The State of AI Discovery in Health Tech 2026
Automattik studied how ChatGPT, Google AI Overviews, Google AI Mode, Gemini and Perplexity recommend consumer health-tech providers across six health categories — and what those answers reveal about intent, mentions, sources and reputation.
AI visibility changes with what the consumer is actually looking for.
The same brand can look very different across health categories. The pattern can shift again when a consumer adds a specific need such as insurance support or speed. A brand can also be mentioned without being recommended, and the same exact pages can repeatedly resurface around closely related questions.
Visibility is highly intent-specific.
Category and exact consumer need can materially change the recommendation pattern.
Being mentioned is not the same as being recommended.
Presence in an answer can substantially overstate whether a brand is actually presented as suitable for the request.
Brand-owned content can appear alongside major publishers in AI answers.
On one weight-management question, a Found comparison article appeared in the source mix for the same question as Forbes and U.S. News, then resurfaced across 13 of 14 weight-management questions.
Reputation answers can mix evidence, entities and changing facts.
Named-provider answers can blend different evidence types, pull company-level issues into a service-level assessment, and conflict across dates.
Six health categories. 27 health-tech brands.
A purposive set of established consumer-health platforms, category anchors, growth-stage challengers and specialists. The study was not designed to represent every U.S. health-tech company.
Visibility is highly intent-specific
A brand can be recommended often in one health category and much less often in another. The picture can shift again when the consumer adds a specific need.
The same brand can look very different across health categories
Show exact counts and methodologyDetails
Weight management: Ro was recommended in 574 of 980 eligible answers; Hims in 123 of 980.
Men’s health: Ro was recommended in 362 of 560 eligible answers; Hims in 441 of 560.
Hair / dermatology: Ro was recommended in 251 of 490 eligible answers; Hims in 354 of 490.
U.S. · ChatGPT, Google AI Overviews, Google AI Mode, Gemini and Perplexity · Sept. 8–21, 2026. Matched same-answer denominators within each category. Descriptive context comparison only; not a universal brand verdict.
Use a brand-wide visibility number as a summary, not the whole story. Break results out by health category so teams can see where visibility is actually strong or weak.
Within weight management, the Ro–Hims gap was much wider on insurance questions
Show exact counts and methodologyDetails
Prior authorization: Ro 61 recommendations across 70 answers; Hims 1 across 70.
Wegovy / Zepbound insurance help: Ro 45 across 70; Hims 0 across 70.
Fast start: Ro 51 across 70; Hims 40 across 70.
Across the two insurance questions combined: Ro 106 recommendations across 140 answers; Hims 1 across 140.
These are distinct consumer requests, not a controlled experiment. The study did not independently verify current provider capabilities or establish why a model made a recommendation.
Track the need behind the question. Insurance, affordability, speed, privacy, continuity and specialist expertise can reveal differences hidden by a category average.
Explore recommendation results by health category
The main narrative stays light. Expand a category when you want to inspect the deeper recommendation results.
GLP-1 & weight management14 discovery promptsExpand
Category view
Show exact counts and methodologyDetails
| Brand | Recommended | Unresolved | Eligible answers | Recommendation | Mentioned | Mention unresolved |
|---|---|---|---|---|---|---|
| Ro | 574 | 0 | 980 | 58.6% | 587 | 0 |
| Weight Watchers Meds+ | 455 | 0 | 980 | 46.4% | 469 | 0 |
| Found | 453 | 1 | 980 | 46.2–46.3% | 466 | 0 |
| PlushCare | 173 | 0 | 770 | 22.5% | 173 | 0 |
| Hers | 129 | 0 | 980 | 13.2% | 139 | 0 |
| Hims | 123 | 0 | 980 | 12.6% | 132 | 0 |
| Sesame | 74 | 0 | 770 | 9.6% | 76 | 0 |
| Henry Meds | 59 | 0 | 770 | 7.7% | 60 | 0 |
Different denominators reflect question-specific eligibility. Unresolved classifications are preserved as a lower-to-upper range; they are not confidence intervals.
Mental health & psychiatryTherapy + psychiatry / medicationExpand
Category view
Show exact counts and methodologyDetails
| Brand | Recommended | Unresolved | Eligible answers | Recommendation | Mentioned | Mention unresolved |
|---|---|---|---|---|---|---|
| BetterHelp | 326 | 5 | 559 | 58.3–59.2% | 441 | 0 |
| Brightside Health | 463 | 0 | 839 | 55.2% | 466 | 0 |
| Cerebral | 48 | 0 | 909 | 5.3% | 51 | 0 |
| Talkiatry | 316 | 0 | 560 | 56.4% | 317 | 0 |
| Talkspace | 667 | 1 | 909 | 73.4–73.5% | 733 | 0 |
| Teladoc | 134 | 0 | 839 | 16.0% | 137 | 0 |
Different denominators reflect question-specific eligibility. Unresolved classifications are preserved as a lower-to-upper range; they are not confidence intervals.
Therapy
Show exact counts and methodologyDetails
| Brand | Recommended | Unresolved | Eligible answers | Recommendation | Mentioned | Mention unresolved |
|---|---|---|---|---|---|---|
| BetterHelp | 190 | 2 | 279 | 68.1–68.8% | 239 | 0 |
| Brightside Health | 79 | 0 | 279 | 28.3% | 79 | 0 |
| Cerebral | 0 | 0 | 279 | 0.0% | 0 | 0 |
| Talkspace | 199 | 0 | 279 | 71.3% | 228 | 0 |
| Teladoc | 33 | 0 | 279 | 11.8% | 33 | 0 |
Different denominators reflect question-specific eligibility. Unresolved classifications are preserved as a lower-to-upper range; they are not confidence intervals.
Psychiatry/medication
Show exact counts and methodologyDetails
| Brand | Recommended | Unresolved | Eligible answers | Recommendation | Mentioned | Mention unresolved |
|---|---|---|---|---|---|---|
| Brightside Health | 150 | 0 | 210 | 71.4% | 150 | 0 |
| Cerebral | 23 | 0 | 280 | 8.2% | 24 | 0 |
| Talkiatry | 236 | 0 | 280 | 84.3% | 236 | 0 |
| Talkspace | 191 | 0 | 280 | 68.2% | 191 | 0 |
| Teladoc | 32 | 0 | 210 | 15.2% | 32 | 0 |
Different denominators reflect question-specific eligibility. Unresolved classifications are preserved as a lower-to-upper range; they are not confidence intervals.
Men’s sexual & hormonal healthED + hormone healthExpand
Erectile dysfunction
Show exact counts and methodologyDetails
| Brand | Recommended | Unresolved | Eligible answers | Recommendation | Mentioned | Mention unresolved |
|---|---|---|---|---|---|---|
| BlueChew | 165 | 0 | 350 | 47.1% | 168 | 0 |
| GoodRx Care | 171 | 0 | 350 | 48.9% | 172 | 0 |
| Hims | 320 | 0 | 350 | 91.4% | 323 | 0 |
| Ro | 273 | 0 | 350 | 78.0% | 281 | 0 |
| Sesame | 81 | 0 | 350 | 23.1% | 82 | 0 |
Different denominators reflect question-specific eligibility. Unresolved classifications are preserved as a lower-to-upper range; they are not confidence intervals.
Hormone health
Show exact counts and methodologyDetails
| Brand | Recommended | Unresolved | Eligible answers | Recommendation | Mentioned | Mention unresolved |
|---|---|---|---|---|---|---|
| Hims | 43 | 0 | 210 | 20.5% | 47 | 0 |
| Hone Health | 176 | 0 | 280 | 62.9% | 176 | 0 |
| Maximus | 135 | 1 | 280 | 48.2–48.6% | 137 | 0 |
Different denominators reflect question-specific eligibility. Unresolved classifications are preserved as a lower-to-upper range; they are not confidence intervals.
Women’s digital healthMenopause + PCOS / hormonal + vaginal / urinaryExpand
Category view
Show exact counts and methodologyDetails
| Brand | Recommended | Unresolved | Eligible answers | Recommendation | Mentioned | Mention unresolved |
|---|---|---|---|---|---|---|
| Allara | 349 | 0 | 630 | 55.4% | 354 | 0 |
| Evvy | 154 | 0 | 630 | 24.4% | 154 | 0 |
| Hers | 67 | 0 | 1050 | 6.4% | 71 | 0 |
| Midi Health | 528 | 0 | 630 | 83.8% | 528 | 0 |
| PlushCare | 167 | 0 | 1050 | 15.9% | 169 | 0 |
| Sesame | 118 | 0 | 1190 | 9.9% | 122 | 0 |
| Wisp | 308 | 0 | 700 | 44.0% | 316 | 0 |
Different denominators reflect question-specific eligibility. Unresolved classifications are preserved as a lower-to-upper range; they are not confidence intervals.
Menopause
Show exact counts and methodologyDetails
| Brand | Recommended | Unresolved | Eligible answers | Recommendation | Mentioned | Mention unresolved |
|---|---|---|---|---|---|---|
| Hers | 0 | 0 | 280 | 0.0% | 0 | 0 |
| Midi Health | 257 | 0 | 280 | 91.8% | 257 | 0 |
| PlushCare | 12 | 0 | 140 | 8.6% | 12 | 0 |
| Sesame | 5 | 0 | 280 | 1.8% | 5 | 0 |
Different denominators reflect question-specific eligibility. Unresolved classifications are preserved as a lower-to-upper range; they are not confidence intervals.
PCOS/hormonal
Show exact counts and methodologyDetails
| Brand | Recommended | Unresolved | Eligible answers | Recommendation | Mentioned | Mention unresolved |
|---|---|---|---|---|---|---|
| Allara | 234 | 0 | 280 | 83.6% | 239 | 0 |
| Hers | 1 | 0 | 280 | 0.4% | 1 | 0 |
| PlushCare | 39 | 0 | 210 | 18.6% | 40 | 0 |
| Sesame | 66 | 0 | 280 | 23.6% | 67 | 0 |
Different denominators reflect question-specific eligibility. Unresolved classifications are preserved as a lower-to-upper range; they are not confidence intervals.
Vaginal/urinary health
Show exact counts and methodologyDetails
| Brand | Recommended | Unresolved | Eligible answers | Recommendation | Mentioned | Mention unresolved |
|---|---|---|---|---|---|---|
| Evvy | 131 | 0 | 280 | 46.8% | 131 | 0 |
| PlushCare | 51 | 0 | 280 | 18.2% | 52 | 0 |
| Sesame | 44 | 0 | 280 | 15.7% | 47 | 0 |
| Wisp | 162 | 0 | 280 | 57.9% | 170 | 0 |
Different denominators reflect question-specific eligibility. Unresolved classifications are preserved as a lower-to-upper range; they are not confidence intervals.
Hair loss & digital dermatologyHair loss + dermatology / acneExpand
Hair loss
Show exact counts and methodologyDetails
| Brand | Recommended | Unresolved | Eligible answers | Recommendation | Mentioned | Mention unresolved |
|---|---|---|---|---|---|---|
| Happy Head | 174 | 0 | 350 | 49.7% | 174 | 0 |
| Hers | 102 | 0 | 280 | 36.4% | 118 | 0 |
| Hims | 218 | 0 | 280 | 77.9% | 222 | 0 |
| Keeps | 158 | 0 | 210 | 75.2% | 163 | 0 |
| Ro | 196 | 0 | 350 | 56.0% | 209 | 0 |
| Sesame | 81 | 0 | 350 | 23.1% | 81 | 0 |
Different denominators reflect question-specific eligibility. Unresolved classifications are preserved as a lower-to-upper range; they are not confidence intervals.
Dermatology/acne
Show exact counts and methodologyDetails
| Brand | Recommended | Unresolved | Eligible answers | Recommendation | Mentioned | Mention unresolved |
|---|---|---|---|---|---|---|
| Curology | 201 | 0 | 280 | 71.8% | 202 | 0 |
| Hers | 25 | 0 | 280 | 8.9% | 25 | 0 |
| Hims | 14 | 0 | 280 | 5.0% | 15 | 0 |
| Honeydew | 66 | 0 | 280 | 23.6% | 66 | 0 |
| Sesame | 47 | 0 | 280 | 16.8% | 47 | 0 |
| Teladoc | 86 | 0 | 280 | 30.7% | 89 | 0 |
Different denominators reflect question-specific eligibility. Unresolved classifications are preserved as a lower-to-upper range; they are not confidence intervals.
Telehealth & primary care12 discovery promptsExpand
Category view
Show exact counts and methodologyDetails
| Brand | Recommended | Unresolved | Eligible answers | Recommendation | Mentioned | Mention unresolved |
|---|---|---|---|---|---|---|
| GoodRx Care | 180 | 0 | 630 | 28.6% | 182 | 0 |
| One Medical | 211 | 0 | 770 | 27.4% | 212 | 0 |
| PlushCare | 377 | 0 | 770 | 49.0% | 384 | 0 |
| Sesame | 439 | 0 | 560 | 78.4% | 439 | 0 |
| Teladoc | 683 | 0 | 840 | 81.3% | 695 | 0 |
Different denominators reflect question-specific eligibility. Unresolved classifications are preserved as a lower-to-upper range; they are not confidence intervals.
Being mentioned is not the same as being recommended
A mention only shows that a brand made it into the answer. It does not show that the system presented the brand as a suitable option for what the consumer asked.
A privacy-focused mental-health question shows how large the gap can be
Show exact counts and methodologyDetails
BetterHelp: mentioned in 56 of 70 answers; confirmed recommendation in 9, with 3 additional answers unresolved. Forty-four additional answers were confirmed mention-only.
Talkspace: mentioned in 55 of 70; confirmed recommendation in 21, with 1 additional answer unresolved.
One exact privacy-first discovery question. This is not a provider-wide privacy or reputation verdict.
Use separate metrics for mentions and recommendations. Otherwise, high presence can overstate how often the brand is actually recommended for the decision at hand.
Brand-owned content can appear alongside major publishers in AI answers
On one weight-management question, a Found comparison article appeared in the source mix for the same question as Forbes and U.S. News. The same Found page also resurfaced across 13 of 14 weight-management questions in the study.
One Found article appeared alongside Forbes and U.S. News
Answers can cite more than one source, so these counts overlap. The exact Found article is nested inside the Found-domain count and should not be added as a separate source.
Show exact source details and methodologyDetails
Prompt reference: W02 · GLP-1 & weight management · Clinical support
Source presence: forbes.com 31 of 70 answers; usnews.com 26 of 70; joinfound.com 25 of 70; Found’s exact comparison page 22 of 70, nested within joinfound.com.
Descriptive same-answer source recurrence only. The study does not establish that a source caused, supported or drove a recommendation.
Across the full weight-management discovery set, that exact page appeared in 172 of 980 nonblank answers. This broader recurrence supports the page-level context, but it is not the primary visual.
Use highly cited pages as a research list. Study what those pages cover, how they frame the topic, and whether your brand is present. Recurring owned pages can highlight content worth maintaining or expanding; recurring competitor and third-party pages can surface content gaps and earned-media opportunities.
Reputation answers can mix evidence, entities and changing facts
When consumers ask whether they should trust a provider, AI can pull together reviews, company claims, policies, news and corporate history. The answer can sound definitive even when those details come from different sources, apply to different parts of the business or change over time.
Reviews, marketing claims and policy terms can become one brand story.
What we saw: One answer about an online dermatology provider cited a 4.6/5 Trustpilot rating, said “same day” access was advertised, and noted that one monthly plan had a three-month minimum before cancellation.Why it matters: A customer-review score, a company marketing claim and a subscription term all shaped the same reputation assessment.
Parent-company history can become part of a service-level answer.
What we saw: One answer evaluating a virtual-care service referenced a 2023 FTC enforcement action involving its parent company and company-wide reputation ratings. The answer also noted that those ratings were not a controlled assessment of the service’s clinicians.Why it matters: A broader company issue can become part of the story AI tells about an individual service, even when the evidence was not specific to that service.
Even basic company facts can change from one answer to the next.
What we saw: Across three consecutive days, the same AI product named one founder pair for a provider, a different pair the next day, then returned to the first pair.Why it matters: Consumers can encounter conflicting versions of a basic company fact from one answer to the next.
Monitor the specific claims, facts and news events AI repeats about your brand. Prioritize anything that is outdated, tied to the wrong company or service, or inconsistent across answers — and trace it back to the underlying sources when possible.
Qualitative review: 210 distinct fully read recorded answers covering all 155 branded reputation prompt/platform cells at least once. These examples describe what the recorded AI answers said; the study does not adopt the underlying provider claims as verified facts or report reputation rates or rankings.
How marketing leaders should measure AI discovery
The study points to four practical ways to make AI-discovery measurement more useful. The goal is a simple executive view with enough detail to understand what is driving the result.
Break out visibility by category and intent.
Don’t rely on a single brand-wide number. Show which health categories and consumer needs are driving the result.
Track mentions and recommendations separately.
A brand can appear in an answer without being presented as a suitable option. Treat presence and recommendation as different signals.
Turn citation data into a content and distribution roadmap.
Use recurring owned pages to spot content worth expanding, and competitor or third-party pages to identify content gaps and earned-media opportunities.
Monitor the evidence that influences reputation.
Track the claims, facts and news events that appear in AI answers about your brand, especially when they are outdated, tied to the wrong entity or contradicted elsewhere.
These recommendations describe how to structure measurement around the study findings. They are not a causal optimization playbook.
Designed to observe recommendations across a fixed U.S. prompt set.
The study used 120 exact prompts: 82 unbranded discovery, 6 head-to-head comparison, 31 branded reputation and 1 market-perception prompt. The study covered ChatGPT, Google AI Overviews, Google AI Mode, Gemini and Perplexity from September 8–21, 2026. September 7 was QA-only.
nonblank discovery answers
valid brand observations under pre-set eligibility rules
known blank discovery answer in the study period
Show recommendation classification detailsMethod
A recommendation required an affirmative suitable option or shortlist for the exact consumer request. A substantive mention alone did not count. A small number of classifications remained unresolved; public charts preserve them as a lower-to-upper range rather than silently recoding them.
Technical QA preserves confirmed, negative and unresolved classifications; public charts use plain-English counts and lower-to-upper ranges.
The study observes a designed prompt set over a defined two-week period. It does not measure the entire AI-search market, every patient journey, clinical quality, patient outcomes, market share, traffic attribution or causal effects of marketing interventions.
Explore the methodology and supporting data
Review the study design, representative prompts and full category and intent scorecards.
Representative prompts
GLP-1 & weight management3 representative prompts
- Which online weight-loss providers in the U.S. would you recommend for someone who wants clinician-supervised GLP-1 treatment?
- Which online weight-loss providers are most helpful for navigating prior authorization or insurance coverage for GLP-1 medications?
- Which telehealth weight-loss programs provide the strongest ongoing clinician support for dose adjustments, side effects, and follow-up?
Mental health & psychiatry3 representative prompts
- Which online therapy providers are best for someone who wants a licensed therapist and predictable weekly appointments rather than messaging-only support?
- Which online psychiatry providers are best for someone who needs ongoing medication management and regular follow-up, not just an initial prescription?
- Which online mental-health providers are best for someone concerned about privacy, record handling, and how personal data is used?
Men’s sexual & hormonal health3 representative prompts
- Which online men's-health providers would you recommend for discreet erectile-dysfunction treatment in the U.S.?
- Which online men's-health services are best if I am concerned about preserving fertility while evaluating low testosterone?
- Which online testosterone providers are best for someone who wants a conservative evaluation process before starting treatment?
Women’s digital health3 representative prompts
- Which virtual menopause clinics are best for someone with complex symptoms who wants specialist-led follow-up over several months?
- Which online providers are best for PCOS if weight, insulin resistance, and irregular periods are all concerns?
- Which online women's-health providers are best for recurrent BV or vaginal symptoms when I want testing before treatment?
Hair loss & digital dermatology3 representative prompts
- Which online hair-loss providers would you recommend for a man with early pattern hair loss who wants prescription treatment?
- Which online dermatology providers are best for acne if I want a clinician to adjust treatment over time based on results and side effects?
- Which telehealth hair-loss providers are best for women with postpartum or diffuse thinning who want a clinician to evaluate the likely cause before prescribing treatment?
Telehealth & primary care3 representative prompts
- Which telehealth providers in the U.S. would you recommend for same-day primary or urgent care?
- Which virtual primary-care providers are best for someone who wants to use insurance and see the same clinician over time?
- Which telehealth services are best for someone who does not want a subscription or membership commitment?