How a pulmonology hospital ran Growth-tier ChatGPT Ads for its ILD programme with clinician-facing intent
The situation
Picture a hospital with a dedicated interstitial lung disease programme sitting inside its broader pulmonology department — a multidisciplinary set-up combining a pulmonologist, a radiologist reading high-resolution CT scans, and often a rheumatologist for the connective-tissue-disease-associated cases, the kind of programme most general pulmonology departments don't run with any real depth. ILD is a comparatively rare and diagnostically difficult category of lung disease, and the hospital's actual patient flow reflected that: a meaningful share of its ILD patients arrived not through direct patient search at all, but through referral from another physician — a general pulmonologist elsewhere who suspected ILD but didn't have the diagnostic or multidisciplinary infrastructure to confirm and manage it, or a rheumatologist managing a connective-tissue-disease patient who needed a pulmonology partner for the lung-involvement side of the case.
This referral pattern was the hospital's real paid-acquisition blind spot. Its existing Google Ads presence was built entirely around patient-facing search terms — symptom-based queries like breathlessness or chronic cough — and had no mechanism at all for reaching the referring-physician audience, because that audience doesn't typically type symptom-based search queries; a physician evaluating where to refer a suspected ILD case researches differently, often asking more technical, capability-specific questions about diagnostic infrastructure and specialist availability, exactly the kind of nuanced, multi-part question a conversational assistant handles better than a keyword search.
A second complication shaped the brief: ILD, unlike many conditions patients recover fully from, is generally a progressive disease category, and the hospital's own clinical team was firm that any advertising touching the programme could not imply that treatment reverses existing lung damage or halts progression outright, since that overstates what current ILD management can genuinely deliver for most patients. The hospital needed a campaign that could speak credibly and technically to a referring physician's evaluation criteria, speak accessibly to a directly-searching patient or family, and do both without ever drifting into an outcome claim the clinical team couldn't stand behind.
The ChatGPT Ads campaign structure ICG designed
Growth-tier's structure ran as three buckets under one hospital brand: a clinician-referral bucket built specifically to capture conversations from physicians researching where to refer a suspected or confirmed ILD case, a patient-and-family second-opinion bucket for people who had already received an ILD diagnosis elsewhere and were evaluating treatment options, and a smaller general-education bucket for earlier-stage symptom-exploration conversations from people who hadn't yet received any diagnosis. Separating the clinician-referral bucket from the patient-facing buckets was the single most consequential structural decision in the account, since a referring physician and a directly-searching patient are, in effect, two different audiences who happen to be discussing the same disease category in almost entirely different registers.
Bid ladder design for the clinician-referral bucket weighted toward conversations using more technical framing — references to specific diagnostic tools like high-resolution CT or lung biopsy, mentions of connective-tissue-disease-associated ILD, or explicit language about seeking a referral partner or tertiary evaluation — at bids illustratively 6-8x above the bucket's baseline, reflecting how much more valuable a genuine clinician-referral conversation is relative to its lower overall volume compared with patient-facing search. The patient-facing second-opinion bucket ran its own ladder, weighted toward conversations referencing an existing ILD or pulmonary fibrosis diagnosis and evaluating treatment centres, while the general-education bucket stayed intentionally low-bid, since early-stage symptom exploration in a rare disease category converts far less predictably and the bucket's real job was awareness rather than direct acquisition.
Landing infrastructure ran three matched pages, and the clinician-referral page was built differently from anything else in ICG's typical hospital landing-page playbook — it read closer to a clinical capability brief than a patient-marketing page, listing the multidisciplinary team's composition, the specific diagnostic infrastructure available (HRCT protocols, biopsy capability, pulmonary function testing depth), typical turnaround time for a referred case's initial evaluation, and a direct clinician-to-clinician contact pathway rather than a general enquiry form, since a referring physician wants a fast, credible answer about capability, not a patient-style lead-capture experience. The patient-facing second-opinion page ran in more accessible language, explaining what an ILD evaluation and ongoing management pathway actually involves without technical jargon, and the education page addressed common early-symptom questions with appropriately cautious, non-diagnostic framing.
Conversion tracking distinguished clinician-referral conversions (a specific referral-pathway contact event) from patient-facing enquiry conversions, letting the hospital's programme lead see referral-source growth as its own metric separate from direct patient acquisition — a distinction the hospital's leadership specifically wanted visibility on, since growing the referral network was, if anything, a higher strategic priority for the programme's long-term case volume than direct patient search given how rare the disease category is.
The compliance discipline
NMC Section 6's outcome-guarantee restriction shaped the account's most sensitive line of copy: no variant across any of the three buckets, patient-facing or clinician-facing, implied that treatment would reverse existing lung fibrosis or halt disease progression outright, language that would have overstated what ILD management can currently deliver for most patients and would have been both non-compliant and clinically dishonest. Copy instead focused on accurate, defensible framing — early and accurate diagnosis, multidisciplinary management, and slowing progression where the specific ILD subtype allows for it — language the hospital's own pulmonology consultants reviewed and approved specifically for clinical accuracy before any variant reached ASCI-level compliance screening.
The clinician-referral bucket required a different compliance lens than ICG's typical patient-facing hospital work, since clinician-directed content is permitted more technical and comparative detail than patient-facing copy — a referring physician reasonably wants to know about diagnostic turnaround time and specialist availability in specific terms — but the account still held firmly to ASCI Chapter III's restriction on comparative claims against another named provider or hospital, describing the programme's own capability in specific, factual terms without any framing that implied another hospital's ILD service was inadequate, a distinction that mattered since referring physicians often have existing relationships with multiple tertiary centres and copy that read as disparaging a peer institution would have damaged credibility with exactly the audience the bucket was built to reach.
ASCI Chapter III's general accuracy standard applied to the patient-facing buckets with the same rigour ICG applies across its respiratory and chronic-disease work, avoiding any language suggesting ILD is easily or fully treatable. DPDP 2023 governed the patient-facing enquiry form, which collected existing diagnosis and current treatment detail as sensitive health data requiring explicit consent language, while the clinician-referral contact pathway, collecting professional rather than patient health data, ran under a lighter but still DPDP-conscious consent framework appropriate to a B2B-adjacent referral contact rather than a patient enquiry.
The 90-day outcome pattern
In the pattern ICG has observed on comparable Growth-tier sub-specialty programme accounts, the general-education bucket generated the highest raw volume but the clinician-referral bucket, despite far lower volume, delivered the strongest downstream value. Illustrative combined conversation-completion events across the three buckets: roughly 20-30 a week in month one, climbing to an illustrative 55-70 a week by month three, with the clinician-referral bucket contributing a comparatively small illustrative 3-5 conversations a week throughout, consistent with how narrow the referring-physician audience genuinely is relative to general patient search.
Despite that low volume, the clinician-referral bucket converted to an actual referred-case contact at illustratively 45-55%, far above either patient-facing bucket's conversion rate, consistent with the expected pattern that a physician actively researching a referral pathway for a specific case is about as high-intent a conversation type as exists in healthcare advertising. The patient-facing second-opinion bucket converted at illustratively 20-25% within 7 days for conversations referencing an existing diagnosis, while the general-education bucket, as designed, converted at a much lower single-digit rate, functioning more as an awareness and future-pipeline layer than a direct-conversion one.
The hospital's programme lead reported that the clinician-referral bucket, though small in absolute numbers, represented genuinely new referral relationships the programme hadn't had before — physicians who found the programme through a conversational assistant rather than through the hospital's existing conference and peer-network referral channels, which the programme lead considered a meaningfully different and additive acquisition source rather than a substitute for existing referral relationships. Blended cost per qualified conversion across all three buckets ran illustratively 10-15% below the hospital's blended Google Ads cost per pulmonology lead over the same quarter, with GA4's AI Assistant channel showing a key-event rate consistent with ICG's wider healthcare book, though the more meaningful metric for this particular account, per the hospital's own framing, was the qualitative shift of gaining a functioning clinician-referral acquisition channel where none had existed before.
What we'd do differently next time
The clinician-referral landing page launched in month one with reasonably generic multidisciplinary-team language before the hospital's own pulmonology consultants had a chance to weigh in on the specific diagnostic-capability detail referring physicians actually care about — HRCT protocol specifics, typical time-to-biopsy-result, named consultant availability — and the page's conversion rate improved noticeably once that detail was added in month two. Involving the clinical team in landing-page copy from day one, rather than treating clinician input as a month-two refinement, would have gotten the bucket to its stronger conversion rate faster.
The general-education bucket's low conversion rate was expected by design, but its budget allocation, set at roughly equal thirds across the three buckets at launch, meant it consumed spend that, in hindsight, could have been more productively shifted toward the clinician-referral bucket once that bucket's outsized conversion rate became clear by week six. Building in an earlier budget-reallocation checkpoint — day 30 rather than the full 90-day review — would have let budget move toward the clinician-referral bucket sooner.
Finally, the account didn't initially track which referring physicians' specialties were generating clinician-referral conversions — pulmonologist versus rheumatologist versus general physician — data that would have helped the hospital's programme lead understand which referral-source relationships the channel was strengthening most. Adding that classification from launch, rather than recognising the gap only once the hospital asked for it in month two, would have given a fuller picture sooner.
How this maps to your own sub-specialty programme
If your hospital runs a genuinely sub-specialised programme within a broader department — ILD within pulmonology, a comparable rare-disease or complex-case service within any specialty — and a meaningful share of your actual case volume arrives through physician referral rather than direct patient search, this Growth-tier structure of running a dedicated clinician-facing bucket alongside patient-facing buckets, with landing content built specifically for how a referring physician evaluates capability, is worth building even at modest budget, since the referral audience's low volume is more than offset by its conversion quality.
The core transferable lesson is that clinician-facing and patient-facing healthcare audiences genuinely need separate campaign architecture, not just separate ad copy within one bucket, because the two audiences ask fundamentally different questions of a conversational assistant, and a campaign built only around patient-facing symptom language will structurally never reach the referring-physician conversations that, for a sub-specialty programme, may represent its most valuable acquisition channel.
