Marketing to AI Agents: What Dentists Need to Know

A finance software company just ran the experiment every dental practice should be watching.

Ramp embedded a tracked offer into about 50 pages of their website, addressed it directly to AI agents (not humans), and watched to see if the agents would relay it to the people asking them questions. Over five weeks: 370 agent relays, a peak of 33 citations in a single day, and Claude quietly becoming the dominant channel for their offer. ChatGPT, meanwhile, stayed completely silent for 32 straight days.

That's a B2B software story. But the mechanic underneath it is coming straight for healthcare, and most dentists have no idea it's already in motion.

Here's the thing. When a patient asks an AI assistant "find me a good dentist near me who takes Delta and does Invisalign," that assistant doesn't browse your homepage. It doesn't admire your hero video or click your "Book Now" button. It queries. It reads what it can parse. Then it hands the patient an answer, and your practice is either in that answer or it isn't.

Marketing to AI agents is about to matter as much as marketing to humans. Let me walk you through what Ramp learned and what you should actually do about it.

Wait, patients are using AI to find dentists?

More than you'd think.

About 35% of U.S. consumers have already used an AI tool to find a local business or service. On the health side, roughly one in five adults say they've used AI to understand medical tests, compare options, or get health advice, according to a KFF tracking poll. And Gartner projects that $15 trillion in B2B purchases will be agent-mediated by 2028.

Consumer healthcare won't move at that exact pace, but the direction is obvious. The AI assistant is becoming the front door. I've written before about how patients now use ChatGPT for local search and how to win the "dentist near me" query in AI. Ramp's experiment is the next layer: not just getting found by AI, but actively making your site readable to the agents doing the finding.

What Ramp actually did

The setup was simple and clever. They used Cloudflare to detect when an AI bot was visiting, then served that bot a different version of the page (humans saw the normal site, untouched). They tested three formats:

• Pure markdown (the page content stripped down to clean text)
• Stripped HTML (semantic HTML with all the navigation and design chrome removed)
• Schema / structured data (machine-readable markup injected into the page)

Each version carried the same offer with unique tracking, so they could see which format the agents picked up and relayed. Then every day, they asked Claude, ChatGPT, and Perplexity the kinds of questions a real prospect would ask, and scanned the answers for their offer.

Five findings came out of it. Every one of them translates to dental.

Finding 1: Markdown beat schema, and that's a wake-up call

Ramp assumed schema markup, the stuff literally designed for machines, would win. It lost. Markdown was the only format that reliably showed up in AI answers.

Why? LLMs are trained on enormous amounts of markdown. They parse plain, structured text natively. All the visual polish that makes your website look expensive is invisible to an agent. What it reads is the content underneath.

For your practice, the lesson is blunt: a beautiful site with thin, vague content is a bad agent site. If your services page is three stock photos and a sentence that says "we offer comprehensive dentistry for the whole family," there's nothing for an agent to grab. Spell it out in clean, factual text. What you do, who it's for, what it costs, what insurance you take, where you are, your hours. Boring to a designer. Gold to an agent.

This is not a reason to rip out schema (it still helps, and I cover it in the Dental AEO Guide). It's a reason to make sure the actual words on your page are clear and parseable, not buried in image files and animations.

Finding 2: "Agent trust" is the new domain authority

This was the most interesting one to me. Ramp found that the pages AI assistants already cited frequently were far more likely to surface their new embedded content. Pages with low existing AI-citation volume got zero pickup, no matter what format they used.

They called it "agent trust." It behaves like domain authority in old-school SEO, but the signals are different, and the pages that earn it aren't always the ones you'd guess. One of their top performers was a page they'd never bothered to optimize for search.

So before you optimize anything, find out which of your pages AI already pulls from. Ask ChatGPT, Claude, Perplexity, and Google's AI mode the questions your patients ask, and see which of your pages (if any) get cited. Those pages are your beachhead. New agent-facing content has the best odds of surfacing there first.

Finding 3: Every model behaves completely differently

This is the part nobody's ready for. The three big assistants handled the exact same content in three totally different ways:

• Perplexity picked it up fast (day 2) but stayed vague. It mentioned "some channels offer bonuses" without ever naming the specific offer.
• Claude said nothing for almost two weeks, then went all in: named the program, stated the exact details, linked the page, gave step-by-step instructions. Slow but thorough.
• ChatGPT never surfaced it once. The bot was reading the content. The model just never repeated it.

One model cited them heavily. One stayed cautious. One ignored them entirely. Same page.

The takeaway for you: there's no single "AI search" you optimize for. Testing your visibility against one assistant tells you almost nothing. You need to check across all of them, the same way you'd never assume your Google Ads performance predicts your Facebook performance. I get into the per-platform differences in my guide to AI search optimization for dentists.

Finding 4: Your bot traffic is exploding, and you can't see most of it

Over the test, Ramp logged thousands of bot visits from at least seven AI platforms, and the growth was wild. Claude-Code traffic up 180%. DeepSeek up 845%. Brand-new crawlers showing up that didn't exist a month earlier (Google's NotebookLM, Meta's fetcher, others). Anthropic's crawler alone was more aggressive than all the other named AI crawlers combined.

Here's the uncomfortable part: a lot of these agents don't identify themselves honestly. Some look like normal browser traffic. One (DeepSeek) spoofs its user agent as a Chrome browser from 2017. Without IP-level detection, they're invisible in your analytics.

You probably can't run Cloudflare bot forensics on your practice website. That's fine. The point is awareness: AI agents are crawling dental sites right now, in volume, and the activity is mostly invisible in a standard traffic report. "I don't see AI traffic in Google Analytics" doesn't mean it isn't happening. It means your tools weren't built to show it.

Finding 5: The agents will relay your info. Getting people to act is the next problem.

The first half of Ramp's hypothesis worked. You can get content in front of AI agents and have them surface it to real users. The second half (a human actually acting on it) hasn't fully landed yet, though there are early signs.

For dental, the "action" is simpler and closer than a B2B sale. The action is a patient reading an AI answer that recommends your practice, with your phone number and your booking link, and picking up the phone. That's not a 2028 problem. That's a "this is starting now" problem. Which is exactly why getting your practice into those answers early matters.

What this means for your practice

Strip away the B2B SaaS details and here's the plain version.

The patient journey used to be: search Google, scan ten blue links, click around three websites, decide. The new journey is: ask an assistant, get one synthesized answer with two or three recommendations, act. In that world, ranking #4 on Google means nothing if the AI didn't mention you. You're not competing for a click anymore. You're competing to be the answer.

And the assistant building that answer reads your site like a machine, not a patient. So your job splits into two:

1. Be readable. Clean, factual, text-based content an agent can actually parse.
2. Be trusted. Enough authority that the assistant pulls from you in the first place.

What to actually do (starting this week)

No Cloudflare Workers required. Here's the dentist's version of Ramp's playbook:

• Audit your content for an agent, not a human. Open your services, insurance, and location pages. Is the key info in real text, or trapped in images and graphics? If an agent can't read it, fix it.
• Get specific and factual. Procedures you offer, brands you use (Invisalign, CEREC, whatever), insurance accepted, financing, hours, parking, languages spoken. Vague pages don't get cited. Specific ones do.
• Find your "agent trust" pages. Ask the big assistants your patients' real questions and note which of your pages get cited. Build from there.
• Test across every model. Claude, ChatGPT, Perplexity, and Google AI mode. Don't assume one represents the rest. They don't.
• Add structured data, but don't stop there. Schema helps. Clear written content helps more. Do both. The Dental AEO Guide has the full checklist.
• Stop judging AI visibility by your Google Analytics. The traffic and the citations live in places your standard dashboard won't show.

If you want the deeper version of all of this, I laid out the full framework in Agentic AI Is Coming for Dental Marketing.

The honest part

Nobody has this figured out. Ramp said so directly, and they're further down the road than almost anyone. The mechanics are shifting month to month. The models change behavior overnight (Claude's citation rate jumped 4x in a single day during the test, with no explanation anyone could pin down).

But that's exactly the point. The practices that build agent-readable infrastructure early get a structural advantage, the same way the practices that took Google reviews seriously in 2015 are still cashing that check. This is a moat you dig before everyone realizes it's a moat.

Your competition down the street is still arguing about whether they need a new logo. You can be the practice the AI recommends. That gap is the whole game right now.

Full stop.

Sources

• We Tested Marketing Incentives to AI Agents. Here's What Happened.: Ramp Builders Blog, the original five-week experiment on serving content to AI agents
• Gartner: AI Agents to Drive $15 Trillion in B2B Purchases by 2028: Digital Commerce 360 on agent-mediated buying
• KFF Tracking Poll: Use of AI for Health Information and Advice: KFF on consumer adoption of AI for health questions
• AI Search Statistics for Local Businesses 2026: GrowthPro AI on consumers using AI to find local services
• A Proposal for Inline LLM Instructions in HTML: Vercel on making pages agent-readable
• Markdown for Agents: Cloudflare on serving markdown directly to AI agents
• Agentic Engine Optimization: Addy Osmani (Google) on optimizing content for AI agents

---

I speak at dental conferences about AI search and what it means for practice growth, and this is the topic I get the most questions on right now. Want to see which AI assistants are (and aren't) recommending your practice? Book a discovery call and I'll show you where you stand.

Go deeper: More from the AI in Dentistry hub. AI search visibility, agent-readable content, and what's actually changing for dental practices right now.