence that earned media, alone, does not cover the commercial-discovery surface, and that the surface needs a different, wider set of tools.

That distinction is the difference between a reckless conclusion and a useful one, and we hold it firmly.

Key Finding 4, The Regional-Language Gap: ~0%, and the Vernacular Blind Spot

Regional-language outlets, Dainik Bhaskar, Dainik Jagran, Amar Ujala, Eenadu, Malayala Manorama, Lokmat, Anandabazar Patrika and the rest, captured effectively 0% of the citations in this round. Not a small share: essentially none. And this despite the fact that these very titles command the majority of India's real newspaper readership. Eight of India's ten highest-circulation dailies publish in regional languages, not English. Inside AI's answer to a commercial discovery query, that readership is invisible.

The causes are structural, and for commercial queries they compound with the same forces that keep the English business press out:

The queries themselves were English, and the citable comparison surface is English-first. Even framed around India, "best neobanks in India" is answered from the densest, most-linked comparison content available, which is in English, and which lives on global platforms (YouTube, Reddit, LinkedIn) and English listicles and brand pages. The vernacular press does not, today, produce a large, structured, crawlable body of "best X, ranked" content in Hindi, Telugu or Malayalam for the engine to reach.

Vernacular digital footprints are thinner in exactly the formats these queries reward. Comparison pages, structured listicles, review-style content, well-maintained brand and encyclopedic entries, the vernacular web is under-built on all of them relative to the English web. The properties that got youtube.com, reddit.com and the brand-owned domains cited are the properties the vernacular commercial-content surface has least of.

Why this matters strategically, not just descriptively:

For Bharat and vernacular-first brands, AI commercial discovery is currently a total blind spot. A D2C brand whose customers read Dainik Bhaskar and search in Hindi may be beautifully covered in the vernacular press and still be entirely absent from the AI answer those same customers get when they ask "best skincare brand in India." The coverage reaches the human; it does not reach the machine mediating the human's decision.

But the gap is an open early-mover lane. Because the vernacular commercial-content surface is under-built, it is contestable. Brands that seed structured, crawlable, comparison-friendly content, reviews, community engagement, brand-owned explainers, encyclopedic presence, in and around vernacular audiences, and mirror it in structured English, can capture citation share before the segment matures and before Indic-language model capability (improving fast) makes the vernacular web more citable. The 0% is not a permanent ceiling; it is an unclaimed frontier.

The operational point stands, uncomfortably: to be seen by AI on a commercial query in India today, you must be present in English, on social and community platforms, in listicles and on your own well-structured pages. The vernacular press, for this query type, does not carry you there.

Key Finding 5, No Stable Cross-Category Leader: Essentially Every Sector Differs

Fragmentation is not only a within-query phenomenon; it holds across sectors too. When we break the round into its five sectors, essentially every sector surfaces a different leading domain. There is no single source that leads across categories, the top-cited domain in Fintech is not the top-cited domain in Healthcare, SaaS, D2C or Manufacturing. The absence of a cross-category leader is itself a signature of the fragmented, query-fitted way these engines source.

The pattern of leaders reflects each sector's content ecosystem rather than a common press bench:

• Fintech, grounded heavily in community and social comparison (reddit.com, youtube.com) plus market-research framing (ibef.org, mordorintelligence.com). Buyers debate neobanks and payment apps in threads and videos, and the engine follows them there.
• SaaS, led by SEO aggregators and comparison directories (builtin.com, customfit.ai) alongside brand-owned product pages (greythr.com, mysa.io) and app-store listings (play.google.com). "Best HR/SaaS tool" queries map directly onto software-listing and review sites.
• D2C, pulled toward social and creator platforms (instagram.com, youtube.com) and brand-owned domains, where consumer product discovery actually happens.
• Healthcare, anchored on brand-owned category leaders (pharmeasy.in, 1mg.com, onsurity.com) plus Wikipedia for the neutral reference layer. Health-service comparison is dominated by the providers' own authoritative pages.
• Manufacturing, leans on market-research and industry-body sources (ibef.org, mordorintelligence.com) plus company-owned pages, reflecting a sector where structured market reports substitute for consumer reviews.

The strategic reading is that *there is no universal "citable set" to memorise, the winning source type shifts by sector, from community-and-social in Fintech and D2C, to software-directory-and-app-store in SaaS, to brand-owned-and-encyclopedic in Healthcare, to market-research-and-industry-body in Manufacturing. A Manufacturing brand and a D2C brand face genuinely different citation surfaces and should build genuinely different footprints. The one constant is what is absent* everywhere: the business press and the vernacular press, neither of which led any sector.

This is also the finding that most directly binds this report to its trilogy siblings. Report 6 asks whether your content is answer-ready in your sector. Report 8 asks what share of the sector's AI answer you win. Report 7 tells you which source types to build to move those numbers, and shows that the answer is sector-specific, platform-and-brand-heavy, and press-light.

The Citation Map: What PR Chases vs. What AI Actually Cited

Two tables tell the story. The first is the table this series was designed to produce, the share captured by the five press segments PR teams traditionally target. Populated honestly with the real data, it is a table of near-zeros, and that emptiness is the finding.

Table 1, The Press Segments PR Usually Targets (commercial discovery queries)

These near-zero shares are specific to commercial "best/top X in India" discovery queries, measured across Gemini, Perplexity and ChatGPT on 8 July 2026. They are not a claim that these outlets are unimportant to AI in general. For news, event, crisis and reputational queries, we would expect the business and general press to reappear strongly, that is where their factual density, freshness and authority fit the question. This report measured commercial discovery only, and in that arena the press did not surface.

The second table is the one the data actually wrote, the source types that dominated the answers. This is where the 334 citations really went.

Table 2, The Source Types That Actually Dominated (334 citations, 237 domains)

Shares are directional groupings of the observed 334 citations; classification of individual domains into types involves judgment (e.g., a brand blog vs. a brand product page), and the long-tail row absorbs the ~227 domains cited once or twice. The point is the ranking of types, which is robust: social/UGC, brand-owned and SEO/market-research together carried the clear majority of citations, while the traditional press carried essentially none.

Read side by side, the two tables are the whole report in miniature. Table 1 is the map PR has been navigating by. Table 2 is the territory AI actually traversed. They barely overlap.

Data and Methodology

This report is a primary-data audit, and we describe the method in full because the credibility of a contrarian finding depends entirely on the transparency of how it was produced.

The query round. The foundation is a standardised set of 20 prompts, the commercial-discovery slice of the same query round used across Report 6 ("State of AEO in India") and Report 8 ("AI Answer Share"). The 20 questions are distributed evenly across the five sectors this trilogy tracks, Fintech, SaaS, D2C, Healthcare and Manufacturing (four questions each), and all share the high-commercial-intent "best/top X in India" archetype: the recommendation question a real buyer asks at the point of choosing a product or provider. Running one shared round three ways is what makes the trilogy internally consistent: the readiness diagnostic (6), the source-citation index (7) and the brand answer-share measure (8) all interrogate the same underlying reality from different angles.

Engines, configuration and date. The round was executed on 8 July 2026 across three answer engines, each with web search enabled: ChatGPT (gpt-4o), Gemini (2.5-flash) and Perplexity (sonar-pro). This gives 60 answers in total (20 questions × 3 engines). Two engines returned cited answers on 100% of their questions: Gemini and Perplexity. ChatGPT's gpt-4o answered closed-book in this configuration, it generated answers but exposed no source citations, and therefore contributed zero citations to the tally. All 334 citation instances in this study come from the 40 sourced answers produced by Gemini and Perplexity.

The tally. For each of the 40 sourced answers, we recorded every cited domain instance. Aggregated, this produced 334 citation instances resolving to 237 distinct domains. We then counted per-domain frequency (yielding the top-10 ranking and the ~18% top-10 share), and classified domains into source types (social/UGC, brand-owned, SEO aggregator/market-research, encyclopedic, app store, long tail) to produce Table 2. Counts are literal; type classification involves documented judgment.

What is a count vs. a judgment. The core numbers, 334 citations, 237 domains, the per-domain tallies (youtube.com 11, reddit.com 11, and so on), the ~18% top-10 share, and youtube.com's ~3.3%, are direct counts, reproducible by re-running the round and re-tallying. The source-type shares in Table 2 involve classification judgment and are presented as directional. The "~0%" press shares are observed absences: those outlets did not surface in the tally.

Why this posture. We deliberately did not model or extrapolate the headline figures. A real, dated, reproducible round with transparent counts is more defensible, and more useful, than a triangulated estimate, precisely because the finding is contrarian. The reader can, in principle, re-run these 20 questions on these engines and check the shape. What the reader must not do is over-generalise from it, which is why the limitations below are extensive and emphatic.

Why AI Sourced This Way, The Mechanics of the Long Tail

The ~18%-to-the-top-ten shape and its social-and-brand tilt are not random. They are the predictable output of how retrieval-augmented answer engines source a commercial recommendation query. Six forces drive the pattern.

1. Query to content fit dominates. For "best X" queries, the engine reaches for pages that are literally about ranking and comparing X. Those are listicles, directories, community threads and creator reviews, not news articles. This single force explains most of the distribution: the business press is absent not because it lacks authority but because it rarely produces "best X, ranked" content on demand.

2. Community and creator content is decision-relevant. Real buyers compare products in Reddit threads and YouTube reviews. That text is dense with the exact features, complaints and preferences a recommendation query needs, so the engine grounds its answer there. UGC wins because it is about the decision.

3. Brand-owned pages are primary sources. For factual product claims, pricing, features, coverage, the vendor's own site is the authoritative source, and the engine cites it directly (pharmeasy.in, greythr.com, onsurity.com). This is why brand-owned domains rank so highly for commercial queries and barely at all for news queries.

4. SEO and market-research content is purpose-built to be cited. Sites like builtin.com, customfit.ai and mordorintelligence.com exist to publish structured, comparison-shaped, list-formatted content optimised for exactly these questions. They are, almost by design, the ideal extractable source for a "best X" answer.

5. Fragmentation is inherent to the query, not the market. Because hundreds of pages plausibly answer a comparison query, and none is the definitive "record," citations spread thin. There is no equivalent of "the newspaper that broke the story" to concentrate onto. The long tail is the natural shape of a question with many valid, partial answers.

6. Encyclopedic and app-store anchors provide validation. Wikipedia supplies neutral facts; app stores supply install-and-rating validation. These are low-share but structurally reliable anchors the engine returns to across sectors.

Put together, these forces explain the whole pattern: the fragmentation (many pages fit), the social-and-brand leadership (that is where decision-relevant comparison lives), and the near-absence of the press (news content does not fit a "best X" question). The same mechanics predict the inverse for news queries, where the press, being the record, would concentrate the citations.

Strategic Implications for PR and AEO, The Real Payoff

If AI's answer to a commercial discovery query is fed by Reddit, YouTube, LinkedIn, Wikipedia, brand-owned pages and SEO listicles, then the strategic conclusion is unavoidable: a press-only PR strategy is nearly invisible to AI at the point of purchase. That is not an argument against earned media; it is an argument that earned media alone does not cover the commercial-discovery surface. The playbook has to widen. Six shifts follow directly from the data.

Build a Wikipedia and encyclopedic presence. Wikipedia was the fifth most-cited domain and appeared across sectors as the neutral factual anchor. A well-sourced, accurate, notability-compliant Wikipedia entry is one of the highest-leverage, most durable AI-visibility assets a brand can hold, and one most Indian brands neglect. It is not PR; it is reference-layer hygiene, and it pays off in citations.

Engineer structured, citable brand-owned content. Brand-owned domains were roughly a fifth of all citations. Your own site is not just a brochure; it is a primary source the engine quotes. Build structured, factual, comparison-friendly pages, clear feature tables, pricing, "how we compare," dated updates, schema markup, because those are exactly the pages pharmeasy.in, greythr.com and onsurity.com won citations with. This is the single most controllable lever you have.

Get into the listicles, directories and reviews. SEO aggregators and market-research pages (builtin.com, customfit.ai, mordorintelligence.com, ibef.org) were ~18% of citations. Inclusion in the "top 10 best X in India" listicles, the software directories, the review sites and the relevant market reports is now a distribution channel, not a vanity mention. Treat placement in these, through outreach, product-listing programs and analyst relations, as a core AEO activity.

Build a YouTube and community footprint. The top four domains were social/UGC. That means credible creator reviews, founder explainers and comparison videos on YouTube; an authentic, non-astroturfed presence in the Reddit and community threads where your category is debated; an active LinkedIn narrative; and, for consumer brands, Instagram. This is earned and owned social, not press, and it is where the highest-share citations actually live. Astroturfing is both unethical and self-defeating here, the engines (and communities) surface authentic discussion; manufactured threads get discounted and can backfire.

Optimise your app-store presence. For Fintech and SaaS especially, play.google.com surfaced as a cited source. App-store listings, ratings and descriptions are an AI-visibility surface. ASO is now partly an AEO activity.

Keep earned media, but aim it at the right queries. None of this retires the business press. For news, funding, crisis and reputational queries, "what happened to X," "is Y trustworthy," "explain Z's regulatory situation", the business and general press very likely dominate the citation set, because there the press is the record. The correct posture is a two-track strategy: community, brand-owned, listicle, encyclopedic and app-store presence for commercial discovery; earned media for news and reputation. Most Indian communications functions run only the second track. The data says they are missing the first entirely, and that the first is where buyers are asking "which one is best."

The meta-implication: AEO for PR is not "get cited by the same outlets, but for AI." It is a broader discipline that spans owned, earned, social, community and reference surfaces, sequenced to the query types that matter for your brand.

The AEO Playbook for Commercial Discovery

A practical, sequenced playbook to convert the index into AI visibility for "best/top X" queries. Use it alongside Report 6 (are you answer-ready?) and Report 8 (what share do you win?).

Step 1, Run your own commercial-query audit. Take the 8 to 12 "best/top X in India" questions a real buyer would ask in your category, and pose them to Gemini, Perplexity and ChatGPT with search on. Record every cited domain. This is your baseline citation surface, and it will look nothing like your media list.

Step 2, Map the source types, not the outlets. Classify what got cited: how much is social/UGC, brand-owned, listicle/market-research, encyclopedic, app store. That distribution is your target portfolio. In SaaS it will skew to directories and app stores; in D2C to social and brand-owned; in Healthcare to brand-owned and Wikipedia; in Manufacturing to market-research and industry bodies.

Step 3, Harden your brand-owned surface first. It is the most controllable ~20% of the citation pool. Build structured, factual, comparison-friendly, schema-marked pages; keep them fresh and crawlable. This is table stakes and it is entirely within your control.

Step 4, Establish encyclopedic presence. Create or improve a well-sourced, notability-compliant Wikipedia entry and keep it accurate. It is durable, cross-sector and disproportionately cited.

Step 5, Win the listicles and directories. Identify the "top 10 best X in India" pages, the software directories, the review sites and the market reports that AI actually cited for your category, and pursue inclusion through outreach, listings and analyst relations.

Step 6, Build authentic social and community presence. Seed genuine YouTube reviews and explainers; engage honestly in Reddit and community threads; maintain a substantive LinkedIn and (for consumer brands) Instagram footprint. Earn mentions; do not fabricate them.

Step 7, Optimise app-store listings where relevant (Fintech, SaaS), treating ratings and descriptions as an AEO surface.

Step 8, Keep a parallel earned-media track for news and reputation, where the business press dominates, and re-run both audits quarterly. The engines re-index constantly; citability is earned continuously, and the commercial surface shifts fast.

How to Use This Benchmark

This index is a strategic instrument, not a scoreboard to memorise. Five ways to put it to work:

As a myth-buster. Use the ~18%-to-the-top-ten headline and the two vivid facts (youtube.com and reddit.com lead; the business press is ~0%) to break the room's assumption that AI visibility equals press placements. It is the fastest way to reframe an AEO investment conversation honestly.

As a portfolio blueprint. Use Table 2 to rebuild your AI-visibility plan around source types, social, brand-owned, listicle, encyclopedic, app store, rather than a media list. If your entire plan is earned media, you have found your gap.

As a sector diagnostic. Use Finding 5 to tune the portfolio to your sector's actual citation surface. A D2C brand and a Manufacturing brand should invest in visibly different source types, and this report tells you which.

As the middle panel of the trilogy. Read this report between Report 6 and Report 8. Six tells you if you are answer-ready; Seven tells you which source types AI actually cites for commercial queries; Eight tells you what share of the answer you win. Together they move from diagnosis to targeting to outcome, all on one shared query round.

As a baseline to beat. These are the counts as of 8 July 2026. Re-run the round for your own brand and category and track the delta over time. The value is in the movement, not the snapshot.

Limitations

Intellectual honesty is what makes a contrarian finding credible, so we state the constraints plainly and in full. These limitations do not weaken the conclusion; they define its scope precisely.

Bounded sample. The audit covers 20 questions, 60 answers, 40 of them sourced, on a single date (8 July 2026), with one model per engine. 334 citations is a directional dataset, not a census of AI behaviour. The shape it reveals is striking and reproducible in principle, but the exact percentages would move on a re-run.

Commercial discovery only. Every question was a "best/top X in India" commercial recommendation query. This is the report's single most important boundary. The near-zero press shares are specific to this query type. News queries, event queries, funding queries, crisis and reputational queries almost certainly show a very different, far more press-heavy source mix, and communicators should assume the business and general press do dominate there. Do not generalise the ~0% press finding beyond commercial discovery.

ChatGPT/gpt-4o did not browse. In this configuration, gpt-4o answered closed-book and contributed zero citations. Every citation here comes from Gemini (2.5-flash) and Perplexity (sonar-pro). A different ChatGPT configuration (a browsing-enabled model) would add citations and could shift the distribution. The findings describe the citing behaviour of two engines, not all three.

Single engine-versions and a single date. Answer engines are non-deterministic, personalised, geographically variable and continuously re-indexed. Model versions change. Two users, two locations or two dates can see different citations for the same prompt. This is a dated snapshot, not a permanent map.

Classification judgment. The per-domain counts are literal, but sorting 237 domains into source types (Table 2) involves judgment at the margins, a brand blog versus a brand product page, an industry body versus a market-research firm. The type-share figures are directional; the type ranking is robust.

Directional, not census. 334 citations is enough to reveal a clear long-tail shape and a clear source-type ranking. It is not enough to publish decimal-precise market shares, and we have not. Treat "~18%," "~3.3%" and the Table 2 shares as observed central figures within honest bands, and re-measure rather than quote them as permanent truth.

Appendix A, Glossary

• AEO (Answer Engine Optimisation): Optimising content and presence to be selected, cited and surfaced by AI answer engines. Broader, this report argues, than earned media alone.
• GEO (Generative Engine Optimisation): Broadly synonymous with AEO; optimising for visibility inside generative AI outputs.
• Citation / citation instance: An instance where an AI answer attributes a claim to, links, or footnotes a source domain. This round had 334.
• Distinct domains: The number of unique source domains across all citations, 237 here, a measure of breadth and fragmentation.
• Top-10 share: The proportion of all citations captured by the ten most-cited domains, ~18% in this round.
• Source type: The category of a cited domain, social/UGC, brand-owned, SEO aggregator/market-research, encyclopedic, app store, long tail.
• Commercial discovery query: A high-intent recommendation question ("best/top X in India") asked when choosing a product or provider. The exclusive focus of this report.
• Long tail / fragmentation: A distribution where no source holds a large share and hundreds of domains each hold a little, the defining shape of these results.
• Closed-book answer: An AI answer generated without live retrieval or surfaced citations, as ChatGPT/gpt-4o produced here.
• Brand-owned domain: A vendor's own website, cited by the engine as a primary source for product facts.

Appendix B, The Source Types and Representative Domains

• Social / UGC: youtube.com, reddit.com, linkedin.com, instagram.com, creator reviews, community comparison threads, professional and brand social presence.
• Brand-owned: pharmeasy.in, 1mg.com, greythr.com, onsurity.com, mysa.io, vendors' own product, pricing and comparison pages.
• SEO aggregators & market-research: builtin.com, customfit.ai, mordorintelligence.com, ibef.org, listicles, software directories, market reports.
• Encyclopedic / reference: wikipedia.org, the neutral factual scaffold.
• App stores: play.google.com, install-level listings and ratings.
• Notably near-absent (for commercial queries): national business dailies (ET, Business Standard, Mint, Financial Express), startup-tech pubs (YourStory, Inc42, Entrackr), general news (TOI, HT, NDTV, The Hindu, Indian Express), trade titles, and all regional-language outlets.

Appendix C, The Trilogy at a Glance

• Report 6, State of AEO in India: Are Indian brands answer-ready? Diagnostic of AEO maturity across the five sectors.
• Report 7, India Media Citation Index (this report): Which sources does AI actually cite for commercial discovery queries? The source-level map of AI's grounding, and the finding that it is fragmented, social-and-brand-led, and press-light.
• Report 8, AI Answer Share: What share of the AI answer does each brand win? Brand-level outcome measure.

All three are built on the same 2026 query round across Fintech, SaaS, D2C, Healthcare and Manufacturing. Report 7's slice was the 20 commercial "best/top X in India" questions, executed across ChatGPT (gpt-4o), Gemini (2.5-flash) and Perplexity (sonar-pro) with web search on, on 8 July 2026. Read together, the three move from readiness to targeting to outcome.

Melivana | PR Intelligence Series 2026, Report 7 of 8. Prepared by the Melivana PR Intelligence unit. Figures are observed counts from a real, reproducible AI query round: 334 citation instances across 237 distinct domains, from 40 sourced answers, gathered on 8 July 2026 across Gemini, Perplexity and ChatGPT. They are directional strategic guidance for commercial discovery queries, not an audited census and not a claim about AI behaviour for news, crisis or reputational queries. For methodology questions or a category-specific citation audit, contact the Melivana PR Intelligence unit.