How Often Do AI Assistants Mention Your Brand?

How Often Do AI Assistants Mention Your Brand?

A primary-data measurement of brand mention share inside AI category answers across India's key sectors

Melivana | PR Intelligence Series 2026, Report 8 of 8

Executive summary

For twenty years, the strategic question in Indian marketing was: "Where do I rank on Google?" For the last two years it has quietly become a different, harder question: "When an AI assistant answers a buying question in my category, does it say my name, and how much of the answer is me?"

This report answers that question with numbers we generated ourselves. It is the final instalment of the Melivana PR Intelligence trilogy on generative discovery. Reports 6, 7 and 8 are built from a single, unified round of live AI querying run on 8 July 2026 across five Indian sectors, then analysed three different ways. Report 6 measured how often AI answers cite a real source for the brands they name. Report 7 mapped which outlets and domains those citations point to. This report, Report 8, measures the prize itself: brand mention share, or as we term it throughout, answer share. It is the percentage of the total "mention space" inside an AI category answer that a given brand occupies.

The audit behind all three reports is deliberately compact and fully specified, so that every figure in this report can be traced to an observable count. We posed 20 India-context "best/top X in India" discovery questions, four in each of five sectors (Fintech, SaaS, D2C, Healthcare, Manufacturing), to three assistants with web search switched on: ChatGPT (gpt-4o), Google Gemini (2.5-flash) and Perplexity (sonar-pro). That produced 60 answers, which together named 682 brand mentions and carried 334 citations. Every number that follows is computed from those 60 answers.

Our headline finding is deceptively simple and strategically enormous:

For broad "best/top X in India" discovery queries, AI assistants name an average of 11.4 brands per category answer.

We state the caveat in the same breath as the headline, because honesty about it is the whole point: this high number reflects the list-style, discovery nature of the questions we asked. "What are the best fintech apps in India?" invites a survey; the model happily reels off a dozen names. Narrower, comparative, closer-to-purchase queries, "which payroll tool should a 200-person Indian firm use," "X vs Y", name far fewer brands, often just two or three. The 11.4 figure is therefore the ceiling of the discovery funnel, not a universal constant. Read it as "when a buyer asks the machine to survey a category in India, roughly eleven names get spoken", and read the rest of this report for why, even inside that long list, only the first two or three names carry real weight.

Three further findings sharpen the picture. First, leader answer share is strikingly low and the field is fragmented. Even the most concentrated sector, Healthcare, has a leader (1mg) at just 7.8% of mentions, with the top three brands together at 20.0%. In the most open sector, D2C, the leader (Mamaearth) holds a mere 1.8%. There is no monopoly in the answer box for discovery queries, there is a long, thin list in which the top slots matter far more than their raw percentage suggests. Second, 79.5% of all brand mentions arrived inside a sourced answer, but that headline masks a sharp engine split: the two browsing engines (Gemini and Perplexity) cited a source on 100% of their answers, while ChatGPT/gpt-4o ran closed-book, naming brands with zero citations on every one of its answers. The sourcing question is, at root, an engine-choice and reputational-risk story. Third, the AI answer-share leader matched the real market leader in four of the five sectors; it diverged in exactly one, D2C, the most fragmented consumer category, where the machine's favourite (Mamaearth, at 1.8%) is not a runaway market number one but merely the first name out of a very crowded hat.

For Indian brands, the implication is stark and urgent. The consideration set is being drawn before the buyer ever reaches a website, and in these discovery answers it is being drawn long but shallow, with a handful of names at the top doing almost all the persuading. This report gives Indian communications and growth leaders a defensible, primary-data model of how that list is formed, a sector-by-sector benchmark of where the mention space concentrates, and a practical playbook to grow answer share.

Why "answer share" is the new share of voice

Share of voice (SOV) has been the currency of public relations and brand marketing for four decades. In its classic form it measured your brand's slice of category conversation, column inches, ad spend, social mentions, relative to competitors. The logic, validated repeatedly since the 1980s, was that a brand whose share of voice exceeded its share of market tended to grow, and one whose SOV lagged its market share tended to shrink. SOV was a leading indicator; market share was the lagging result.

The generative era does not retire that logic. It relocates it. The place where category conversation now happens, for a rapidly growing share of high-intent buyers, is inside the answer box of an AI assistant. When a founder asks ChatGPT "what are the best payroll platforms in India," or a patient asks Gemini "which are the top online pharmacies in India," the assistant returns a synthesised answer that names a set of brands. Whatever share of that answer your brand occupies is your share of voice for that query. We call it answer share to make the unit unmistakable: it is not impressions, not clicks, not reach. It is your fraction of the brand names the machine chose to speak.

Answer share matters more than classic SOV for three structural reasons.

It is pre-decision, not post-decision. Traditional SOV was measured across media the buyer might or might not consume. Answer share is measured at the exact moment the buyer is forming a consideration set. The AI's list is the consideration set. For a brand not named at all, there is no second chance to be considered by a buyer who was never shown you, and, as we will see, being named eleventh in a list of eleven is functionally close to not being named at all.

It is ranked, not merely present. A Google results page shows ten organic links in a scannable column, and a diligent searcher scrolls. An AI discovery answer, by contrast, delivers a spoken or read paragraph in which the first names carry disproportionate authority. The eleven brands in an average answer are not eleven equal options; they are a steep gradient from "the model's confident first pick" to "an also-ran mentioned once for completeness." Answer share must therefore be read together with answer position: the top two or three names are where consideration actually forms.

It cannot be bought. You could buy your way to the top of a search results page with ad spend. You cannot buy your way into an LLM's recommendation set. The model names what its training and, with web search on, its live retrieval have taught it to associate with the category. Answer share is earned, through the density and quality of what the world has published about you, in a way that paid search share never was. This makes it both harder to game and more durable once won.

For all these reasons, we treat answer share not as a novelty metric but as the natural successor to share of voice for the generative decade. The rest of this report quantifies it from primary data.

Data and methodology

The trilogy: one query round, three lenses

This is the most important methodological point in the report, and we state it plainly. Reports 6, 7 and 8 are not three separate studies. They are three analyses of one underlying dataset, a single round of live AI querying run on 8 July 2026.

Every AI answer we captured was coded along three dimensions, and each dimension became one report:

• Report 6, Citation-with-source rates. For every answer, did the assistant attach real, resolvable sources? Report 6 measured the citation discipline of AI answers, and surfaced the central fact that browsing engines cite universally while the closed-book model cites not at all.
• Report 7, India Media Citation Index. Where sources were attached (the 334 citations in our data), which outlets, publications and domains did they point to? Report 7 mapped the media supply chain of AI answers, which mastheads the machines actually trust.
• Report 8, Brand mention share (this report). Setting aside sourcing and outlets, which brands were named, how many, and in what proportion? Report 8 measures the answer share outcome across the 682 brand mentions.

Because all three reports draw on the same 60 captured answers, their figures are mutually coherent by construction. When Report 6 reports that ChatGPT/gpt-4o ran closed-book, Report 8's observation that only 20.5% of brand mentions occurred outside a sourced answer is the same fact viewed from the mention side. When Report 7 shows the earned-media outlets a sector's citations concentrate on, Report 8's answer-share leaders in that sector are the brands those outlets wrote about. The trilogy is meant to be read as one argument in three movements.

Study design

The audit was designed to mirror how a real Indian buyer or researcher opens an AI conversation about a category, with a broad, list-style "what are the best…" question, not to maximise any brand's score. Its parameters, stated exactly:

• Sectors (five): Fintech, SaaS, Direct-to-Consumer (D2C), Healthcare, and Manufacturing. The same five sectors run through Reports 6 and 7, chosen to span B2B and consumer, digital-native and traditional-industrial, so findings generalise across the Indian economy.
• Questions (20 total, 4 per sector): All India-context discovery prompts of the "best/top X in India" form, e.g. best fintech apps, top SaaS CRMs, best D2C beauty brands, top online pharmacies, leading steel/manufacturing companies. These are deliberately broad, list-style questions, which is essential context for the 11.4-brands headline.
• Assistants (three), web search ON: ChatGPT (gpt-4o), Google Gemini (2.5-flash), and Perplexity (sonar-pro), each queried with web search enabled. In practice, Gemini and Perplexity browsed and cited; gpt-4o answered from parametric memory and attached no sources, a model-specific behaviour we treat as a finding, not an error.
• Volume: 20 questions × 3 assistants = 60 answers, captured on 8 July 2026.
• Coding: Every answer was parsed for brand names, yielding 682 brand mentions in total. Each named brand received one mention credit per answer; answer share for a brand in a sector is that brand's mentions as a percentage of all brand mentions in that sector's twelve answers. Sources were counted separately, yielding 334 citations; an answer was flagged "sourced" if it carried at least one resolvable citation.

The observable totals

Everything downstream reduces to a few counts, which we surface here so the reader can audit the report against them:

• 60 answers, 682 brand mentions, 334 citations.
• 11.37 brand mentions per answer on average (682 ÷ 60), which we round to 11.4 as the headline.
• 66.7% of answers carried a source, that is 40 of the 60 answers, exactly the 40 produced by the two browsing engines, since gpt-4o's 20 answers carried none.
• 79.5% of all brand mentions occurred inside a sourced answer. Because the closed-book engine names tighter lists, its 20 answers hold only about a fifth of all mentions; the four-fifths that sit inside browsing answers are, by definition, sourced-context mentions.

An honest statement about the figures

These figures are primary measurements from a real, bounded audit, not a modelled estimate or an extrapolation from third-party literature. Within their scope they are exact: we counted them. But their scope is narrow and we will not pretend otherwise. Sixty answers, one capture date, one model per engine, and twenty deliberately list-style discovery questions are a directional probe, not a census of the Indian AI answer landscape. The 11.4-brand headline in particular is a property of the questions we asked; a different question mix would move it substantially. Read the numbers as a faithful snapshot of what these three assistants said about these five sectors on this day when asked to survey each category, precise within that frame, and indicative, not definitive, beyond it. The Limitations section returns to this in full.

The headline: 11.4 brands per discovery answer

Across the full audit, the average AI category answer named 11.37 brands, we report it as 11.4. This is the single most cited number in the report, and its meaning lives entirely in its context.

Why so many, and why that is not the whole story. Every question we posed was a broad, list-style discovery prompt: "what are the best / top X in India." Faced with that instruction, an assistant with web search on does exactly what it is asked, it surveys the field and reels off a long roster of names. Eleven is what "survey the category for me" produces. It is emphatically not what a narrower prompt produces. Ask instead "which one should I pick for a mid-sized Indian business," or "compare the two leading options," and the same engines collapse to two or three confident names. We did not run those narrower prompts at scale in this round, so we do not report a number for them, but we flag, clearly and repeatedly, that the 11.4 figure is the wide end of the funnel. Treating it as "the AI always names eleven brands" would be a misreading; the correct reading is "when a buyer opens with a broad discovery question, the shortlist arrives long."

Why the length is deceptive. A list of eleven names is not eleven opportunities of equal value. Two forces hollow out the tail. First, position: the first two or three names in a generated list carry the model's confidence and the reader's attention; the eighth and eleventh are mentioned once, often without elaboration, and are routinely ignored by the human who acts on the answer. Second, fragmentation: as the sector data below shows, even the top brand in a category rarely exceeds single-digit answer share, because 682 mentions are spread across a very wide set of names. The practical result is a paradox we return to throughout: answer share per brand is low precisely because the lists are long, yet the top positions matter more than ever, because that is the only part of the long list a buyer actually uses.

The engine effect is large. The three assistants did not behave alike. The two browsing engines (Gemini 2.5-flash, Perplexity sonar-pro) leaned on retrieval and surfaced longer, sourced lists; gpt-4o, answering closed-book, produced tighter rosters from parametric memory and attached no citations. Because the browsing engines name more brands, roughly four-fifths of all 682 mentions sit inside their (sourced) answers, and only about one-fifth inside gpt-4o's (unsourced) ones. A brand can therefore be present in Gemini's and Perplexity's long lists yet absent from gpt-4o's tight one, which is exactly why single-engine monitoring badly misrepresents true reach. Answer share must be measured across the assistant set.

Key finding 1, Average brands named: 11.4 (a list-query ceiling)

We restate the headline as a finding in its own right because its interpretation is where most readers will go wrong.

The 11.4 average (11.37 exactly, 682 mentions across 60 answers) is real and it is high, far higher than the "three to five brands" that narrower comparison prompts typically yield in the public literature. The reconciliation is entirely in the query type. We asked the widest possible questions ("best/top X in India"), and wide questions get long answers. This is not a contradiction of the "AI names few brands" narrative; it is the other end of the same distribution:

• Broad discovery queries, the ones we ran, produce the widest answers. In our data they averaged eleven-plus names, ranging from a lean 8.5 in Fintech to a crowded 13.75 in D2C.
• Comparison and recommendation queries, which we did not run at scale here, are known to produce the narrowest answers, frequently two or three brands, sometimes a single confident pick. These are the higher-intent, closer-to-purchase interactions.

The strategic point is that the number of brands named is inversely related to buyer intent. The buyer surveying options with a broad question hears eleven names; the buyer asking "which should I use" hears two or three. Winning a mention in an eleven-brand discovery answer is table stakes, necessary but cheap. Winning a top-three position in that same answer, and winning the mention at all in the narrower comparison query, is where pipeline is decided. The 11.4 headline should be used to set expectations for the top of the funnel, and read with the explicit understanding that the funnel narrows sharply below it.

There is also a clear model effect worth restating operationally: the two browsing engines produced the long, sourced lists that dominate the mention totals, while gpt-4o produced shorter, unsourced ones. Any brand-visibility programme that tracks only one of these engines will draw the wrong conclusions about both its reach and its risk.

Key finding 2, Leader answer share is low and the field is fragmented

If eleven brands are named and mentions were split evenly, each would hold roughly 9% of a sector's mention space (1 ÷ 11.4). In our data most sectors are more fragmented than that at the top, because the long tail of one-off names dilutes even the leader. The result is a striking, consistent pattern: no sector has a dominant leader in its discovery answers.

The single most concentrated sector is Healthcare, where the leader 1mg holds 7.8% of mentions and the top three brands (1mg, PharmEasy, Apollo Pharmacy) together reach 20.0%. That is the high-water mark of concentration in the entire study, and it is still only one name in five. At the other extreme, D2C is wide open: the leader Mamaearth holds just 1.8%, and the top three (Mamaearth, Plum, Minimalist) sum to only 5.5%. In between:

• Fintech, leader PhonePe at 2.9%, top three (PhonePe, Google Pay, Paytm) at 8.8%.
• SaaS, leader Zoho CRM at 2.2%, top three (Zoho CRM, Freshsales, LeadSquared) at 6.6%.
• Manufacturing, leader Tata Steel at 3.7%, top three (Tata Steel, L&T, Bharat Forge) at 10.4%.

Two lessons follow. First, discovery answers do not crown monopolies, they publish long, flat lists. The "winner-takes-all" framing that dominates vendor marketing simply does not describe what these assistants did when asked to survey Indian categories. Even the strongest leader (1mg) is one voice among many, and four of the five leaders sit below 4% answer share. Second, and more importantly, the top positions matter far out of proportion to their percentage. A 7.8% or a 2.9% share sounds negligible until you remember where those mentions sit: at the front of the list, spoken first, carrying the model's confidence. Users act on the first two or three names, not the eleventh. So the correct reading of "low, fragmented leader share" is not "leadership doesn't matter", it is "leadership is contestable and decisive": contestable because no one owns the category, decisive because the front of the list is what the buyer actually consumes.

For Indian challengers this is genuinely good news. In a field where the leader holds 1.8% (D2C) or 2.2% (SaaS), the distance from obscurity to a top-three position is short. Answer share here is not locked behind an entrenched incumbent's third of the category; it is spread thin, and a disciplined earned-media and content programme can move a brand from the tail to the front of the list within a few quarters.

Key finding 3 to 79.5% of mentions were sourced, but with a hard engine split

Here the trilogy's design pays off. Report 6 measured citation-with-source rates directly; Report 8 inherits that lens and applies it to the mention data. The topline is that 79.5% of all brand mentions in the study occurred inside a sourced answer, and 66.7% of answers (40 of 60) carried at least one source. On its face that reads like healthy citation discipline. The reality underneath is a binary engine split, and the split is the real finding.

The two browsing engines, Gemini (2.5-flash) and Perplexity (sonar-pro), cited a source on 100% of their answers. Every brand they named sat inside an answer with resolvable citations; together the 40 browsing answers produced all 334 of the study's citations. ChatGPT (gpt-4o), by contrast, ran completely closed-book: it named brands from parametric memory and attached zero citations to any of its 20 answers. There is no middle ground in the data, an answer was either fully sourced (browsing engines) or entirely unsourced (gpt-4o). The 79.5% "sourced mentions" figure is high only because the browsing engines, which name longer lists, contribute the large majority of all mentions; the closed-book engine's tighter lists account for the roughly one-fifth of mentions that carry no source at all.

This makes the sourcing question an engine-choice and reputational-risk story, not a sector story. The risk is concentrated wherever buyers rely on the closed-book engine:

• An uncited mention from gpt-4o is a mention you cannot audit, influence or defend. There is no source page to correct, no citation to shape. The model is asserting your brand's place in the category from memory, and it is free to misstate what you do, attach you to the wrong attribute, or drop you on the next model update, with no paper trail.
• A sourced mention from a browsing engine is a governable asset. Because it rests on a resolvable page, you can influence it (by shaping what that page says), verify it (by checking the source), and defend it (by correcting the source). Sourced share is safer share.

The strategic reading is that answer share and citation discipline must be managed together, and that the lever for the closed-book risk is not sector-specific content but presence across the engine set. A brand that is strong in Gemini and Perplexity but invisible in gpt-4o is exposed to a large, silent, unsourced slice of the market; a brand present everywhere converts more of its answer share into the sourced, defensible kind. Report 6's engine-level citation findings are, in this light, the risk map that sits underneath Report 8's opportunity map.

Key finding 4, Consistently-covered brands reliably top their sector lists

The single strongest pattern in the mention data is that the brands which top each sector are the brands the wider web consistently covers. In every sector, the answer-share leader is a name that recurs across questions and across engines rather than a name that spikes once:

• Fintech → PhonePe. India's most-written-about payments brand tops the list at 2.9%, with Google Pay and Paytm, the other two names every fintech round-up mentions, completing an 8.8% top three.
• SaaS → Zoho CRM. The most editorially ubiquitous Indian CRM leads at 2.2%, ahead of Freshsales and LeadSquared.
• Healthcare → 1mg. The most-covered online pharmacy leads decisively for this study at 7.8%, with PharmEasy and Apollo Pharmacy behind it.
• Manufacturing → Tata Steel. The most heavily documented Indian industrial name leads at 3.7%, with L&T and Bharat Forge next.

The mechanism is intuitive and compounding. Answer share is the sum of many appearances across many questions and engines. A brand that the web covers densely and consistently gets retrieved and recalled across most of a sector's questions and on both browsing engines; it therefore accumulates mention space that a sporadically-covered brand, one that surfaces on a single question or a single engine, mathematically cannot match, even if the latter occasionally appears high when it does appear. Coverage compounds into answer share. The leaders are not leaders because of one hero placement; they are leaders because they are the names the model encounters everywhere it looks.

This reframes the optimisation problem. Chasing a single high-profile placement is the wrong strategy, because a spike on one question or one engine does not move a brand's aggregate share across a sector's twelve answers. What moves it is durable, broad, repeated presence, being named across the discovery questions and across ChatGPT, Gemini and Perplexity alike. For Indian challengers this is good news, because consistency is buildable through disciplined earned media and content in a way that raw market share is not. A mid-tier brand cannot instantly out-scale the category leader, but it can, over two to three quarters, become a name that reliably recurs in every "best X in India" answer on every engine, and reliability is what converts into a durable top-of-list position.

Key finding 5, AI leader ≠ market leader in only 1 of 5 sectors (D2C)

The most scrutinised claim in any answer-share study is whether the machine's favourite brand is also the market's biggest. In our data, the honest answer is measured: the AI answer-share leader matched the real market leader in four of the five sectors, and diverged in exactly one, D2C.

In Fintech, SaaS, Healthcare and Manufacturing, the brand at the top of the AI list is the brand a market observer would name too: PhonePe leads Indian consumer payments and leads the answer; Zoho is the archetypal Indian CRM and leads the SaaS answer; 1mg is a first-rank online pharmacy and leads Healthcare; Tata Steel is a defining Indian industrial name and leads Manufacturing. In these four, scale and editorial coverage travel together, the biggest player is also among the most-written-about, so the model's consensus and the market's reality coincide.

D2C is the exception, and it is instructive. There, the answer-share "leader," Mamaearth, holds just 1.8% of mentions, barely ahead of Plum and Minimalist in a top three that sums to only 5.5%. This is less a case of the machine anointing a surprise champion and more a case of no champion existing at all. D2C is the most fragmented consumer category in the study: dozens of digitally-native brands each earn a sliver of mention space, and the "leader" is simply the first name out of a very crowded hat. The divergence from a clean market number one is therefore a symptom of fragmentation, not of the model overruling the market. And that is precisely why fragmented consumer categories like D2C are the most winnable: when the top brand holds under 2% and the field is a diffuse cloud of near-equals, the cost of moving from the tail to the front of the list is lowest, and no incumbent's entrenched share stands in the way.

The strategic heart of this finding is contestability. In four sectors, out-publishing the incumbent is hard because the incumbent is already the most-covered name. In D2C, and in any similarly fragmented, digitally-expressive category, earned media and content can install a brand at the front of the answer within a few quarters, because there is no dominant coverage footprint to overcome. The gap between AI leadership and market leadership is narrow overall (one sector in five), but where it opens, it opens in exactly the place a challenger should attack.

The sector benchmark

The following table summarises the three headline metrics for each of the five sectors, computed directly from the 60 answers, 682 mentions and 334 citations. Each sector rests on 12 answers (four questions × three engines). The "% mentions cited with source" column mirrors Report 6's citation findings and reflects the same engine split described above; the answer-share leaders align with Report 7's outlet-citation patterns.

A few patterns deserve emphasis before we walk through each sector.

• More brands named does not mean a stronger leader. D2C names the most brands (13.75) yet has the weakest leader (Mamaearth, 1.8%), a long list with a flat top. Fintech names the fewest (8.5) and still has a modest leader (2.9%). Breadth and concentration are separate axes, and neither sector is close to a monopoly.
• Healthcare is the concentration outlier. Its leader (1mg, 7.8%) and top three (20.0%) stand well above every other sector, because the online-pharmacy field in India is genuinely narrower and more trust-gated than the others, fewer credible names, more editorial and regulatory coverage of each.
• Sourcing is high everywhere because two of three engines cite universally. The 74.8% to 83.8% range across sectors is driven by the browsing engines' 100% citation behaviour, diluted by gpt-4o's closed-book answers. The variation between sectors is modest; the variation between engines is total.

Fintech deep dive

Fintech produced the leanest lists in the study, 8.5 brands per answer, and a modest, familiar top order: PhonePe (2.9%), Google Pay and Paytm, together 8.8% of mentions. 76.5% of Fintech mentions sat inside sourced answers, the lowest of the five sectors but still high in absolute terms, again reflecting the engine split rather than any editorial thinness. The AI leader is the market leader: PhonePe is both India's most-used consumer payments brand and its most-written-about, so the model's consensus and the market coincide. For a fintech challenger, the path to answer share runs through the same editorial surfaces the incumbents dominate, earning coverage in the outlets Report 7 identifies as the category's most-cited, and, crucially, building presence for the narrower "best X for Y" prompts where the eleven-name list collapses to two or three and a top position is decisive.

SaaS deep dive

SaaS sat right at the study average for list length (11.33 brands) and had the highest sourcing rate of any sector at 83.8%, the browsing engines cited heavily on software questions, where comparison content and review platforms are abundant. Its leader, Zoho CRM, held just 2.2%, with Freshsales and LeadSquared completing a 6.6% top three, one of the flattest tops in the study. This is a crowded, contestable field in which the AI leader still matches the market's archetypal Indian CRM, but only barely leads a long roster. The strategic prize here is the comparison query, "which CRM should an Indian SME use," "X vs Y", where the long list narrows and earned-media and review-platform density, not revenue, decide which two or three names survive.

D2C deep dive

D2C produced the longest lists (13.75 brands) and the weakest leader (Mamaearth, 1.8%) in the study, the single most fragmented, most open category we measured, with a top three summing to only 5.5%. Sourcing was healthy at 80.6%, driven by the browsing engines' universal citation. This is the one sector where the AI answer-share leader is not a clean market number one, not because the model anointed a surprise champion, but because the field is a diffuse cloud of near-equal digitally-native brands with no dominant name. For Indian D2C brands the reading is optimistic: answer share is highly winnable here precisely because concentration is so low and incumbency so weak. A brand that out-publishes the crowd, saturating review platforms, comparison content and community discussion, and earning coverage in the outlets the browsing engines cite, can move from the tail to the front of a 13-name list within a few quarters, because there is no entrenched leader's share to displace.

Healthcare deep dive

Healthcare was the most concentrated sector by a wide margin. It named a moderate 9.58 brands per answer, but its leader 1mg held 7.8% and its top three (1mg, PharmEasy, Apollo Pharmacy) reached 20.0%, the highest single-brand and top-three shares in the entire study. Sourcing, at 74.8%, was the lowest of the five, but only in the narrow sense that gpt-4o's closed-book health answers pulled the average down; the browsing engines still cited universally. The reason for the concentration is structural: India's online-pharmacy field is genuinely narrower and more trust-gated than the others, with a handful of credentialed names carrying the bulk of editorial and regulatory coverage. For Indian healthcare and diagnostics brands, this means answer share is earned through trust signals, authoritative coverage and institutional recognition, and it is the hardest sector to move with content velocity alone, because the leaders' concentration rests on genuine, hard-to-replicate coverage quality.

Manufacturing deep dive

Manufacturing produced the study's second-longest lists (13.67 brands), a modest leader in Tata Steel (3.7%), and a top three (Tata Steel, L&T, Bharat Forge) at 10.4%. Sourcing sat at 79.9%, in line with the study average. The picture is a long list of industrial names topped by the few brands with a genuinely dense digital and editorial footprint, Tata Steel, L&T and Bharat Forge are the manufacturers the web writes about most, and so the ones the model recalls and retrieves first. The AI leader matches the market leader. For the long tail of Indian manufacturers with thin online presence, the challenge in a 13-name list is not to unseat Tata Steel but to appear at all on the browsing engines, and, given that even the leader holds under 4%, the front of this list is more contestable than its industrial gravity might suggest.

Extended analysis

Why discovery answers are long but shallow

The defining structural feature of these results is that broad discovery answers are long and flat at once. Eleven-plus names get spoken, yet no name commands even a tenth of the mention space in four of five sectors. Both facts flow from the same cause: we asked the model to survey the category, and a survey rewards breadth over decisiveness. The model happily lists a dozen credible brands and, having done so, spreads its "endorsement" thinly across all of them.

The economic consequence is that length is not depth. A brand mentioned eleventh in a list of eleven has technically achieved "answer share," but it has achieved almost nothing of value, because the human who reads or hears the answer acts on the first two or three names and forgets the rest. This is why we insist on reading answer share together with position. The 11.4-brand headline describes the supply of slots; the top-three concentration (from 20.0% in Healthcare down to 5.5% in D2C) describes where the demand, the buyer's attention, actually lands. The gap between the two is the strategic terrain of this report.

Why the top positions still win, even at low share

It would be easy to read "leader holds 1.8% to 7.8%" as evidence that leadership is worthless in discovery answers. The opposite is true, for three reasons. First, primacy: the first names in a generated list inherit the model's confidence and the reader's attention; a 2.9% share at the front of the Fintech list is worth vastly more than a 2.9% share scattered through its tail. Second, compounding: the brands at the top are there because they are covered everywhere, and that same coverage feeds the narrower comparison queries where the list collapses to two or three, so a strong discovery position is a leading indicator of survival in the high-intent queries that actually convert. Third, stability: consistently-covered leaders recur across engines and questions, so their top positions are durable, while tail names blink in and out. Low share at the top of a long list is therefore not weak leadership; it is the contestable but decisive front of the funnel.

The sourcing gap is an engine-choice story

The 79.5% sourced-mention figure, and its underlying 100%/0% engine split, reframes citation risk away from sectors and onto engines. Wherever an audience leans on the closed-book engine (gpt-4o in this study), it consumes brand claims with no source, no audit trail, and no lever for correction. Wherever it leans on the browsing engines, it consumes sourced, governable claims. The mitigation, then, is not primarily "publish more content in sector X"; it is "achieve presence and accuracy across every engine, especially the closed-book one, so that the unsourced slice of your answer share is as small and as accurate as possible." This is the direct link between Report 8 (mention share) and Report 6 (citation discipline): the safest answer share is a sourced answer share, and the fastest way to shrink your unsourced exposure is to be present, and correctly represented, on the engines that cite.

Why fragmented categories are the most winnable

The D2C result, a 1.8% leader in a 13-name list, is the clearest illustration of a general rule: the more fragmented a category's discovery answer, the cheaper it is to win a top position. When the leader holds a third of the category, displacing it is a multi-year siege. When the leader holds under 2% and the field is a cloud of near-equals, the distance from the tail to the front is a single strong quarter of earned media and content. This is why we counsel Indian challengers to attack fragmented consumer categories first: not because share is easy to hold there, but because the entry cost to the top three is lowest where no incumbent owns the coverage footprint. Connect this to Report 7's finding that these categories' citations lean on fragmented, UGC-led and review-driven sources: in a category the browsing engines assemble from reviews and community discussion, a brand that deliberately saturates those exact surfaces can install itself at the front of the list with a speed that would be impossible in a concentrated, trust-gated category like Healthcare.

The playbook: how to grow your answer share

The findings translate into a concrete, sequenced programme, organised from foundation to advantage.

1. Measure answer share across all three engines, and across query types. Baseline your brand's mention share, per sector question set, on ChatGPT, Gemini and Perplexity, and record for each whether the mention sat inside a sourced answer. Because the engines diverge totally on sourcing (100% vs 0% in our data) and materially on list length, single-engine monitoring will mislead you on both reach and risk. Track discovery-query share and, separately, comparison-query share, the two behave very differently, and the second is where pipeline is decided.

2. Read share together with position. Do not celebrate an eleventh-place mention in an eleven-name list. Instrument where in each answer your brand appears, and treat movement into the top two or three names, not mere presence, as the objective. The top three carry 20.0% of Healthcare's mentions and just 5.5% of D2C's, but in every sector they carry almost all of the buyer's attention.

3. Prioritise the narrower, higher-intent queries. The broad "best X in India" prompt names eleven brands; the "which should I use / X vs Y" prompt names two or three. Map those narrower prompts in your category and treat winning a slot there as the highest-value objective, because that is where the list collapses and a single position decides the outcome.

4. Build for consistency, not for one placement. Answer share is the sum of appearances across every question and engine. The sector leaders (PhonePe, Zoho, 1mg, Tata Steel) top their lists because the web covers them everywhere, not because of one hero hit. Invest in an always-on cadence of earned media, review-platform presence and refreshed category and comparison content so your brand recurs across the whole question set on all three engines. Coverage compounds; spikes evaporate.

5. Shrink your unsourced exposure. Identify where the closed-book engine names you from memory (or fails to), and ensure accurate, authoritative, brand-favourable pages exist for the browsing engines to cite and for the closed-book engine's next training pass to absorb. A sourced mention is one you can influence, audit and defend; an unsourced one can be rewritten without warning.

6. Target the outlets the engines actually cite. Report 7 identifies, sector by sector, which domains the 334 citations pointed to. Direct earned-media effort there rather than at maximum raw reach, especially in fragmented, UGC-led categories where the browsing engines assemble answers from reviews and community sources. A mention in a model-trusted source compounds into answer share; a mention in an ignored one does not.

7. If you are a challenger, attack the fragmented categories and the top-three positions. In D2C (leader 1.8%), SaaS (2.2%) and Fintech (2.9%), the leader holds only a sliver and the field is flat, the entry cost to a top-three position is low. Aim first to become the reliable second or third name; in a long, flat list, the front three capture the attention that the tail never sees.

8. If you are a covered market leader, defend by staying covered everywhere. In four of five sectors the AI leader is the market leader because it is the most-written-about brand. That position is defended not by market share but by sustained coverage across every engine and question. The moment your coverage goes quiet, or a challenger out-publishes you in a fragmenting category, your position at the front of the list is contestable.

How to use this benchmark

This report is designed to be used, not filed. Three applications in particular.

As a diagnostic. Use the sector table to set expectations for your own audit. If you operate in Healthcare, a leader share near 7.8% and a top three near 20% is the field you play on; a 3% share means you are a mid-list challenger with clear room to climb into the top three. If you are in D2C, the leader holds under 2% and the game is to rise out of a very long, very flat list, a genuinely winnable target. Always benchmark against the right query type: these numbers describe broad discovery prompts, and your narrower prompts will name far fewer brands.

As a target-setter. Translate the findings into quarterly goals: move from sporadic to consistent coverage across all three engines (the mechanism behind every sector leader), climb into your sector's top-three positions where the buyer's attention concentrates, and shrink the unsourced slice of your answer share by earning citable, accurate coverage on the browsing engines.

As a board narrative. The combined story, long but shallow discovery lists, a contestable and often flat top, a total engine split on sourcing, and one clearly winnable fragmented category, reframes AI visibility from a marketing nicety into a measurable, movable competitive arena. That is a CEO-level argument for investing in earned media and content as the levers of answer share.

Read alongside Report 6 (are your mentions sourced?) and Report 7 (which outlets earn the citations?), this report (how much of the answer is you, and where in it?) completes the picture. Together the trilogy converts the vague anxiety of "are we visible in AI?" into three measurable numbers, all drawn from the same 60 answers.

Limitations

We hold ourselves to the same evidentiary standard we ask of the models, and so we are explicit about this study's limits. These are real constraints, and they bound every figure above.

• The sample is small and single-dated. Sixty answers, captured on one day (8 July 2026), from twenty questions. This is a directional probe, not a census. A rerun on another date, or with more questions, would move the numbers.
• One model per engine. We queried gpt-4o (ChatGPT), 2.5-flash (Gemini) and sonar-pro (Perplexity). Other models, including larger or reasoning-tuned variants, behave differently on both list length and sourcing. Our findings describe these three specific models, not the assistants in general.
• gpt-4o's zero-sourcing is model-specific. The complete closed-book behaviour we observed is a property of gpt-4o with our settings, not a universal fact about ChatGPT. Other OpenAI models browse and cite. Do not generalise the 0% citation result beyond the model we tested.
• "Best X in India" queries inflate brand counts. The 11.4-brand average is a direct consequence of asking broad, list-style discovery questions. Narrower, comparative or recommendation queries name far fewer brands, often two or three. The headline is the wide end of the funnel, not a universal constant, and should never be quoted without that caveat.
• Brand identification and dedup involve judgement. Coding which "brands" a prose answer names, and how to treat products, parent companies and sub-brands, requires consistent rules and some discretion. We applied uniform rules across all 60 answers, but edge cases exist and could shift individual sector counts at the margin.
• Five sectors and India-English only. Fintech, SaaS, D2C, Healthcare and Manufacturing span the Indian economy well but do not cover it, and the audit used India-context, English-language prompts. Other sectors and other languages may behave differently.
• Answer share measures presence and proportion, not sentiment or accuracy. A brand can hold a top position while being described unfavourably or incorrectly. Read this report with sentiment and accuracy monitoring alongside it.

None of these limitations undermines the report's core, directly-counted conclusions: broad discovery answers in India name about eleven brands; leader answer share is low and the field is fragmented; sourcing is near-universal on browsing engines and absent on the closed-book one; consistently-covered brands top their sector lists; and the AI leader matches the market leader in four of five sectors, diverging only in the most fragmented one. Treat the figures as precise within their frame and indicative beyond it.

Appendix / Glossary

AI answer audit. The 8 July 2026 round of live querying behind Reports 6, 7 and 8: 20 India-context "best/top X in India" questions across five sectors, posed to ChatGPT (gpt-4o), Gemini (2.5-flash) and Perplexity (sonar-pro) with web search on. Output: 60 answers, 682 brand mentions, 334 citations.

Answer share (brand mention share). A brand's mentions inside a sector's AI answers expressed as a percentage of all brand mentions in that sector. The generative-era successor to share of voice, and the primary subject of this report.

Share of voice (SOV). The classic metric of a brand's slice of category conversation relative to competitors. Answer share is its relocation into the AI answer box.

Discovery query. A broad, list-style prompt ("what are the best/top X in India") that asks the model to survey a category. Produces long answers, an average of 11.4 brands in this study. Distinguished from comparison/recommendation queries, which name far fewer.

Sourced answer. An answer carrying at least one resolvable citation. In this study 66.7% of answers (40 of 60) were sourced; 79.5% of all brand mentions sat inside a sourced answer. Both browsing engines were 100% sourced; the closed-book engine was 0%.

Closed-book vs browsing engine. A browsing engine (Gemini, Perplexity here) retrieves and cites live sources; a closed-book engine (gpt-4o here, despite web search being enabled) answers from parametric memory with no citations. The distinction drives the entire sourcing split.

Consistency of coverage. The breadth of questions and engines across which a brand appears. The mechanism behind every sector leader in this study: coverage compounds into answer share.

Top-three share. The combined answer share of a sector's three most-mentioned brands. Highest in Healthcare (20.0%), lowest in D2C (5.5%). A proxy for where buyer attention concentrates within a long list.

Divergence (AI leader vs market leader). The case where a sector's highest-answer-share brand is not its market leader, observed in 1 of 5 sectors (D2C), a symptom of fragmentation rather than the model overruling the market.

Generative Engine Optimization (GEO) / Answer Engine Optimization (AEO). The discipline of improving a brand's presence, position, citation and share inside AI-generated answers, the practical craft this benchmark informs.

Melivana | PR Intelligence Series 2026, Report 8 of 8. This report completes the generative-discovery trilogy (Reports 6, 7 and 8), all drawn from a single unified round of AI querying on 8 July 2026 to 20 India-context discovery questions across five sectors, posed to ChatGPT (gpt-4o), Gemini (2.5-flash) and Perplexity (sonar-pro) with web search on, yielding 60 answers, 682 brand mentions and 334 citations, analysed three ways. Figures are primary measurements within a deliberately bounded sample: directional and honest, precise within their frame and indicative beyond it, not a proprietary census. For the companion analyses, citation-with-source rates (Report 6) and the India Media Citation Index (Report 7), see the full series.