Five health and wellness brands run a verifiably live, shopper-facing AI assistant as of July 2026: Thorne (Taia), HUM Nutrition (Holly), The Vitamin Shoppe (Shoppe Advisor), Inno Supps (Nova, built on Alhena), and NutraBio (Rep AI). All five advise and recommend. None completes checkout on its own.
Everything else marketed as "wellness AI" is usually one of three other things: a clinician tool, a back-office copilot, or a wearable's health coach.
Supplement brands using AI are far rarer than the press releases suggest. We opened every claimed deployment in the category, one storefront at a time, and this is what survived the click.
Which supplement brands are using AI right now?
Ten claimed deployments, opened and confirmed during the week of July 6 to 12, 2026. Ordered by evidence strength.
| Brand and assistant | What it actually does | Status | Evidence and disclosed result |
|---|---|---|---|
| Thorne "Taia" |
Ingredient, benefit and dosing questions, personalized recommendations, product comparisons on thorne.com. No checkout. | Live, advisory | Brand product page, verified Jul 2026. Thorne's chief science officer told Glossy (Apr 2026): 200,000+ messages and 350,000+ recommendations in six months, roughly 8% higher AOV among Taia users. Company-reported, engaged versus unengaged. |
| HUM Nutrition "Holly" |
AI Nutritionist trained on HUM's dietitian-reviewed content, multilingual. Gated behind the on-site quiz and dashboard. | Live, advisory | Brand help centre and official channels, verified Jul 2026. No results disclosed. |
| The Vitamin Shoppe "Shoppe Advisor" |
In-store AI touchscreen: product information, goal-based semantic search, live inventory. | Live, in-store | Company press release, Jan 28, 2026. Launched at the NYC Innovation Store; the COO has since described expansion to associate iPads (Retail Brew, Apr 2026). No results disclosed. |
| Inno Supps "Nova" |
AI Performance Coach on innosupps.com. Storefront chat plus an on-page "Have questions?" module. Built on Alhena. | Live, advisory | Scoped live-site observation, Jul 12, 2026. Widget carries an "AI Agent by Alhena AI" mark and an AI-generated-responses notice. No results public. Alhena customer, disclosed. |
| NutraBio Rep AI assistant |
On-site sales and support assistant across 400+ SKUs. | Live, advisory | Vendor case study (Rep AI), 2026: 11.46% conversation-to-sale and roughly 10% higher chat-assisted AOV. Vendor self-reported, selection effects likely. |
| Grüns Sidekick + ChatGPT |
Two unrelated things: a back-office operator copilot, and direct purchasing through ChatGPT as an acquisition channel. | Off-storefront | Official Shopify case study. Founder Chad Janis frames ChatGPT purchasing as a new channel for a business that is more than 90% recurring revenue. No sales data disclosed. |
| Hims & Hers "MedMatch" |
Recommends treatment options to the prescribing provider. Beta scoped to anxiety and depression. | Not a shopping agent | Company newsroom, Nov 2023. Provider-facing. No results disclosed. |
| Hims & Hers "Labs AI" |
Finds patterns across up to 130 biomarker tests in 10 health areas. Educational, never diagnostic, escalates to clinicians. | Not a shopping agent | Company newsroom, May 7, 2026. Corroborated by Fierce Healthcare. No results disclosed. |
| Wellness Extract Salesforce Agentforce |
Support and engagement automation. Adoption announced, shopper-facing status not confirmed. | Announced | Salesforce Agentforce 2.0 press release, Dec 17, 2024, quoting the brand's program manager. An announcement, not a verified shipped shopper agent. |
| Persona Nutrition Wellness Advisor |
AI wellness advisor reachable at a dedicated subdomain, not linked from the main storefront. | Unconfirmed | Live at ai.personanutrition.com, observed Jul 2026. Did not resolve into a working end-to-end assessment. Production status genuinely unconfirmed. |
Read the last column before the first. It is where most market coverage stops.
Start with the results column. For eight of the ten deployments, the strongest disclosed result is "none." Only two brands have put a number into the public record, and both numbers compare shoppers who chose to engage against everyone else.
That is not a knock on either brand. It is the state of the evidence. In health and wellness, almost nobody is publishing outcome data on consumer AI, so a census has to grade on what is live and how it is built.
What does "live" actually mean in wellness AI?
A brand can truthfully say it "uses AI" while shipping something a shopper will never see. Four different things get called the same word, and they carry different value and different risk.
1. Shopper-facing product advisors
Answer a shopper's questions and recommend from the brand's own catalog. The rarest of the four. Taia, Holly, Nova, NutraBio.
2. Provider-facing or clinical support
Recommends to a licensed clinician or interprets medical data. Never aimed at a shopper filling a cart. MedMatch, Labs AI.
3. Merchant and back-office copilots
Help the operator forecast, write, or analyze. The shopper never touches them. Shopify Sidekick is the clearest case.
4. Biometric coaching and diagnostics
Interpret your own wearable or lab data and offer guidance. Genuine AI products that sell nothing. Oura Advisor, WHOOP Coach.
Only the first category is what "AI shopping" coverage usually claims to be about. Keeping the other three separate is the single most useful thing an operator can do when reading a competitor's announcement.
How do you grade an AI deployment claim?
Four questions, asked of every row above, and worth asking of any vendor demo or press release you are handed.
- What is live? A shipped, reachable feature, or a plan named at a launch event.
- What evidence supports it? See the ladder below.
- What result is disclosed? A specific, defined number, or nothing.
- What does it not prove? The limitation that stops you over-reading the claim.
Two further distinctions decide how much a deployment is worth.
- Action scope. Does the agent only advise, or can it complete a purchase?
- Grounding. Does it answer from the brand's live product and policy data, or from the open internet?
In a regulated category, grounding is not a nicety. An advisor that free-associates about what a supplement can do for a disease is a compliance problem, which is exactly why the live deployments are so conservative.
Which brands are live, and what does each one do?
Does Thorne have an AI shopping assistant?
Yes. Taia is a shopper-facing AI wellness advisor on Thorne's own site. It answers ingredient, benefit and dosing questions, gives personalized supplement recommendations, and compares products against each other.
It does not process checkout. It advises, then hands the shopper back to the normal buying flow.
What it does not prove: the disclosed 8% AOV difference compares Taia users against plain site visitors. Higher purchase intent, not the assistant, could explain part of that gap.
What is HUM Nutrition's Holly?
Holly is HUM's AI Nutritionist, trained specifically on the brand's dietitian-reviewed content rather than the open web, and available in multiple languages.
Access is gated behind HUM's quiz and member dashboard, which makes it a narrower surface than a storefront-wide agent.
What it does not prove: no performance data has been disclosed, and the exact surface coverage is unspecified.
Does The Vitamin Shoppe have an AI advisor?
Yes, but in a store rather than on a website. Shoppe Advisor is an AI touchscreen launched at the retailer's NYC Innovation Store, offering product information, goal-based search and live inventory.
It is the only physical-retail deployment in this census, and the COO has since described extending the capability to associate iPads.
What it does not prove: one pilot location, in-store only, with no published usage or sales outcome.
What powers Inno Supps' Nova?
Nova is Inno Supps' AI Performance Coach, running on Alhena's AI shopping assistant. It is the clearest two-surface deployment in this census: a storefront chat agent plus an on-page "Have questions?" module that seeds real product questions directly on the product page.
Both surfaces answer from the brand's own catalog and content. The widget tells shoppers that responses may be AI-generated, links the privacy terms, and carries an "AI Agent by Alhena AI" mark. In a category this regulated, visible disclosure is part of the design, not decoration.
What it does not prove: no performance numbers are public, so by this census's own rules the row gets none. Inno Supps is an Alhena customer and the relationship is disclosed.
Is NutraBio's assistant a real shopping agent?
Yes. NutraBio runs an on-site sales and support assistant across a large supplement catalog, and it is the only deployment besides Thorne with published numbers attached.
Those numbers come from the technology vendor's own case study: an 11.46% conversation-to-sale rate and roughly 10% higher average order value on chat-assisted purchases.
What it does not prove: vendor self-reported figures with no independent verification and no defined control group. Treat them as a hypothesis to test, not a benchmark. Some wider ROI figures circulating for this deployment appear to be vendor-level rather than NutraBio-specific.
Is Grüns selling supplements through ChatGPT?
Grüns is the cleanest illustration of why the category label matters, because one brand is doing two unrelated things at once.
Shopify's case study documents Sidekick as a back-office copilot for the operator, and separately describes the brand embracing direct purchasing through ChatGPT, where an outside AI agent can recommend and start a subscription. Founder Chad Janis frames that second surface as a new acquisition channel for a business that is more than 90% recurring revenue.
What it does not prove: neither surface is a brand-owned shopper-facing agent on the storefront, and no sales data has been disclosed for either. Read only the headline and you would merge them.
Are Hims & Hers MedMatch and Labs AI shopping assistants?
No, and this is the most commonly miscounted entry in the category. MedMatch recommends treatment options to the prescribing provider, not to the shopper, and its beta is scoped to anxiety and depression.
Labs AI interprets a user's results across up to 130 biomarker tests spanning ten health areas. It is built to never diagnose and to escalate to a licensed clinician.
What it does not prove: both are substantial AI systems. Neither is a consumer product-shopping assistant, and counting them as one overstates how much retail AI is actually live.
Is Wellness Extract's Agentforce agent live?
The brand is named in an official Salesforce release, which is good sourcing. But the quote is a statement of intent about the possibilities the platform brings, and the adoption described is support and engagement automation.
What it does not prove: an announcement is a real signal, not a shipped shopper-facing agent. Verify current status before listing this brand as live.
Is Persona Nutrition's AI advisor in production?
Persona's wellness advisor is reachable at a dedicated subdomain, but on observation it did not resolve into a working end-to-end assessment and is not linked from the main storefront.
What it does not prove: this is genuinely unconfirmed rather than confirmed-absent. A reachable URL is not evidence of a production shopper experience.
Why does almost nothing complete the purchase?
Every confirmed shopper-facing advisor in this census stops at the same place. They answer, they recommend, and then they hand the shopper back to the standard cart.
The reason is regulatory posture, not technical difficulty. In April 2026 the FDA issued what compliance specialists describe as its first warning letter centred on AI use, to a contract manufacturer, citing the use of AI agents to generate specifications, procedures and production records without adequate human review under 21 CFR 211.22(c).
That letter concerns factory records, not a shopping chatbot, so do not over-read it. What it establishes is a posture: regulators now treat unreviewed, ungoverned AI in a health context as a problem in itself, not only when a specific claim is wrong.
Now consider what a supplement advisor is actually asked: can I take this with my blood pressure medication, is this safe during pregnancy. An advisor answering those from the open web, or drifting from "supports healthy sleep" into "treats insomnia," produces exactly the unsupported health claim that draws enforcement. The full structure-and-function versus disease-claim analysis is its own subject, covered in our guide to AI compliance for supplement brands under FDA and FTC rules.
Which wellness brands do not have an AI shopping assistant?
A census is only as honest as its negatives. Publishing who is not live is the part almost no market map does, and it is often more useful than another logo grid.
| Brand | Finding as of the July 2026 check | Status |
|---|---|---|
| Ritual | No public AI shopping assistant found. Competes on browsable category navigation, ingredient transparency and third-party verification rather than a conversational advisor. | Not found |
| Seed | No public customer-facing AI shopping assistant. Its named AI effort, CODA, is an internal computational-biology and R&D platform, not a shopper tool. | Internal AI only |
| Nutrafol | No public AI shopping assistant found. Standard storefront, no announced conversational agent. | Not found |
| Care/of | Brand shut down. Subscriptions were cancelled June 17, 2024 after majority owner Bayer ceased further investment. It never shipped an AI-branded assistant, only a rules-based quiz. | Closed |
Two cautions apply to every "not found" above. Absence of public evidence is not proof that nothing exists internally, and a JavaScript-loaded widget can hide from a static check. These are scoped findings with a date, not permanent verdicts.
The direction of travel is still real. Care/of, once the most visible personalized-vitamin brand in the category, is gone, and the survivors have mostly not replaced the static quiz with a conversational agent.
How do you spot AI-washing in this category?
During research for this census, at least one widely repeated claim about a named vitamin-brand "AI assistant" traced back to content-optimization articles rather than a live product.
- If the claim is unnamed, treat it as unverified. Real deployments have a product name.
- If the claim is unlinked, treat it as unverified. Open it yourself before you repeat it.
- If the only source is a roundup article, find the brand's own page or press release.
- If the source is the vendor, keep the number and keep the label attached to it.
Do AI shopping assistants increase supplement sales?
The public evidence is thin but no longer empty. Here is everything that exists, with the label that belongs on it.
| Disclosed result | Source and date | Evidence label |
|---|---|---|
| Roughly 8% higher AOV for Taia users, plus 200,000+ messages and 350,000+ recommendations in six months | Thorne chief science officer via Glossy, Apr 2026 | Company-reported, engaged vs unengaged |
| 11.46% conversation-to-sale and roughly 10% higher chat-assisted AOV | Rep AI case study for NutraBio, 2026 | Vendor self-reported |
| AI-referred retail visitors generated 53% more revenue per visit and converted 54% higher than non-AI traffic | Adobe Analytics, reported mid-2026, from over one trillion US retail site visits | Third-party analytics, cross-category |
| 3x conversation-to-purchase conversion versus site average and 38% AOV uplift among AI-engaged shoppers | Alhena case study, Tatcha | Alhena customer, beauty not wellness, disclosed |
Every row above compares shoppers who chose to engage against everyone else. Self-selection explains some unknown share of each gap. That caveat belongs on your own dashboard too.
Consumer appetite is not the constraint. Roughly 32% of US adults told Rock Health they had used an AI chatbot for health information as of December 2025, double the 16% a year earlier. US supplement sales reached $74.15 billion in 2025, up 7.1% year over year according to the Nutrition Business Journal. The gap between that demand and ten claimed deployments is the whole story of this census.
Why is being cited by AI a different game from running one?
Put this census next to the AI-visibility data and you are looking at two separate capabilities that get discussed as one.
In 5W's Supplements AI Visibility Index, built from 3,800 prompts across ChatGPT, Claude, Perplexity, Gemini and Google AI Overviews between January and April 2026, four brands captured over 47% of observed citation share: Thorne, Seed, AG1 and Momentous.
Of those four, only Thorne runs a live shopping agent. Seed's named AI is an internal R&D platform. AG1 and Momentous have no public assistant at all.
This matters because of where AI answers get their material. Research reported through eMarketer in late 2025 found that 86% of citations in AI-generated responses come from sources brands directly control or strongly influence, such as their own sites, listings and reviews.
The engines also disagree with each other. Analyses of hundreds of millions of citations put the overlap between domains cited by ChatGPT and by Perplexity at only around 11 to 12%, and Perplexity cites roughly three times as many sources per answer as ChatGPT does. Publishing for one engine is not publishing for all of them.
Being the brand AI engines recommend and running an agent that serves your own shoppers are separate capabilities. The brands in this census are building the second. Most of the visibility winners have only the first, which is the subject of Alhena AI Visibility.
What should you do next if you operate a wellness brand?
Read this census as a set of decisions rather than a leaderboard. Three conclusions hold up.
1. Grade competitors on surface, not on the word "AI"
The useful question is never whether a competitor has AI. It is which of the four surfaces they shipped, whether the agent transacts or only advises, and whether it is grounded in vetted product data or in the open web.
A back-office copilot and a shopper-facing advisor are both "AI" and they compete for none of the same ground.
2. Measure your own results instead of borrowing a vendor's
Outcome data for consumer wellness AI is mostly not public yet. Do not wait for the category to publish proof before you build, and do not repeat unverified numbers either.
- Define the comparison group before launch, not after.
- Separate shoppers who chose to engage from those who did not.
- Measure lift against a real baseline, not the site average.
- Track advisory outcomes, such as returns and support deflection, alongside revenue.
3. Build the boundary before the feature
The deployments that are live decided what the agent will not do, which is give medical advice, imply a disease claim, or answer from an ungrounded source, before they decided what it would do.
That constraint is why the confirmed advisors advise and hand off. If you are working out the build order, our six-decision playbook for deploying a wellness AI agent covers the sequence, and the operator's guide to AI shopping agents for wellness brands covers whether to deploy at all.
Key takeaways
- Five brands are verifiably live. Thorne, HUM Nutrition, The Vitamin Shoppe, Inno Supps and NutraBio. Everything else is a different surface or an announcement.
- None of them transacts. Every brand-owned wellness agent stops before checkout. The only purchase surface in the category sits inside a third-party assistant.
- Eight of ten deployments disclose no results. Only Thorne and NutraBio have published numbers, and both compare engaged shoppers against everyone else.
- Most "wellness AI" is not shopping. Provider tools, back-office copilots and biometric coaches get counted as shopping agents and inflate the picture.
- Compliance sets the ceiling, not technology. The FDA's first AI-focused warning letter in April 2026 signalled that ungoverned AI in a health context is itself the problem.
- Visibility and deployment are different games. Four brands hold over 47% of AI citation share in supplements. Only one of them runs a live agent.
See a grounded wellness agent on your catalog
Alhena agents answer from your live product and policy data, not the open web, so recommendations stay inside content you have already vetted.
Frequently asked questions
Five are verifiably live as of July 2026: Thorne (Taia), HUM Nutrition (Holly), The Vitamin Shoppe (Shoppe Advisor, an in-store kiosk), Inno Supps (Nova, built on Alhena) and NutraBio (Rep AI). Persona Nutrition operates an advisor at a subdomain with unconfirmed production status. Most other "AI" in the category is provider-facing, back-office or biometric coaching rather than shopping.
Taia is a shopper-facing advisor. On thorne.com it answers ingredient, benefit and dosing questions, gives personalized recommendations and compares products. Based on the product page it does not process checkout. It advises and hands the shopper back to the normal buying flow.
No. MedMatch is provider-facing and recommends treatment options to the prescribing clinician, with a beta scoped to anxiety and depression. Labs AI interprets a user's biomarker results across up to 130 tests, is designed never to diagnose, and escalates to a licensed clinician. Both are significant AI systems, but neither is a consumer shopping assistant.
A shopping assistant is shopper-facing: the customer interacts with it for recommendations or support. A back-office copilot such as Shopify Sidekick helps the operator with analysis and content, and the shopper never sees it. Grüns is documented using both, which is a common source of confusion.
Nova, Inno Supps' AI Performance Coach, runs on Alhena. It appears as a storefront chat agent plus an on-page "Have questions?" module, both answering from the brand's own catalog and content, and the widget carries an "AI Agent by Alhena AI" mark with an AI-generated-responses notice. Inno Supps is an Alhena customer, disclosed here, and no performance numbers are public.
Yes, Care/of shut down. After majority owner Bayer ceased further investment, the brand cancelled subscriptions on June 17, 2024 and laid off staff. It never shipped an AI-branded assistant, only a rules-based quiz. As of the July 2026 check date there is no confirmed revival.
The public evidence is thin but no longer empty. Thorne disclosed roughly 8% higher average order value among Taia users alongside 200,000+ messages in six months. A vendor case study for NutraBio reports an 11.46% conversation-to-sale rate and about 10% higher chat-assisted AOV. Both compare engaged shoppers against everyone else, so selection effects likely explain part of the difference. Treat single-source figures as hypotheses to test, not benchmarks.
Compliance risk, more than technology. Supplements sit in a regulated, high-stakes category, and an assistant that gives medical advice or implies a disease claim can generate exactly the unsupported health claim that draws FDA or FTC attention. In April 2026 the FDA issued what specialists call its first AI-focused warning letter, over ungoverned AI use in regulated records. Live advisors respond by narrowing scope: answer from vetted brand-owned content, decline medical advice, and hand off before diagnosis or checkout.
Grade it on four things: what is actually live versus merely announced, what evidence supports it, what result is disclosed, and what the claim does not prove. Then classify the surface as shopper-facing, provider-facing, back-office or biometric coaching, and check whether the agent transacts or only advises. Many impressive announcements resolve into a back-office tool, a stated intent, or a coaching feature that sells nothing.
In 5W's Supplements AI Visibility Index, built from 3,800 prompts across ChatGPT, Claude, Perplexity, Gemini and Google AI Overviews between January and April 2026, Thorne, Seed, AG1 and Momentous together captured over 47% of observed citation share. Only Thorne among them runs a live shopper-facing agent, which is why citation share and deployment should be tracked as separate capabilities.
