For most of 2025, the marketing conversation about Google’s AI search was collapsed into a single phrase: “AI Overviews.” The AI-generated summary that appears above the classic blue links absorbed all the attention, all the reporting, and most of the panic.
Treatment center CMOs asked how to get into AI Overviews. Consultants sold AI Overview optimization. Search Console rolled out AI Overview impression data.
That framing has quietly stopped being accurate. Google now runs AI Mode as its own standalone conversational search surface, alongside the AI Overviews that sit on top of classic results.
The mistake I keep watching operators make is treating AI Mode as an extension of AI Overviews, a bigger version of the same feature. It is not. It is a separate search surface with its own citation logic, its own URL selection behavior, and its own reporting implications.
The data that made this concrete for me was an Ahrefs analysis of September 2025 US data that found AI Mode and AI Overviews cited the same URLs only 13.7% of the time for the same query.
Most citations show up on one surface but not the other. That is not a rounding error. It means a treatment center website could be optimized well for AI Overviews and still be effectively invisible to a family running the same query through AI Mode, or the reverse.
This piece is the operator’s read on why the two surfaces diverge, what that divergence means for a behavioral health website specifically, and the specific moves that put a rehab site into the citation pool on both surfaces rather than one.
Key Takeaways
- AI Mode and AI Overviews are two distinct Google search surfaces with only 13.7% citation overlap for the same query, per Ahrefs, meaning a page cited in one can be invisible in the other for the same query.
- AI Overviews pulls citations from queries that could be answered with a paragraph or two of consolidated information, favoring pages with strong classic ranking signals and answer-first formatting.
- AI Mode fans a single query out into multiple sub-queries and citations, favoring pages that satisfy multi-faceted compound prompts and often surfacing sources that never made the classic top ten.
- Treatment center sites need a two-track optimization strategy: reinforcing the existing pages that already win AI Overview citations while also building compound-prompt pages that give AI Mode multiple sub-query hooks to cite.
- Reporting has to split into two lanes: AI Overview citation share and AI Mode citation share need to be tracked separately, because gains and losses on one surface do not predict the other.
What AI Overviews is actually doing behind the scenes
AI Overviews is the surface most treatment center operators have already spent a year adapting to. It is the AI-generated summary that sits above the traditional blue links on about 39% of US desktop searches as of June 2026, per Comscore data reported by Search Engine Land.
The Overview pulls from a set of cited sources, typically three to six URLs, and presents a consolidated answer to the searcher’s query.
The selection logic behind AI Overviews is closer to a hardened version of classic ranking than most operators realize. Pages that already rank in the top ten for the query are heavily favored.
The Overview tends to select sources that lead with a direct answer to the query in the first paragraph, have clear semantic structure (H2s, FAQ blocks, definition patterns), and carry credible authority signals for the topic (author bios, editorial review markers, first-party expertise).
For a treatment center, the pages that tend to win AI Overview citations are the pages that were already winning classic organic traffic.
Cornerstone service pages, well-structured level-of-care explainers, glossary and definition entries, and long-form guides with strong FAQ blocks. If your top-ten organic pages are structurally clean, you are already partway to AI Overview eligibility.
DEFINITION
AI Mode is Google’s deeper conversational search surface that generates responses using Gemini-family models and a broader multi-source retrieval pattern than AI Overviews. It cites different URLs, weighs authority differently, and often surfaces on multi-step or comparative queries where AI Overviews stays silent.
What AI Mode is actually doing, and why it is different
AI Mode is the standalone conversational search experience Google launched globally in April 2026. It looks less like classic Search and more like ChatGPT with a Google logo.
A searcher types a question, or a compound question, and AI Mode returns a conversational answer with citations, then allows follow-up questions in the same session.
The mechanic that makes AI Mode different from AI Overviews is what Google internally calls query fan-out. A single user query does not get answered by a single retrieval.
AI Mode decomposes the query into multiple sub-queries, retrieves sources for each sub-query, and synthesizes the whole thing into one answer.
A user asking “what is the difference between residential and PHP for someone with dual diagnosis” is not answered by one search.
It is answered by five or six sub-searches (definitions of both levels of care, common criteria for choosing between them, dual diagnosis modifiers, cost considerations, insurance implications), each of which pulls its own set of citations.
That fan-out mechanic is why the low citation overlap is not surprising in retrospect. AI Overviews is answering the query. AI Mode is answering five or six adjacent sub-queries and stitching them together.
The pages eligible to be cited on one surface are only sometimes the same pages eligible to be cited on the other.
The concrete divergence a rehab operator sees on the same query
Run the query “how long does inpatient drug rehab take” through both surfaces and the difference becomes visible. AI Overviews will typically return a consolidated answer citing three or four established sources: a large chain website, a directory, maybe a health authority, and possibly one mid-authority behavioral health blog.
The same query run through AI Mode returns a longer conversational answer citing eight or ten sources. Some of those citations overlap with the AI Overview set. Many do not.
The AI Mode citations tend to include more mid-authority operator sites, more clinical explainers, more insurance and cost pages, and more compound-prompt content that would never have ranked in the classic top ten for the seed query but does satisfy a downstream sub-query the fan-out generated.
That divergence has a specific implication for a behavioral health website. Winning AI Overview citations rewards the pages you already have working well.
Winning AI Mode citations rewards content architecture, cluster density, and compound-prompt page design.
A treatment center site that has six or seven pages in a cluster addressing every angle of a topic (definition, cost, timeline, criteria, insurance, outcomes) gets substantially more AI Mode citations than a site with one strong page on the seed query.
AI Mode is a cluster-density surface. AI Overviews is a ranking-strength surface.
Why this changes rank reporting for treatment centers
The most common reporting mistake I have watched over the last quarter is treating AI visibility as a single number.
A client CMO asks “how are we doing in AI search?” and the answer they get back is an aggregate impression count from Google’s new Generative AI performance reports, or worse, an aggregate share-of-voice number from a third-party monitoring tool.
Aggregates hide the picture. A site can be losing AI Overview citation share while gaining AI Mode citation share, and the aggregate will look flat.
Another site can be holding on AI Overviews but leaking AI Mode citations to competitors with denser cluster architecture, and the aggregate will look healthy right up until admissions volume drops.
The reporting shift for a behavioral health site in the second half of 2026 is to split AI visibility into two lanes. Track AI Overview citation share (or the classic Search Console impression proxy for it) as a rank-adjacent metric that responds to page-level ranking work.
Track AI Mode citation share as a cluster-density metric that responds to compound-prompt content and cluster architecture. Report them side by side, and stop trying to average them into a single number.
OPERATOR INSIGHT
AI Overviews and AI Mode are two different citation surfaces, and treating them as one optimization problem produces weaker results on both. The two-track split: AI Overviews wants concise, structured, snippet-friendly content with named-author credibility. AI Mode wants deeper compound-prompt pages that answer multi-faceted questions in one place. A rehab site running both tracks in parallel produces citations neither optimization alone would generate.
What the research shows
Several independent studies measured clicks on queries with AI Overviews. They differ in method and period, so read them as a range.
What the click studies found
8% vs 15%
Share of visits with a click on a traditional result, with vs. without an AI summary (Pew, 2025)
58%
Lower CTR for the top-ranking page on keywords with AI Overviews (Ahrefs, Dec 2025)
+120%
More organic clicks per impression when a page is cited in the AI Overview (Seer, 2026)
Pew Research Center. Pew’s July 2025 analysis tracked 900 US adults’ Google searches in March 2025. Users clicked a traditional result in 8% of visits with an AI summary, versus 15% without, and ended their browsing session more often after a page with a summary.
Ahrefs. Ahrefs’ February 2026 update compared desktop Search Console data from December 2023 and December 2025 and found a 58% lower average CTR for the top-ranking page on keywords with AI Overviews. Its April 2025 study had measured 34.5%.
Seer Interactive. Seer’s November 2025 update found a 61% decline in organic CTR on informational queries with AI Overviews from mid-2024 to September 2025. Its April 2026 update, using a larger dataset and reported by Search Engine Journal, found the 2025 decline didn’t extend into 2026, and that being cited in an AI Overview meant 120% more organic clicks per impression.
Google. Google’s Liz Reid wrote in August 2025 that total organic click volume from Search to websites was relatively stable year over year, and that clicks were slightly higher quality. Google didn’t publish data behind the statement.
DEFINITION
Zero-click query. A search that ends without a click to any website, because the results page answered the question or the searcher gave up. AI Overviews can raise the share of zero-click searches for informational questions, while Seer’s 2026 data found AI Overviews on far fewer commercial and transactional queries.
A diagnostic you can run in Search Console
Run these six checks in order.
Split branded and non-branded
Google’s branded queries filter separates searches for your facility’s name from everything else. A branded drop is a demand or reputation question; a non-branded drop is a visibility question.
Group similar queries
Google’s query groups cluster related searches, which makes it easier to see which topics lost clicks.
Segment by page type
Compare blog posts, program pages, location pages, and insurance pages separately. The pattern is often concentrated in one group.
Check the results page
For the queries that lost the most clicks, search them yourself, logged out. Note whether an AI Overview appears and whether your site is cited.
Line up the dates
Compare the drop against the num=100 change, Google’s update dates, and your own site changes.
Follow the outcome
Check whether calls, forms, and admits from organic search fell with the clicks. If they held, the lost clicks weren’t the ones producing admissions.
Our guide to Search Console’s AI performance reporting covers the reports in more detail.
How to re-weight a content program
If the diagnosis points to AI Overviews absorbing informational clicks, the response isn’t to write more informational posts.
Shift effort toward action pages. Program pages, insurance pages, admissions pages, and location pages answer the questions families ask once they’re deciding. Our guide to the nine types of content that drive admits covers which formats tend to lead to inquiries.
Make informational posts worth citing. Seer’s 2026 data found pages cited in an AI Overview earned far more clicks per impression than those that weren’t. Clear, sourced, self-contained answers are the citation candidates. Our guide to getting cited in AI Overviews and ChatGPT covers the patterns.
Connect informational posts to the next step. A post on signs of alcohol use disorder should link clearly to the facility’s program and admissions pages, so the readers who do click have somewhere to go.
Refresh before you write. Updating posts that still earn clicks often does more than adding new ones. Our content refresh framework covers how to choose.
Measure the right outcome. Sessions were always a proxy. Report organic inquiries and admits by landing page, so a drop in blog sessions is judged against what it actually cost.
OPERATOR INSIGHT
Before reacting to a traffic drop, pull organic admits for the same period. If admits held while blog sessions fell, the content program is working where it matters and the lost clicks were likely from readers further from a decision.
If admits fell too, the diagnostic above tells you where to look, and the fix is usually on program and location pages rather than the blog.
The two-track optimization strategy for treatment center sites
The optimization work for a treatment center site under this framing splits into two parallel tracks, each targeting a different surface.
The AI Overview track is the closer cousin to classic SEO. The moves that improve AI Overview citation eligibility are the moves that improve classic ranking.
That means clean semantic structure with clear H2s and answer-first paragraph openers, FAQ blocks with schema, strong internal linking from high-authority pages, author bios with named clinical credentials, and freshness signals through periodic updates.
Ahrefs’ study of 17 million citations found AI-cited content is 25.7% fresher than pages ranking in organic Google results, so a refresh cadence on cornerstone pages is table stakes. The May 2026 Core Update reinforced that freshness signal at the classic ranking layer, which feeds AI Overview eligibility downstream.
The AI Mode track is different work. What AI Mode rewards is cluster density and compound-prompt coverage.
That means building content architectures where a single service topic is covered by six or seven distinct pages, each addressing a different sub-query angle (what is X, how much does X cost, how long does X take, who is X for, what happens after X, X vs Y).
It means writing pages that themselves satisfy compound prompts, with structural depth across multiple facets rather than one narrow angle.
And it means internal linking that binds those pages together into a topic silo the AI Mode fan-out can pull from at multiple points.
The Full AI Search Stack for Treatment Centers covers the technical scaffolding for both tracks. What the two-track view adds is the reporting discipline to see whether each track is actually producing citation gains.
The pages a rehab site should already have on each track
Working backward from the two surfaces, there is a page inventory that puts a treatment center into the citation pool on both.
For AI Overview eligibility, the pages that carry disproportionate weight are the cornerstone level-of-care pages (residential, PHP, IOP, detox, telehealth), the modality pages (medication-assisted treatment, DBT, CBT, EMDR, trauma-informed care), and the definition/glossary content that answers “what is X” queries.
The payer and financial pages sit alongside these (insurance verification, cost of treatment, out-of-network reimbursement). These pages need to be well ranked, well structured, and freshly updated.
For AI Mode eligibility, the layer above those pages matters. That is where cluster hub pages, compound-prompt guides, and cross-topic explainers do their work.
A residential treatment cluster that includes only one page on residential is going to lose AI Mode citation share to a competitor whose residential cluster includes seven pages covering criteria, cost, timeline, family involvement, aftercare planning, and comparison against other levels of care.
Cluster architecture is the leading indicator of AI Mode share.
The internal linking discipline binds them together. Every cornerstone page needs to point down into its cluster spokes, and every spoke needs to point back up to the cornerstone plus laterally to sibling spokes. That is the wiring AI Mode fan-out follows.
What the reporting layer needs to catch up on
The Search Console Generative AI performance reports that launched in June 2026 give some of this reporting for free, but the current version has limits.
The reports show impressions inside generative AI features but do not currently break AI Overview impressions apart from AI Mode impressions inside the aggregate view. That split will presumably come, but as of the current writing, you have to piece it together manually.
For a treatment center CMO who wants the two-track reporting now, the practical workaround is a monthly manual sample. Pick twenty high-intent queries that map to your admissions funnel. Run each of them through both AI Overviews and AI Mode.
Log which sources got cited on each surface, whether your pages made either list, and what percentage share of citations you captured relative to your competitor set. It is manual work, but it produces the two-lane visibility the aggregate reports do not.
Third-party brand monitoring tools that track citation share inside ChatGPT, Perplexity, Claude, and Gemini also provide a useful cross-check. The overlap between those platforms and Google’s AI Mode is imperfect but directionally correlated.
What the reports actually show, in plain language
The Generative AI performance section lives inside the same Performance area of Search Console you already use. You will see two report types there.
The first is Search, which covers AI Overviews (the AI-generated summary that sits above traditional blue links on about 39% of US desktop searches as of June 2026, per Comscore data reported by Search Engine Land) and AI Mode (the standalone conversational search experience Google launched globally earlier this year).
The second is Discover, which covers AI-generated summaries surfaced inside the Discover feed on Android.
For a treatment center, Search is where the operational value lives. Discover matters for brand-side content teams but is a smaller signal for admissions-driven SEO.
Each report shows you four filterable views. Impressions is the count of times a URL from your site appeared inside a generative AI feature. Pages tells you which URLs got cited. Countries breaks the impressions out geographically.
Devices splits mobile from desktop from tablet. That is the full data set. Google has been explicit that clicks, CTR, and queries are not currently available and gave no timeline for when they might be.
How to actually find the reports in your Search Console
Open Performance in Search Console and look for the section labeled Generative AI features. Since August 31, 2026, every property has it.
The opt-out control lives in a different place: Settings, then Search generative AI. If your team has multiple properties, check it on each one before you read the numbers, because a property that has been excluded will show no AI impressions at all.
Once you have access, the top-line view shows total impressions across generative AI features over your selected date range.
From there, you can filter to isolate AI Overviews impressions specifically, or filter down to a single URL, country, or device to see how a specific page is performing inside AI-driven surfaces.
The date range picker works the same way as the classic Search Console performance report, but the report only has data from around its launch forward, so most properties will have limited historical depth for another year.
The four questions a treatment center operator should be asking of these reports
Once you have access, the reports become useful only when you ask them the right questions. I use four when I open the report for a client site.
The first is: how big is my AI impression footprint relative to my classic Search impression footprint? There is no published benchmark for that ratio yet, so track it against your own baseline month over month rather than against someone else’s number.
An AI footprint that stays tiny relative to classic impressions suggests a citation problem, either technical (schema, structured data, crawl issues) or editorial (content that does not answer the compound prompts AI Mode fans out into sub-queries).
The second is: which pages are being cited, and do those pages match the intent I want to be cited on? A common pattern I see in rehab sites is glossary and definition pages absorbing a disproportionate share of AI Overview impressions.
That is not a problem in itself, but if your service pages and location pages are not showing up at all, your citation footprint is aimed at informational intent rather than conversion intent, and that misalignment will show up in call volume before it shows up in traffic.
The third is: what is the device split on AI impressions? If your AI impressions skew noticeably more mobile than your classic Search impressions, the AI surfaces are reaching a different slice of your audience than your blue-link results, and mobile page speed and mobile FAQ visibility deserve more attention.
That has implications for how you think about mobile page speed and mobile FAQ visibility, both of which the Full AI Search Stack for Treatment Centers covers in more depth.
The fourth is: is my AI impression trend line moving in the same direction as my classic Search impression trend line? If classic impressions are up and AI impressions are flat, your pages are ranking for classic queries but losing the citation eligibility inside generative surfaces.
If AI impressions are up and classic impressions are down, you are winning citation share while losing rank, which is the modern pattern most YMYL sites are living inside now.
WATCH OUT
The opt-out toggle in Search Console removes your pages from AI Overview and AI Mode responses entirely. Some sites are considering it out of frustration with the missing click data. For behavioral health, opting out is a strategic mistake. AI Overviews alone appeared on about 39% of US desktop searches by June 2026, according to Comscore, so opting out removes you from a large share of the results page. Opting out sacrifices future traffic for present-day reporting purity.
The opt-out toggle: why no behavioral health facility should use it
Google added a control inside the new report that lets you block your content from AI Overviews, AI Mode, and Discover’s AI features. The toggle does not affect your normal Search rankings. Opting out forfeits any AI traffic and impressions entirely.
A few details from Google’s documentation for the control matter for multi-site operators. Including your site is the default. Child properties, such as a URL-prefix property for one facility’s section of the site, inherit the setting of their parent unless someone overrides it. A change generally takes a few days to fully apply. And the control does not affect AI training, which is handled separately through Google-Extended.
I want to be direct on this: for a treatment center, this toggle is a trap dressed up as a privacy control.
The impulse to opt out usually comes from watching zero-click behavior climb and reasoning that if AI features are keeping searchers from clicking through, blocking them will restore the click. That reasoning collapses on contact with the data.
A page that is not cited inside AI Mode does not send those searchers back to the classic blue links. It sends them to whichever competitor did get cited. You give up your citation footprint without gaining a single traditional click in exchange.
For a family in the first hour of researching addiction treatment, being cited by AI Mode as one of the three or four facilities the assistant recommends is worth substantially more than a click that never comes. Opting out removes you from that consideration set entirely.
The only defensible reason I can think of for a behavioral health facility to hit the opt-out toggle is if compliance counsel has flagged a specific hallucination risk on your entity, and even then, an AI Information page is a better tool than a blanket surrender.
How to fold AI impressions into monthly reporting for the operator or CMO
The reports will absorb some of the space in your monthly SEO report that classic Search impressions currently occupy. My recommendation is not to replace one with the other, but to run them side by side, with an explicit narrative about what each metric is actually measuring.
Classic Search impressions still tell you about traditional ranking. That number is still useful, even as CTR compresses.
AI impressions tell you about citation eligibility inside the surfaces where increasingly more high-intent searchers spend the first phase of their research. Together, they draw a fuller picture of your site’s visibility in the modern search environment.
The KPI shift I have been recommending across our client book is to add AI citation share as a first-class KPI alongside sessions, phone calls, and admits.
It does not replace admits as the north star. It becomes the new leading indicator that sits two steps before the admit, in the same slot classic organic sessions used to occupy. For how that KPI fits the rest of the plan, see our complete guide to behavioral health marketing.
What to change this quarter
The specific moves that reflect the two-surface reality inside a treatment center’s SEO program over the next 90 days fall into three tracks.
Audit your existing cornerstone pages for AI Overview eligibility (semantic structure, FAQ blocks, freshness).
Build out cluster density around your top three revenue-driving service topics (six to seven pages minimum per cluster, each addressing a distinct sub-query angle).
Split your AI reporting into AI Overview citation share and AI Mode citation share as separate lanes rather than a single aggregate number.
The KPI shift I have been recommending across our client book is to add both metrics to the monthly report alongside classic organic sessions, admissions calls, and admits.
Position 1 CTR is compressing. Aggregate AI visibility is a lagging summary. The two-lane citation share numbers are the leading indicators that tell you where the top of your funnel is actually moving.
Frequently Asked Questions
Why do AI Mode and AI Overviews cite different URLs for the same query?
AI Overviews consolidates one query into one consolidated answer with a small number of citations, favoring pages with strong classic ranking signals for that specific query. AI Mode decomposes each query into multiple sub-queries through its query fan-out mechanic, then pulls citations for each sub-query independently and stitches the answers together.
The result is that AI Mode ends up citing a broader set of pages that satisfy any of the sub-queries, including pages that would never have ranked in the classic top ten for the seed query.
An Ahrefs analysis of September 2025 US data measured citation overlap between the two surfaces at 13.7%, meaning the citation pools are largely distinct even for identical seed queries.
Do I need to write different content for AI Mode versus AI Overviews?
The base structural moves overlap. Both surfaces reward clean semantic structure, answer-first paragraph openers, FAQ blocks, credible author bios, and freshness signals. Where the two surfaces diverge is in what they reward at the site-architecture level.
AI Overviews rewards individual page strength. AI Mode rewards cluster density. A site that publishes one strong page per topic will do well in AI Overviews and poorly in AI Mode.
A site that publishes six or seven pages per topic covering different sub-queries will do well in AI Mode and no worse in AI Overviews. The economical answer is to build clusters, which serves both surfaces.
How do I measure AI Mode citation share separately from AI Overview citation share?
Google’s new Generative AI performance reports in Search Console show combined impressions inside generative AI features but do not currently split AI Overview impressions from AI Mode impressions cleanly. That split will presumably come but is not available as of the current writing.
The practical workaround is a monthly manual sample. Pick fifteen to twenty high-intent queries that map to your admissions funnel, run each through both AI Overviews and AI Mode, and log which sources got cited on each surface. It is manual work but produces the two-lane visibility that the aggregated reports do not.
Should I be worried if my classic organic traffic is dropping but my AI impressions are climbing?
Not necessarily. That specific pattern (classic organic down, AI impressions up) is the modern pattern most mid-authority YMYL sites are living inside now. Position 1 CTR on classic Search has compressed from 27% to 11% over the past two years, and much of that compression reflects clicks being absorbed by AI Overviews and AI Mode.
The metric that actually matters for a treatment center is what happens downstream.
If your AI impressions are climbing while admissions calls are flat or dropping, the gap is usually in name recognition (do families remember your facility as a name to call after seeing you cited), phone number visibility on the page a searcher does click through to, or admissions call quality. It is rarely a reporting error.
Trevor Gage is Director of Marketing at Webserv, a behavioral health marketing agency working with residential, outpatient, and telehealth treatment providers across the United States. He leads Webserv’s SEO, content, and AI search practice. If you want the two-track AI citation audit against your own site, start with a Visibility Gap conversation.
