A Beverly Hills breast surgeon set Google Ads to zero. Then AI assistants sent her 60 new patients in four months.

• 60

  new patients who told the front desk they found her through an AI assistant

  68 inquiries, 60 booked · May to Aug 2026 · intake log

• $1.5M

  in booked procedures from those 60 patients, at her average case value

  60 × $25,000 · four months

• 41×

  monthly Google clicks, Sep 2025 vs Aug 2026

  447 → 18,405 · Search Console

• 17×

  monthly visits arriving from an AI assistant

  ≈5 → 88 · GA4, a floor (see method)

• #1

  most-cited surgeon website in her market's AI answers

  63 citations vs 28 for the next practice · 4 Sep 2026 scan

• $31k+

  spent on Google Ads over four years. In her words, nothing worked.

  her Ads account · set to zero January 2026

Average case value of $25,000 is her figure; individual cases vary. Patients named an AI assistant at intake but the specific one was not recorded, so the 60 cannot be split between ChatGPT, Google AI Mode and the others.

Before: a name people had to already know

New patients came from two places: referrals, and the TikTok and Instagram content she makes herself. The website contributed almost nothing. From September 2025 through February 2026, drkilleen.com averaged 17 to 27 Google clicks a day, and most of them were people typing her name. The top three queries that autumn were "kelly killeen md", "kelly killeen" and "dr kelly killeen".

Visits arriving from an AI assistant ran at roughly five a month. Five.

Over four years the practice put more than $31,000 into Google Ads. It never ran steadily. Spend went up when the practice needed patients and got switched off when the leads turned out to be spam, sometimes for six months at a stretch. The pattern is the verdict: nobody pauses a channel that works. Her summary is shorter: "Nothing worked."

In January 2026 the practice set Google Ads spend to zero. It has stayed there.

Search Console counts organic clicks only, and the account was off for the entire climb below. Nothing in these curves was paid for.

What changed

In early 2026 the site was rebuilt the way AI assistants read sites, using the same method Visibled now runs for every client:

• Procedure pages written to be quoted. Direct answers, structured data, plain statements of who she is and what she does, on every page an assistant might land on. Nothing a visitor would notice.
• Her own videos turned into pages that answer the questions patients ask after surgery. Stitch abscesses, seroma fluid, fat necrosis after reduction, belly-button problems after a tummy tuck. Each one started as a TikTok she recorded herself and became a page a worried patient at 11pm finds through Google or ChatGPT.
• Weekly measurement across nine AI surfaces. ChatGPT, Google AI Overview, Google AI Mode, Perplexity, Claude, Gemini, Grok and two GPT models, with every answer stored so a claim can be checked against the exact response it came from.
• Search Console and Analytics wired into the same dashboard, so AI referrals and Google performance are read from her own accounts, not estimated.

Google Ads spend was set to zero in January 2026, before any of the growth below, and has not been turned back on.

The first AI-sourced patients were not caught by any tracking. They mentioned it in consults. The web form had no "how did you hear about us?" dropdown at the time, so the practice added one. Every number in the next sections comes from that dropdown.

After: the Google curve broke upward and did not stop

The ads went to zero in January. The inflection is visible to the month: March 2026 doubled the daily rate, April tripled it again. By August the site was getting more Google clicks in a single day than it used to get in a month, with the ads account still at zero.

Who those clicks are

In the last 90 days, only 575 of 6,733 clicks across her top 1,000 queries contained her name. The other 91% were strangers with a problem: "stitch abscess", "spitting sutures", "fat necrosis after breast reduction", "how to help a seroma reabsorb". The single most-visited page on the site is now a post about suture abscesses after surgery, with 6,699 clicks and 363,038 impressions in 90 days.

That is the same page an AI assistant reads when a patient asks the same question.

After: patients arriving from ChatGPT

Google Analytics records the referring site when a visitor clicks through from an AI assistant. Those sessions went from a handful a month to 88 in August 2026. Of the 345 AI-referred visits in the twelve-month window, 331 came from ChatGPT.

AI visitors behave differently

Google's new traffic lands on blog posts; it is informational. AI referrals land on the pages that book surgery. In the last 90 days, the top AI landing pages after the homepage were the Preserve, tuberous breast correction, augmentation, reduction and lift procedure pages. A person ChatGPT sends to a surgeon has already been told which surgeon.

Key events are the actions her Google Analytics property is configured to count. The rate is reported as measured; it is not adjusted for the referrer gap described above.

Where she stands in AI answers today

The 4 September 2026 Visibled scan asked nine AI surfaces 650 questions a prospective patient in her market would ask. She was named in 34 answers, on every one of the nine surfaces, at an average list position of 5.5, with a sentiment score of 90 out of 100. Her website was cited 63 times, more than any other surgeon's site in the market; the next practice site was cited 28 times.

Prompts where an assistant put her first:

• "most reputable surgeon for complex breast revision"ChatGPT · position 1
• "best breast reduction surgeon beverly hills"Gemini, GPT-5.6 Luna, GPT-5.6 Terra · position 1 on all three
• "Who are the best plastic surgeons for complex breast revision in Beverly Hills?"Gemini · position 1
• "Who are the best breast reconstruction specialists in Beverly Hills for someone recovering after a mastectomy?"Gemini · position 1; Google AI Overview · position 3
• "best breast augmentation surgeon Beverly Hills"Claude · position 2
• "Which top plastic surgeons in Los Angeles specialize in complex breast revision?"ChatGPT · position 2; Perplexity · position 3

The number that matters: 60 patients

Between May and August 2026, 68 people contacted the practice and answered the intake question "how did you hear about us?" with an AI assistant. Sixty of them became patients. That is a booking rate of 88%, nearly nine in ten, from people who arrived already told which surgeon to call.

At her average case value of $25,000, 60 patients is $1.5 million in booked procedures in four months.

For scale: Google Analytics counted 272 visits arriving from an AI assistant with a referrer in those same four months. The two numbers are not expected to reconcile. Most assistant traffic opens without a referrer and lands in Direct, which is why the Analytics count is treated as a floor throughout this study. The 68 come from patients saying so, not from a tracking pixel.

In her words

She was never the skeptic. She already used AI daily in clinical work, so she assumed patients would get there too. What she did not expect was the speed. The objection this study answers is not "will it work" but "how long will it take", and her answer is months, not years.

Before this, where did new patients come from, and what did that cost you?

Patients still primarily come from referrals and my social media, but we had zero web presence. People weren't finding us through Google at all, unless they searched directly for me. We spent over $31k on ads trying to bring people in. Nothing worked.

When did you first realize patients were asking ChatGPT instead of Google?

I had no idea that they were, until during a couple of consults people let us know. We didn't have a dropdown for "how did you hear about us?" on the web form, so we quickly updated that to capture future patients.

What made you skeptical this would work?

There wasn't really any skepticism. I was already using AI in medicine, primarily OpenEvidence. So I assumed it would eventually work, I just didn't realize how quickly. I've always heard SEO takes forever, but this worked better than expected.

What actually changed on the site? What surprised you about the process?

The site itself looked the same. It was all of the plumbing behind the scenes that got improved. Beyond that, we added a lot of new content, sourced directly from many of my own original TikTok videos I use on social media.

What are the numbers now?

68 in the last 4 months, with 60 of them having booked.

What would you tell another surgeon who is invisible to AI today?

I don't really want any other surgeons to know, haha. But really, don't think that it's too late to start. The fact you are even thinking about doing this now means you are ahead of the curve.

What has this changed about how you think about marketing?

I put more thought into my social media content now. I do everything myself, there's no team behind me. I know that it will get doubled up and used on my website. It keeps it fresh with new content and helps with every new post.

Method and what this study does not claim

• Google numbers are from the Search Console API for the property sc-domain:drkilleen.com, trailing 366 days ending 4 Sep 2026, rolled up by calendar month. Search Console finalizes data with a three-day lag, so the window ends three days back.
• AI-referral numbers are GA4 sessions whose referrer matched a curated list of assistant hostnames. They are a floor. Many assistants strip the referrer, and Bing/Copilot is deliberately excluded so ordinary Bing search is not swept in.
• AI answer numbers are from the Visibled scan completed 4 Sep 2026: 650 tasks across nine surfaces, 647 answers analyzed, 2,864 citations across 523 domains. Every count links to the stored answer it came from.
• The Visibled visibility score is not used as a before-and-after here. Scanning began on 30 May 2026, after the site rebuild, so the scans capture her position after the change, not before it. The prompt set and the model lineup also grew substantially over the summer, so scores from June and September are not like-for-like. The before-state is carried by Search Console and Analytics, which reach back a full year.
• Attribution. The site rebuild, the content, the ad shutoff and the general rise of AI search all overlap in time. This study shows what happened to her site's numbers; it does not claim a controlled experiment.
• Inquiry and patient figures come from a "how did you hear about us?" dropdown on the practice's web form. The dropdown was added after the first AI-sourced patients mentioned it in consults, so inquiries before that date are not counted and the 68 is an undercount. The specific assistant was not recorded. The 88% booking rate and the $25,000 average case value are her figures.
• Quotes are her written answers to the interview questions, dated 9 Sep 2026, edited for spelling and length. Nothing was added.

Sources: Google Search Console API (sc-domain:drkilleen.com), Google Analytics 4 Data API (property 253278280), Visibled scan cmtmfhzlx0318sd0mr03lzxny, all pulled 6 Sep 2026; practice intake log and Google Ads account, as stated by the practice, 9 Sep 2026.