Vertical Performance Benchmarks for AI Advertising in Health and Wellness

Health and wellness advertisers sit on more benchmark data than almost any other category in digital marketing, and almost none of it answers the question that actually matters: what happens when an ad meets someone while they're still figuring out what they need, before that need gets flattened into a search term or a demographic bucket. Facebook and Google can tell you what a click cost. Neither can tell you why the person clicked, or what they were afraid of five minutes before they did. This piece works through what the legacy numbers say, what conversational AI surfaces that those numbers structurally can't, and what the gap between the two means for anyone buying media in this category right now. The intent signal in a health-related AI prompt already beats anything social or search has ever handed an advertiser, and buyers who wait for a finished scorecard before acting on that will spend a year behind the ones who didn't.

What the legacy social benchmarks actually look like for wellness advertisers right now

Start with the anchor number. Wellness and Holistic Health averaged a 1.99% click-through rate on Facebook over the 13 months ending January 2026, against a 1.86% global benchmark, according to Superads' analysis of $3 billion in ad spend. Call it 7 to 8% above market. That's real, but it's thin. It doesn't come close to explaining the attention and budget this category commands.

The average also buries the more telling detail: how unstable it is. First-half 2025 CTR ran 1.82%; second half jumped to 2.15%, an 18% swing inside a single calendar year. Q3 hit 2.17%, Q4 landed close behind at 2.14%. Wellness doesn't perform consistently well on Facebook. It performs well in bursts and sits flat the rest of the time, and treating the annual average as a steady-state number is the first mistake most media planners make in this category.

Cost tells a matching story, and here's where the confusion usually creeps in. CPC fell from about $1.47 in early 2025 to roughly $0.93 by January 2026, a 37% drop, and it's tempting to read that as an efficiency win. It isn't one. CPMs sat above the global market for the entire period, so advertisers paid a premium throughout, not just during the hot quarters. The falling CPC lines up with demand patterns outside the hot quarters, not any real improvement in targeting or match quality. Confusing the two is the kind of mistake that gets a budget approved on the wrong logic.

The "wellness" label itself hides more than it reveals, and lumping its sub-verticals together is the second mistake. Mental health and therapy advertising, pharmaceutical, and general wellness each carry different creative constraints and buyer profiles, which tends to push their CTR and CPA numbers apart even within the same blended benchmark.

What these benchmarks measure well: reach, auction pricing, how a piece of creative lands in a crowded feed. What they can't measure at all: the reasoning that got someone into that feed in the first place.

Why search benchmarks tell a similarly incomplete story in this category

Search gets treated as the serious channel in health and wellness, the one where intent is supposedly proven rather than guessed at. Industry benchmarks for fitness and wellness bear that out on cost: search tends to deliver a higher CPA than display or social formats, alongside stronger conversion rates and ROAS. Every performance marketer recognizes that trade. You pay more per click because the person clicking already narrowed a category down, even if not yet a brand.

But that narrowing is exactly what search can't get past. It only sees demand after that demand has been compressed into a keyword. Someone typing "magnesium supplement" already did the mental work of reducing a decision to three words, so the search engine gets a real signal, just a thin one. It has no visibility into the thyroid diagnosis behind it, the soy allergy, the six weeks of half-abandoned browser tabs that led to those three words.

That earlier phase is where health decisions actually get made. It used to happen with a doctor, a friend, or a forum thread at 11pm. Increasingly it happens in a chat window with an AI assistant, a phase search was never built to see. Social has the same blind spot from a different angle: it intercepts a demographic proxy, not a decision in progress. Both channels measure what happens after intent gets flattened into something an ad system can read. Neither can see it while it's still forming, and that structural gap is what the rest of this piece is built around.

What conversational AI surfaces that search and social cannot: the prompt as intent evidence

The contrast in raw signal is stark. A typical Google search runs three or four words. A ChatGPT prompt, per Pacvue's analysis, runs closer to a full paragraph, laying out the objective, the constraints, the context, and the trade-offs at once, unprompted, in the user's own words.

OpenAI's own data puts the scale of this in perspective: roughly 20% of ChatGPT conversations carry shopping intent, spanning categories that include fitness and beauty, out of a corpus running 2.5 billion daily queries as of July 2025. That's a commercial layer too large to wave off as noise.

Health prompts in particular stack several intent signals into a single message. A symptom-to-solution prompt can carry a working diagnosis, a budget, a lifestyle constraint, and a stated preference for natural over pharmaceutical options, all in one exchange. A comparison prompt might name two products by brand and ask which one fits a specific medical condition, late-funnel intent showing up in the middle of a conversation instead of at the end of a click path. A prompt like "is it safe to combine X and Y" is someone actively deciding something in real time, not browsing idly.

Here's the honest caveat, and it matters: most of this traffic isn't commercial in a narrow sense. Profound's 2025 analysis of more than 50 million ChatGPT prompts found only 9.5% classify as commercial and 6.1% as transactional, against 32.7% informational and 37.5% generative or task-completion. But health and wellness is one of the few categories where that informational share still carries real buying signal, because research and purchase sit closer together here than almost anywhere else. Targeting an informational health prompt isn't wasted spend. It's reaching someone at the exact moment they're most open to a recommendation, before they've hardened into a brand-specific search.

Sequence matters as much as content here. Verve's analysis of more than a billion daily signals found the pre-purchase journey now starts with AI chat for over 20% of users overall, and in travel, that figure runs as high as 37%, already producing more first-touch queries in LLMs than in search. Health, with its research-heavy and emotionally loaded decisions, fits the profile of a category where that share could run high. These are research-heavy, emotionally loaded decisions, exactly the kind that pull someone into a long back-and-forth with an assistant instead of a quick search-and-click.

How conversational AI advertising works mechanically, and what makes health a distinctive targeting environment

Mechanically, LLM advertising places ads inside the assistant's own response, matched to what the conversation is about rather than to a webpage, a keyword, or a cookie-built profile. No third-party tracking pixel does the work here. The match comes from the content of the exchange itself.

OpenAI's ad model runs as real-time bidding: an advertiser sets a budget, some contextual targeting parameters, and creative, and the system places a relevant ad below the assistant's response, billed on CPM or CPC depending on the campaign's goal. The real departure from a search auction is that relevance isn't a tiebreaker among similar bids. It's a direct input into cost and placement. A better contextual match can beat a higher bid that fits worse.

A framework called LERA adds another layer: the LLM itself scores the relevance of candidate ads, and that score combines with the bid under a payment rule built around advertiser utility rather than raw price, extended across multi-turn conversations rather than a single exchange.

Search auctions never had to deal with what researchers call the generative externality problem. Inserting an ad into a generated response can change the tone, length, and specificity of that response, not just what sits beside it. In health, that's not a theoretical risk. A response that turns transactional the instant a sponsored product gets slotted in can cost both advertiser and publisher user trust, and fast. Brand-safety filtering and clear ad disclosure aren't nice-to-haves in this category. They're the whole game, and any buyer treating them as a compliance checkbox instead of a trust mechanism is going to get burned.

On the publisher side, the plumbing generally runs like this: install an SDK, make a server-side call carrying prompt context (no personal data attached) and a session token, get back a structured ad object with a title, body copy, a call to action, an advertiser URL, and a disclosure string, then render it inside the publisher's own interface. Evaluation for health-specific integrations comes down to latency (sub-250ms at the 95th percentile is the SLA publishers target), how cleanly the "Sponsored" label meets disclosure norms, how granular the brand-safety filters are for pharma and regulated claims, fill rate on prompts that carry real commercial intent, and revenue split.

Regulation sits over all of it. Ad copy here can't carry a clinical claim that wouldn't survive FTC or FDA review, and a claim delivered inside a helpful-sounding AI response reads as more authoritative than the same line on a banner ad ever would. That raises the cost of a compliance mistake considerably, and it's the reason health advertisers can't treat this inventory the way they'd treat a generic e-commerce placement.

Where the AI advertising market stands today, and what that means for health and wellness buyers

The trajectory isn't in question. eMarketer's 2026 projection puts US AI ad spending at $68.25 billion by 2030. The mix buyers need to sit with is the near-term one: eMarketer estimates that in 2026, more than 80% of AI advertising will run adjacent to AI-generated content (think Google's AI Overviews) rather than inside a genuine chatbot conversation. Search-adjacent AI advertising is forecast to grow 152% to $26.42 billion. Chatbot-native ad spend is forecast to grow 1,641%, a number that sounds enormous and still only lands at $0.96 billion. This market is in its earliest stage of forming. It isn't mature enough to buy on autopilot, and anyone selling a health and wellness brand a fully-baked chatbot media plan right now is overselling inventory that doesn't yet exist at scale. Platforms like Thrad, which runs its own exchange alongside a DSP for AI chat environments, sit closer to that conversational inventory than generalist buying tools can.

Supply is still sorting itself out too, and unevenly. OpenAI Ads reportedly reached a $1 billion annualized revenue run rate across more than 40 countries, a pace that marks it as one of the more striking new ad platform launches in recent memory. Perplexity, by contrast, stopped accepting new advertisers at some point in 2025, so its ad business is a closed pilot, not something a brand can plan a budget around. Google has been deliberately slow rolling out Gemini ads while it builds the user base first; Sundar Pichai referenced "native ad concepts" specific to Gemini in early 2025 without giving a launch date.

None of this supply is charity. Running large language models costs money, on the order of $700,000 a day in compute infrastructure by one reported figure for ChatGPT, which makes ad revenue a financial necessity for free-tier users, not a side experiment platforms can afford to delay. For health and wellness buyers, that points to a workable near-term sequence: treat ChatGPT's ad platform as the entry point today, use AI Overviews and other search-adjacent placements for immediate scale, and build fluency in pure chatbot inventory now, ahead of the expansion rather than in response to it.

Reading the benchmark gap: what the numbers suggest about performance potential in health and wellness AI advertising

Be straight about what doesn't exist yet: there's no published head-to-head benchmark table comparing AI advertising performance to Facebook or Google search performance at the vertical level for health and wellness. The decade of data built up for those legacy channels simply hasn't been built for conversational AI yet. So the case here is structural and inferential, not a finished scorecard, and it should be read as exactly that.

What the legacy data shows: wellness social advertising sits modestly above the market average on CTR (1.99% against a 1.86% global baseline on Facebook, per Superads), pays a real CPM premium to get there, and swings quarter to quarter instead of holding steady. What the richness of AI conversational signal implies, without yet proving, is different in kind: a prompt that names a symptom, a specific product, a budget ceiling, a lifestyle constraint, and a sense of urgency all at once hands an ad system something a Facebook click never has. Matching against that kind of layered context should outperform matching against an interest category or a lookalike audience. The signal is richer. That much is settled. What isn't settled is whether current measurement tools can actually track the conversion once it happens, and that's the real gap, not the quality of the signal itself.

That measurement gap is not a footnote. AI assistant interactions generally don't leave the click-trail that search and social attribution were built to follow. Someone who asks ChatGPT about a supplement, sees a sponsored suggestion, and buys it on the brand's own site the next day may never get credited back to that conversation. This problem isn't unique to health and wellness, but it bites harder here: CPAs across the category's sub-verticals stretch from modest (mental health and therapy) to steep (pharmaceutical), and a blind spot in attribution costs more in real dollars the higher that CPA baseline sits.

The fair framing for a buyer, then, runs like this. The intent-signal advantage over legacy channels is structural and already well-documented. The performance uplift that should follow from it points in one clear direction. The measurement infrastructure to confirm that uplift with precision is still being built, and pretending otherwise misrepresents where this market actually stands. CuFinder's 2026 Fitness and Wellness Benchmarks offer one useful anchor on why the gap matters financially: customer lifetime value in fitness and wellness runs high enough that even a modest CPA reduction, driven by better intent-matching rather than better bidding, meaningfully shortens payback period.

What health and wellness brands can act on now, given where benchmarks and inventory actually stand

Waiting for the channel to mature before getting involved is the wrong call, and the data above is the reason why. The signal advantage exists today, even while the tools to measure it with full precision are still catching up. Brands that spend the next year building real fluency, in how the auctions price relevance, what disclosure requires in a regulated category, how brand-safety review actually works for pharma-adjacent claims, will hold a structural edge once chatbot-native inventory scales the way search-adjacent inventory already has.

The benchmarks reviewed here aren't a reason to sit on the sidelines. They describe a category that already beats the market average on social, already pays premium CPAs for its highest-intent search traffic, and is now watching a new channel emerge that captures intent earlier and in far more detail than either of those two ever could. Waiting for a finished scorecard doesn't protect a media budget. It just hands a year's head start to whichever competitor already learned the auction system, the disclosure standard, and the measurement approach that will define how this category buys media for the next decade.