How a dialysis chain used Scale-tier ChatGPT Ads with 3-language conversational-response support
The situation
Picture a dialysis chain operating a network of centres across a state with a genuinely diverse linguistic population — Hindi-comfortable families in some catchment areas, English-comfortable families in urban pockets, and a substantial regional-language-comfortable population elsewhere across the network. The chain had strong clinical infrastructure — trained nephrology nursing staff, modern dialysis machines, consistent slot availability that many competing centres in the region struggled to match — but its digital presence ran entirely in English, a language choice that made sense for the chain's website copy but badly mismatched how a meaningful share of its actual prospective patient families searched and researched.
Chronic kidney disease is a slow-moving diagnosis for most patients — stage 3 or 4 CKD often precedes dialysis initiation by months or years, giving families a real research window before an urgent decision point arrives. But when that window closes and dialysis becomes necessary, families move fast, and they move in whatever language lets them think most clearly under stress. The chain's leadership had noticed, informally, that a disproportionate share of its highest-quality enquiries — families who showed up already understanding their CKD stage, their nephrologist's referral, and roughly what dialysis involves — were coming through word-of-mouth referral in regional-language-speaking communities, while its English-only digital marketing was pulling in a comparatively shallower, more exploratory audience.
The chain's ambition for this engagement was explicit: build a campaign that met CKD families in Hindi, English and the dominant regional language of its service area, tracked and optimised independently by language rather than folding everything into one blended English-first campaign that happened to occasionally surface in another language through generic translation. That ambition, combined with the chain's multi-centre network and the need to route leads to the correct centre based on catchment and language, put this account at Scale tier.
Campaign structure
ICG built the account around three parallel language buckets — Hindi, English and the regional language — each running its own conversation-stage bid ladder rather than one ladder translated three ways. This distinction mattered because conversation patterns genuinely differed by language, not just by vocabulary: English-language conversations skewed toward more exploratory, comparison-driven queries from younger family members researching on behalf of an older relative, while Hindi and regional-language conversations skewed toward more direct, urgency-marked queries closer to the point of dialysis initiation, often phrased by the patient or an immediate caregiver rather than a research-minded younger relative.
Within each language bucket, the bid ladder weighted toward CKD-stage and urgency signals consistent across languages in substance if not in phrasing — conversations referencing a nephrologist's dialysis recommendation, a specific CKD stage, or "starting dialysis soon" language sat at the top of the ladder, 7-9x baseline, while general "what is dialysis" or "how does kidney failure treatment work" conversations, still valuable for long-window CKD-stage-3 families but far from an immediate booking, sat in the middle tier. Conversations focused purely on cost or insurance questions without any stage or urgency signal sat lowest, since this segment converted unpredictably and needed lower-cost bidding to stay efficient at Scale-tier budget.
Landing infrastructure ran language-specific pages rather than one page with a language toggle, since a toggle-based approach tested poorly in ICG's prior multilingual healthcare accounts — families arriving from a conversational assistant in their chosen language expect the landing experience to continue in that language without an extra click. Each language's page covered the same core information — centre locations and catchment coverage, typical dialysis-session scheduling, staff qualifications, and slot-availability framing kept deliberately conservative for compliance reasons described below — written natively for that language's register rather than machine-translated from an English master, since CKD families reading urgent medical information deserved copy that read naturally rather than mechanically.
Enquiry forms captured language and nearest-centre preference at submission, routing leads directly to that centre's intake coordinator, several of whom were themselves fluent in the language the family had used, closing the loop between the language the family researched in and the language they'd be spoken to in in that critical initial phone call. Reporting ran monthly with language as a primary segmentation dimension alongside the standard centre and bucket dimensions, letting the chain's leadership see clearly which language was converting best in which catchment area rather than reading one blended number that would have obscured which language investment was actually working.
Compliance discipline
NMC Section 6 and ASCI Chapter III governed every ad variant across all three languages, with two risks specific to this category drawing particular scrutiny. First, slot-availability and wait-time language: dialysis centres operate on genuinely constrained capacity, and any implication of guaranteed immediate availability, in any language, risked both a compliance flag and a real harm to a family making an urgent decision based on an inaccurate claim. Every variant across all three languages was reviewed to ensure availability language stayed to "contact us to check current slot availability" framing rather than any stronger guarantee.
Second, outcome and life-expectancy framing: dialysis is a life-sustaining treatment, not a cure, and CKD families researching this topic are often at an emotionally vulnerable point where overstated outcome language could cause real harm. No ad variant in any language implied guaranteed life-expectancy outcomes, transplant-eligibility improvement, or comparative superiority against other named or implied centres. This standard was enforced identically across languages — critically, each language's copy was independently reviewed by a compliance-trained fluent reader rather than assuming a cleared English variant's translated counterpart automatically passed, since compliance risk can hide in connotation and register differences a literal translation review would miss.
DPDP 2023 governed all enquiry forms uniformly across languages, with consent language covering how contact details and CKD-stage information volunteered in conversation would be used and stored, translated with the same independent-review discipline applied to the ad copy rather than machine-translated. Across the full 90-day run and all three languages, every ad variant cleared compliance review before launch, with zero post-launch flags in any language.
90-day outcome pattern
In the pattern ICG has observed on comparable Scale-tier multilingual healthcare accounts, month one showed a clear staggered start across the three language buckets because Hindi and English launched simultaneously while the regional-language bucket, requiring additional native-speaker copy review time, launched roughly two weeks later. Illustrative combined volume across Hindi and English in month one: 35-45 conversation-completion events a week; the regional-language bucket, once live, ran at a lower 8-12 events a week in its first two weeks, reflecting both its later start and the normal ramp period any new bucket needs.
By month two, all three buckets were running at full calibration. Combined weekly volume across all three languages reached an illustrative 70-85 conversation-completion events, and — the number the chain's leadership cared about most — the regional-language bucket's conversion rate from conversation to booked centre visit began outperforming both Hindi and English, reaching an illustrative 30-34% conversion within 10 days compared to 22-26% for Hindi and 18-22% for English, consistent with the chain's earlier informal observation that its regional-language-comfortable audience tended to arrive further along in genuine decision-readiness.
By month three, combined weekly volume across all three languages reached an illustrative 100-120 conversation-completion events, and the chain reallocated a portion of budget from the lower-converting English bucket toward the regional-language bucket mid-quarter, a decision the language-segmented reporting made straightforward to justify to leadership in a way a blended report never could have. Cost per qualified centre-visit booking, blended across all three languages, ran illustratively 20-30% below the chain's prior English-only digital spend, with the regional-language bucket specifically running the most cost-efficient of the three by month three.
In GA4, all three language segments attributed to the AI Assistant channel, tracked as separate audience segments within that channel. The chain's centre-level intake data showed a downstream pattern consistent across all three languages but strongest in the regional-language segment: a higher share of bookings from this campaign proceeded to actual dialysis initiation at the chain's centres, rather than a first visit followed by the family choosing a different provider, compared to the chain's historical conversion rate from first visit to ongoing patient — read by the chain's leadership as evidence that language-matched outreach was reaching families who trusted the chain more fully by the time they walked in.
What we'd do differently
The staggered launch — Hindi and English live from week one, the regional language delayed roughly two weeks for copy review — cost the regional-language bucket real calibration time it didn't need to lose, given that its underlying demand and conversion quality turned out to be the strongest of the three once live. Building the native-language copy review process to run in parallel with, rather than after, the Hindi and English launch would have let all three buckets start calibrating simultaneously and likely pulled the month-two performance divergence forward by several weeks.
The mid-quarter budget reallocation from English toward the regional-language bucket, while directionally correct, was sized conservatively — leadership approved a modest shift rather than the larger reallocation the data supported, a caution ICG understood given the chain's unfamiliarity with the language-segmented data before this engagement, but a more assertive reallocation earlier in month three likely would have compounded the regional-language bucket's stronger unit economics into a larger overall efficiency gain for the quarter.
Finally, routing leads to language-fluent intake coordinators worked well but wasn't formalised until several weeks into the campaign — early regional-language leads were occasionally handled by whichever coordinator was available rather than a language-matched one, a gap that likely cost some conversion quality in exactly the early weeks when the campaign most needed strong first impressions to build momentum.
How this maps to your own vertical
If your practice or chain serves a linguistically diverse patient population — common across most of India regardless of specialty — building genuinely native-language conversation buckets rather than one dominant-language campaign with occasional translated variants is likely to surface exactly the kind of underserved, high-intent audience this dialysis chain found in its regional-language segment. The specific languages and the specialty will differ from this scenario, but the structural lesson generalises: language match is often a stronger conversion signal in Indian healthcare than most practices assume until they test it directly.
The independent per-language compliance review discipline described above isn't optional overhead — it's the safeguard that makes multilingual healthcare advertising defensible, and any practice running ads in more than one language should build that review step in from the start rather than treating a cleared primary-language variant as sufficient cover for its translations.
