Last updated: Aug 2026 · Reviewed by Aaraddhya Bhatalkar, Co-CEO, New Resilience.

The short version: When someone asks ChatGPT, Claude, or Gemini for help finding treatment, the assistant runs a live web search, reads a handful of pages, and answers with a short list of named centers. You make that list for two reasons: the tool can find clear, factual pages about you, and other sites on the web say the same things about you. Most treatment center sites fail the first test because their pages are written to reassure rather than to answer, and they miss the second because few other places mention them consistently. Both are fixable, and almost no one in this field is working on them yet.

What people are actually asking

Families and referrers have started asking assistants full questions instead of typing keywords into Google. The questions are specific:

• "What are the best inpatient rehabs in Arizona that take Aetna?"

• "Where can my son go for dual diagnosis treatment near Denver?"

• "Is there a women's-only program for trauma and addiction in the Southeast?"

• "What should I ask when choosing a detox center?"

The answer comes back as a few named places, usually with a line on why each one fits. If you are not in that short list, the family never sees you. There is no second page to scroll to. AI answers name fewer options than a page of search results, so the cost of being left out is higher.

The data on this is still early, but the centers already paying attention are seeing it in their numbers. One of them put it this way:

"We are up to about 20 to 30 percent of our online referrals coming from ChatGPT and other AI search engines, and that number is only rising. What matters more is that those referrals convert about twice as well. The families have already done a fair amount of the clinical and financial research on their own, so our admissions team does not have to spend as much time on it."

Smaller programs see it too. A ten-bed program told us ChatGPT had brought in about four clients over the past year, and that they still could not explain why it kept surfacing their name. A residential program counted three leads in six months and admitted one, with two of those people volunteering on their own that ChatGPT was how they found the place. An admissions coordinator at another program explained why the ones who arrive this way tend to convert:

"When you ask ChatGPT something, you have to ask a specific question, so the people who come through it are more qualified."

A channel that is growing and converting better than the others is worth building now, while the rest of the field is still ignoring it.

How the tools actually answer a question

There are two ways a model can mention your center. The first is from memory. The model was trained on a snapshot of the web, so it may already know you exist if enough pages mentioned you before its cutoff. That memory is stale and hard to influence on any useful timeline. The second is from live search, and that is the one you can win this year.

When an assistant has web access, it turns the person's question into a few search queries, pulls the top results, reads them, and writes its answer with citations to the pages it used. ChatGPT's search, Google's AI Overviews, and Gemini all sit on top of a web search index, so the pages that already rank are the pages the assistant gets to choose from. Ranking gets your page considered. Being clear is what gets it quoted. For that, the tool has to open your page and lift a clean answer out of it.

Why treatment center sites get skipped

Most treatment center websites are hard for a model to use. Four reasons come up again and again.

The words that matter are missing. To match "inpatient dual diagnosis, in-network with Aetna, near Scottsdale," a model needs those facts written as text. Many sites talk about meeting you where you are and walking beside you on your journey, and never plainly state the level of care, the conditions treated, the insurances accepted, or the city.

The facts are trapped in design. Levels of care drawn as a graphic, insurance shown as a row of logos, the phone number inside a widget. People read those fine. Crawlers read almost none of it.

There is no plain summary. No clear list of programs, no FAQ, nothing that says in one place who this is for and what happens next.

The site disagrees with the rest of the web. The name, locations, and programs on the homepage do not match the Google Business Profile, the directories, or the review sites. When sources disagree, a model trusts none of them and picks a center whose story is consistent.

The two things that get you cited

Two things decide whether you get named.

First, clear pages you control. Write each important page the way a person asks the question. State the conditions you treat, your levels of care, the insurances you accept and whether you are in network, your locations, what admission looks like, and what a stay involves. Put it in real text with headings, lists, and an FAQ, and add the behind-the-scenes labeling, called structured data, that tells machines what each part is. The aim is to state the specific facts a family names, in language a model can copy without guessing.

Second, consistent mentions elsewhere. A model trusts a fact more when several independent sources agree on it. Your Google Business Profile, the directories, review sites, and any press should all say the same things about you. Reviews carry extra weight. They are recent, they name specifics, and they come from outside your own marketing. This is where an alumni program pays off twice. Engaged alumni leave reviews and tell their stories, and those stories become the outside corroboration a model leans on when it decides who to name. The same alumni work that brings back referrals also builds the evidence that makes an assistant comfortable recommending you.

What GEO actually is

This work has a name. Generative Engine Optimization is making sure AI assistants can find you, understand you, and repeat you correctly. It shares a foundation with SEO, because the tools still run on search underneath. It adds three things SEO never cared about: writing pages a model can quote, keeping your facts identical everywhere they appear, and earning the reviews and mentions that give a model the confidence to say your name.

How to tell if it is working

You can measure this directly, which surprises people.

Ask the assistants yourself. Open ChatGPT, Claude, Gemini, Perplexity, and Google, and type the real questions your patients ask. See whether you appear and whether the facts are right. Do it once a month. Most centers have never done it once.

Watch for wrong facts. A model will state the wrong insurance or the wrong location with total confidence if the web is inconsistent about you. Finding and correcting those is half the job.

Expect attribution to be indirect. No family will tell you ChatGPT or Gemini sent them. You see it as more people arriving already knowing your name, and you confirm it by asking every inquiry how they heard about you.

Do this, avoid this

Do

• State the specifics in plain text: conditions treated, levels of care, insurances accepted and in-network status, locations, and what admission involves.

• Publish real insights and data: your outcomes, what you do differently, anything only you can say. Original sources get cited. Brochure copy does not.

• Write an FAQ in the words families actually use, and answer each question directly.

• Earn steady reviews from alumni and families. They are recent, specific, and come from outside your marketing, which is what a model trusts.

• Add structured data that labels your programs, locations, and FAQs for machines.

• Keep every fact identical across your site, your Google Business Profile, and the directories.

Avoid

• Locking facts inside images, graphics, or logo strips. Crawlers cannot read them.

• Leading with feeling and skipping the specifics. "We meet you where you are" tells a model nothing.

• Letting your listings disagree with each other. Conflicting names, locations, or programs get you dropped.

• Relying on your homepage alone. Models often pull from reviews and directories, not you.

• Confusing more with better. Specific and accurate beats long.

• Expecting to pay your way in. There is no ad slot in an AI answer.

If you do nothing else, fix the facts on your five most important pages and turn on reviews from alumni and families. It is earned, it moves slower than paid search, and it compounds. If you want help doing the work, that is what we do. You can book a short demo.