Highlights
- The dataset: This study is an analysis of over 130,000 AI source citations using prompts specific to the mental health care industry. It analyzes AI responses from four AI models run in 12 countries over the course of Q2 2026.
- Reddit makes up 20% of ChatGPT's citations: Its single most-cited source, by a 15x margin over its next domain.
- ~84% local citations in Sweden, ~32% in Spain: Localization swings hard by market, not just by model.
- 12% average overlap between any two models: Grok and Google AI Overview are the closest pair at 33%, the highest agreement we have recorded.
- 27% of Microsoft Copilot's citations are from institutional domains: Public-health authorities, NHS trusts, and regional health services, making Microsoft Copilot the most public-sector-heavy model.
Which sources should you target to get cited as a mental health care brand?
The top-cited domains show how differently the four models source their answers. Each model's five most-cited sources, with their share of that model's total citations:
| Model | Top cited sources | Citation character |
|---|---|---|
| Grok | reddit.com (6.8%), psychologytoday.com (4.5%), yelp.com (4.1%), zorgkaartnederland.nl (1.7%), m.yelp.com (1.5%) | Reviews + community + directories |
| ChatGPT | reddit.com (19.8%), aslcittaditorino.it (1.3%), priorygroup.com (0.8%), allbiz.mx (0.8%), redunitas.com.ar (0.7%) | Reddit + niche clinics |
| Microsoft Copilot | riseclinic.ca (1.8%), doctoralia.com.mx (1.8%), vard.skane.se (1.4%), zorgkaartnederland.nl (1.4%), aslroma1.it (1.3%) | Local directories + public health |
| Google AI Overview | google.com (11.8%), psychologytoday.com (2.4%), instagram.com (2.2%), 1177.se (1.8%), facebook.com (1.6%) | Own results + social + directories |
ChatGPT is the most concentrated, nearly a fifth of everything it cites is Reddit, and no other source breaks 1.5%. Google AI Overview cites its own results page (google.com) more than any external domain. Microsoft Copilot and Grok are flatter, spreading citations across hundreds of directory and clinic sites where no single domain dominates. A provider optimizing only for ChatGPT is absent from roughly 95% of the domains Grok and Google AI Overview prefer, but a listing on a strong national directory or Psychology Today can earn visibility in both Grok and Google AI Overview at once.
How have AI source rankings changed over time in the mental health care industry?
We split the ~16-week window at its midpoint (2026-05-04) and ranked the top domains by their share of citations within each half. We read the movement as suggestive rather than conclusive.
The pattern is stark: the review and appointment-booking platforms that anchored the first half, Yelp, Jameda, the TopDoctors network, Doctolib, fell out of the top 25 entirely in the second half, while their slots filled with public-health authorities (NHS trusts, Italy's ASL sites, Sweden's 1177.se) and google.com itself. We observe an institutional shift in the citation mix; we do not claim the underlying models re-ranked.
What type of content do AI models cite for mental health providers?
We classified every cited domain by category. Commercial sites lead overall, as expected for a service industry, but mental health stands apart in one way: institutional content, public-health authorities, NHS trusts, regional health services, is a major citation category, especially for Microsoft Copilot.
Two contrasts stand out. First, Grok is the community model, 28.1% of its citations are user-generated content (Reddit, Yelp, Facebook), with ChatGPT close behind at 23.7%, while Google AI Overview (9.6%) and Microsoft Copilot (6.5%) sit far lower, and Grok cites the most editorial content too (9.0%). Second, Microsoft Copilot and Google AI Overview are the institutional models, roughly a quarter of their citations go to public-health authorities and health-service sites (NHS trusts, Sweden's 1177.se, Italy's ASL regional health authorities). That institutional weight is a mental-health signature you rarely see in categories like dental or hotels. For a provider, public-sector and directory presence matters far more here than in most industries, because two of the four models lean on it heavily.
Do AI models cite local-language content for mental health providers?
Using the source's country-code top-level domain (ccTLD) as a proxy for local-language content, we measured the share of citations pointing to a local-ccTLD source in each non-English market. Spanish-language markets were scored against their own national ccTLD (.ar.mx.es).
Across all non-English markets, 61.9% of citations point to a local-ccTLD source. But the average hides a sharp split: Northern-European markets cluster high (Sweden 84%, Netherlands 83%), while Spanish-speaking markets sit lower (32-51%). The likeliest explanation is structural, not behavioral, Sweden, the Netherlands, and Germany have dense national directories and public-health portals, whereas Spanish-language mental-health sites frequently publish on .com and pan-regional domains. Treat the Spanish figures as a floor, not a true localization rate: Spain's low mark reflects that many Spanish therapists surface via .com directories rather than .es sites.
Which AI model relies most on local sources for mental health providers?
Holding the market constant, the models differ in how local they go. The heatmap shows each model's local-ccTLD citation rate by language (all cells meet the n≈30 sample threshold; Google AI Overview produced too few French and Dutch citations to report). Spanish combines Argentina, Mexico, and Spain.
Microsoft Copilot is the most aggressively local across the board, 88% in Dutch and 94% in Swedish, consistent with its heavy reliance on national directories and public-health authorities. Grok is the most globalized in Spanish (41%), where it reaches for pan-regional review platforms, but climbs to 77-81% in Dutch and Swedish. ChatGPT lands in a steady band (48-75%), and Google AI Overview is uneven, going very local in Sweden (91%) but reaching for international sources in Italian (43%). The country drives localization more than the model does, but within a country Microsoft Copilot reaches for local directories while Grok and Google AI Overview reach more often for global platforms.
Should mental health providers optimize for each AI model separately?
Mostly yes, but less absolutely than in other categories. We built each model's ranked list of its top-20 most-cited domains and measured how much each pair holds in common. The average overlap is 11.5%, ranging from 2.6% to 33.3%.
The spread is real: Grok and Google AI Overview agree on half their top-cited domains, while ChatGPT shares only Reddit with either of them. ChatGPT is the outlier that drags the average down, it agrees with no other model on more than two domains, while Grok and Google AI Overview form a surprisingly aligned pair built on shared directory and reference sites.
How many sources does each AI model cite per answer?
The models differ enormously in how many sources they cite per answer. Grok is by far the most source-hungry; ChatGPT the most economical.
Grok cites roughly 5x as many sources per response as ChatGPT. For a provider, that cuts two ways: Grok offers far more citation slots to compete for, but each individual citation carries proportionally less weight; ChatGPT cites few sources, so earning one of its handful is both harder and more valuable. ChatGPT produced 3,945 responses but cited sources in only 1,112 of them, most ChatGPT answers cited nothing at all. Google AI Overview is unusually source-generous here at 12.3 per response.
Context
This report covers how four AI systems, ChatGPT, Microsoft Copilot, xAI's Grok, and Google AI Overview, cited web sources when answering mental-health-provider questions between March 10 and June 29, 2026. The data spans 12 markets (Argentina, Australia, Canada, France, Germany, Italy, Mexico, the Netherlands, Spain, Sweden, the United Kingdom, and the United States) and seven languages.
The findings describe which sources AI models cite, not which providers, clinics, or treatments are best. A high citation share means a domain is frequently surfaced by a model, not that it is authoritative, accurate, or clinically sound, an important caveat in a health-sensitive category. Citation behavior is also a moving target: models update their retrieval and ranking continuously, so a snapshot like this captures a period, not a permanent state.
Methodology
How we measured this
We analyzed every AI response and its cited sources for the mental-health-provider category over the observation window: 12,223 responses and over 130,000 cited source citations. A "citation" is a web source a model explicitly surfaced in support of an answer; we count only sources flagged as cited.
For each model we ranked the domains it cited most and measured how much any two models' top-20 lists have in common, an overlap rate of 11.5% means that of all the distinct domains two models name between them, only about one in nine appears on both. We classified every cited domain into a content category (commercial, user-generated, institutional, editorial, reference) to see what kind of source each model trusts. To gauge localization, we used the source's country-code top-level domain (ccTLD) as a proxy for local-language content and measured how often each model cited a source on the country's own national domain; this proxy undercounts local content published on generic .com domains, so the localization rates, especially in Spanish markets, are conservative. Model coverage varied across the observation window, so temporal movements are reported as relative ranks within each half. Finally, we counted cited sources per response per model. All reported per-model and per-language cells meet the n≈30 minimum, and the headline gaps were confirmed with two-proportion z-tests.
Frequently asked questions
Do different AI models cite the same sources for mental-health questions?
Not much, but more than in most industries. Any two models share only about 12% of their top-20 most-cited domains on average, so roughly 88% of the sources one model relies on are absent from another's top tier. The exception is Grok and Google AI Overview, which share half their lists (33%). There is a shared spine of directories and reference sites that two of the four models both lean on.
Which sources does each model favor?
Grok favors community and review platforms (Reddit, Yelp, Facebook) plus national directories. ChatGPT is dominated by Reddit, nearly a fifth of all its citations, plus a thin layer of niche clinic sites. Microsoft Copilot funnels into local directories and public-health authorities. Google AI Overview cites its own results page most, then Psychology Today, social platforms, and national health services.
Does language matter for getting cited?
Yes, but the market matters more than the model. In Sweden and the Netherlands, roughly 83-84% of citations go to local-ccTLD sources; in Spain, only about 32% do. Part of that gap is that Spanish-language providers often publish on .com and pan-regional directories, which our method counts as non-local, so those figures are conservative.
Which model is most likely to cite a local provider's website?
Microsoft Copilot. It reaches 88% local-ccTLD in Dutch and 94% in Swedish, and it leans heavily on national directories and public-health portals. If your priority is being cited as a local mental-health provider, Microsoft Copilot is the most receptive system.
Why does public-health and institutional content matter so much here?
Because two of the four models lean on it. Microsoft Copilot (26.5%) and Google AI Overview (22.8%) send roughly a quarter of their citations to public-health authorities, NHS trusts, and regional health services, a pattern you rarely see in other local-service categories. In mental health, being listed or referenced by a public-health portal is a genuine path to AI visibility.
Which model cites the most sources, and does that help me?
Grok cites about 29.9 sources per response, roughly five times ChatGPT's 5.9. More sources mean more slots to compete for in Grok, but each carries less weight. ChatGPT's economy of citations, and the fact that most of its answers cite nothing at all, makes each of its few slots both harder to win and more valuable. Google AI Overview is unusually generous here at 12.3 sources per response.
Is being on Reddit worth it for a mental-health provider?
For ChatGPT, enormously. Reddit is ChatGPT's single most-cited domain at 19.8% of all its citations, about 15 times its next source, and it ranks #1 for Grok too. User-generated discussion is a real path to AI visibility, but mainly for the models that lean on it; Microsoft Copilot and Google AI Overview barely cite it.
What should a mental-health provider do with this?
Optimize per model, but exploit the overlap. Presence on major national directories and Psychology Today can earn you visibility in both Grok and Google AI Overview at once, a rare two-for-one in AI visibility. Beyond that, treat each model as a channel: a strong Reddit footprint helps with ChatGPT, review platforms help with Grok, and directory plus public-health-portal presence helps with Microsoft Copilot and Google AI Overview.
