YMYL SEO in the Age of AI Search: Why Expert-Backed Content Wins Where Trust Is Non-Negotiable

AI search advice is everywhere right now, and it's mostly written for SAAS. I'm exploring what it takes to rank and be recommended in categories like health/wellness and finance.

TL;DR

• Most of the AI search advice in your feed is built for SaaS. Health, finance, and the other YMYL categories play by different rules, on purpose, because Google holds them to a higher trust bar.
• The biggest sources in these categories are institutions, and they aren't going anywhere. Rewriting what they already published is a losing game.
• What you CAN publish is the thing only you have: your own data, your internal expert’s point of view, your own methodology. That's what AI has a reason to cite.
• Domain size doesn't always predict AI citations. A smaller, expert-backed brand can win the specific questions its expert answers better than the giants bothered to.
• Start with owned, expert-backed content, the layer you control. Then, amplify it with earned media, social distribution and off-site communities.
• Done right, one piece of content does three jobs: it ranks, it gets found by AI, and it builds the brand. Win-win-win.

Search "benefits of NAD+" right now. Go ahead, I'll wait.

What surfaces in the AI answers is a mix of hospitals, big health publishers, and (depending on the day) a couple of YouTube videos. Look closely, and you'll spot a wellness brand's own blog sitting in that lineup now. That one's a former client of ours, and I'll come back to how it got there in a bit.

But here's what that SERP is really telling you. In categories like health and finance, traditional and AI search are looking for true experts to cite. That article that's ranking? It wasn't just another run-of-the-mill generic blog post. It was an information-packed, POV driven piece from their Chief Medical Officer.

That's the kind of informational content that stacks up against medical journals, big brands, and high-authority publishers.

I've spent the last decade creating content across a range of brands, and I've watched a LOT of AI search advice roll through my feed lately. Almost none of it fits the categories that I've been working with — categories that impact a person's health, financial stability, or well-being.

It's written for SaaS: get on G2, build comparison pages, go win some Reddit threads. It's great advice, but the playbook isn't exactly the same for every industry. So I went and did some research to show you how to win in YMYL SEO.

• What YMYL means for SEO now (and why AI raised the stakes)
• Why YMYL SEO is different from SaaS SEO
• Where AI search pulls from in health and finance
• Does domain authority matter for AI citations?
• What earns AI citations in YMYL content
• Owned vs. earned content for AI search (and where to start)

What YMYL means for SEO now (and why AI raised the stakes)

Quick refresher, because we're going to be using this term a lot. YMYL stands for "Your Money, Your Life." It's Google's label for content that can affect someone's health, money, safety, or well-being. It comes from the Quality Rater Guidelines, and the current read on it is less about the topic itself and more about harm: could a wrong answer here actually mess up someone's life?

YMYL isn't a ranking factor, per se; it's more like a category. What it does is raise the bar for the E-E-A-T (experience, expertise, authoritativeness, trust) your content has to exhibit. A generic post about a medication gets judged more harshly than a generic post about the best hiking backpack. Makes sense, right?

And AI search made that level of trust matter even more. The systems answering your customers' questions are extra cautious around the categories where being wrong is unacceptable. So the trust bar you've been raising for Google (or haven't) is even more pronounced when you're dealing with AI.

Why YMYL SEO is different

If you sell project management software, AI has a bunch of shortcuts for figuring out whether your product is any good: G2 reviews, Capterra, a Reddit thread where people discuss the pros and cons. Then there are the comparison pages stacking you against five competitors. A model can build a solid answer about your category without an expert, credentialed human involved.

Health and finance have their own versions of that, to be fair. Healthgrades, Zocdoc, Vitals. Bankrate and NerdWallet on the money side. Plenty of healthtech companies are sitting on G2 right now, and for "which platform should we buy" questions, that layer works exactly like it does in SaaS.

But watch what happens when the question changes. Reviews and directories can tell someone which clinic takes their insurance, or which app has the smoothest onboarding. They can't tell them what actually clears hormonal acne, or when it makes sense to test hormone levels in perimenopause. Star ratings don't contain clinical knowledge. So the second the question shifts from "who should I go with" to "will this actually work for someone like me," the strategy needs to follow.

You can watch the difference play out in the citation data, too. DeltaV looked at 25,337 AI citations across eight brands and found listicles pulling 61% of citations in B2B services and 0% in healthcare. The format that dominates one category shows up nowhere in the other.

And this isn't a side feature you can ignore. Google's AI Overviews show up on roughly 82% of health queries, so for most health searches, the AI-generated answer is what the reader sees first.

Google's been drifting even further this way on the traditional side too. Search Engine Land's read on the March 2026 core update, backed by SE Ranking volatility data, was that it raised the bar for ranking: strong brands, owned data, and direct query value won, and intermediaries now look increasingly exposed.

So you can't write a stack of helpful-ish posts and bolt on "AEO" as some optimization pass at the end. Especially in YMYL, the expert has to be involved from the get-go, or you've got nothing worth citing to optimize in the first place.

Because when AI needs a real answer in these categories, it goes looking for whoever generated the knowledge. The NIH. Mayo. The CDC. Peer-reviewed research. Those are primary sources, and they are not going anywhere.

BUT what if your brand showed up there instead? Most don't, and here's why.

Think about what most brand blogs in this space actually bring to the table right now. A post explaining a condition, or a roundup of what the research says. A "what you need to know about X" explainer. All of it is a rewrite of what the institutions already published. It's secondary content, competing directly against the source it's summarizing.

That's a losing battle, and the proof is in the data (keep reading).

Where AI search pulls from in health and finance

So, where IS AI search pulling from? I gathered the data by vertical, because the answer shifts depending on where you’re looking. (Legal's a different beast, by the way. A handful of directories like Avvo and Super Lawyers own nearly the entire citation layer there, so I'm keeping this to health and finance.)

Where AI cites in health: institutions, YouTube, and where you fit in

Health is the most cautious vertical there is. Surfer's analysis of 36 million Google AI Overviews found health answers leaning on the names you'd expect, like NIH, Healthline, Mayo Clinic, and Cleveland Clinic. Video is doing more work than most people realize, too. YouTube alone accounts for nearly a quarter of all citations across every vertical Surfer looked at.

This tracks with the DeltaV study we mentioned earlier. And you know what else? They tracked how often AI reads a source versus how often it actually uses one.

On diabetes queries, the engines read Healthline, WebMD, and Medical News Today constantly and then mostly didn't cite them. They cited the primary sources: CDC, Mayo Clinic, and NIH. Mayo alone got cited more than five times as often per read as WebMD did.

WebMD has a physician review board. It is not a content farm. But when an engine has the summary and the source sitting side by side, it takes the source.

And if WebMD can't win on rewritten clinical content, your blog post covering the same ground probably can't.

So what CAN you publish that isn't mostly regurgitated from someone else? Your own opinions and findings. The outcome data from your patient population. The protocol your team developed. Nobody can write that ahead of you, because nobody else has it. You're not competing with the NIH on what the NIH published. You're the original source.

And I’d argue that if you’re publishing content that contains facts and findings from an expert (like a Chief Medical Officer) that has also contributed to published studies, your brand has a better chance of getting cited for related content. For example: If your expert has a published study on NAD+, and you’re publishing an NAD+ cluster, branded content has a leg up against others that don’t.

Where AI cites in finance: affiliates, the business press, and the gaps they leave

Finance looks a little different. Goodie's analysis of 109,000 finance citations found NerdWallet is the closest thing to a single source of truth across ChatGPT, Gemini, Claude, and Perplexity, with CNBC, Forbes, and Investopedia right behind it. Conductor's benchmarks put NerdWallet at 10.14% of citations in financial services and Bankrate at 8.47%.

Here’s the number in that Conductor study I’d underline: article content was by far the most-cited page type, with over 110,000 pages pulled into finance AI Overviews. Those sites pull in original research and expert findings. Can your finance content say the same?

And keep this in mind: a lot of those affiliate roundups nail "best high-yield savings account." They're thin on the specific, advisor-level guidance a real planner gives a real client. If that guidance is sitting inside your firm, that's your way in.

Does domain authority matter for AI citations?

OK OK, I can hear it now. "But the sites you mentioned are HUGE behemoths with a TON of domain authority, of course they're going to be cited."

And the big boys DO dominate, but it looks like domain authority has less weight when it comes to AI answers. I can’t help but wonder if it’s just that there’s less credible, original data to be found elsewhere.

Surfer just ran the numbers on this, and the answer is pretty stark. They pulled 20,000 prompts and about 5 million citation URLs across AI Mode, AI Overviews, ChatGPT, and Perplexity, then tested citation likelihood against three different authority metrics. All three came back at essentially zero. Slightly negative, actually. Their takeaway: domain authority doesn't appear to have meaningful impact on whether AI cites you. Which is good news, because content is something anyone can compete on.

AND — smaller sites are legitimately in the mix. AI will cite a page with barely any backlinks when it's current and answers the question in the best way, because relevance to the specific question is doing more work than link equity.

Worth knowing that different sources behave differently here.

Google's AI Overviews cite constantly, including on informational queries. That's what all that Surfer data is measuring. ChatGPT is a different animal: ask it something like "what are the benefits of NAD+" and it'll just answer from what it already absorbed, no sources at all. It goes hunting for live sources when the question is specific enough, fresh enough, or contested enough that what it already knows doesn't cut it.

So specificity isn't a nice-to-have — it’s essential. Broad explainer content earns you AI Overview citations and builds your brand as a known entity in the category. The specific, only-you-have-it content is what gets pulled when someone asks a real question in chat.

So no, you're not beating Mayo on "what is NAD+." But you can be the source AI cites on the specific question your credentialed expert answers better than Mayo has. That's a fight you arguably can win, and most of your competitors haven't even shown up for it.

Now, I'll be straight about what we're seeing for broader citations: top brands DO concentrate at the top. Kevin Indig's research on how AI picks its sources found the top 30 domains in a topic soak up about 67% of citations, and that pages ranking first in Google were 3.5 times more likely to be cited than pages below rank 20.

What earns AI citations in YMYL content

So, of course, if you’re one of the domains we mentioned earlier, you’ve got it in the bag. But what if you’re not?

Three things keep surfacing: having a named author with real credentials, original data or first-hand experience a model can't fabricate on its own, and a complete answer to the actual question. Andy Crestodina at Orbit Media has landed on a similar short list, and Surfer's read on YMYL specifically is that verified authorship, citations from credible studies, and visible credentials are what help AI tell a trustworthy source from a less trustworthy one.

And you know what’s cool? The signals that AI is rewarding are very similar to the ones your SEO has been telling you to implement for years. It’s the same E-E-A-T work that's won YMYL rankings since the Medic update.

Here's how I'd think about what one good piece of content can do for you now:

• Job one: it ranks. Resource content pulls traffic, builds topical authority, and does everything it's always done. SEO is still a HUGE piece of the traffic pie, so yay.

• Job two: it gets discovered. This is the one people miss. When ChatGPT goes looking for something it doesn't already know, it searches Google to find it. Backlinko proved this with a great little test: they made up a fake word, published a page about it, blocked every crawler except Google's, then asked ChatGPT what it meant. ChatGPT Plus found it and quoted the page word for word. Only Google could see that page.

• Job three: it’s more likely to get cited. Being discoverable and being worth citing are two different things. That's the primary-source layer: the data, credentials, and specificity that make an engine pick your page over the summary sitting next to it.

YMYL content teams can brief and source their content this way: go specific enough that the answer isn't already sitting in the model's memory, and build the piece on primary material your own experts and data can supply.

You’re not manufacturing authority out of thin air. It's already sitting in your expert's head. The job is getting it onto a page where AI can find it, which is exactly what a good SME interview does: it pulls out the clinical observation only a practitioner would make, the position only someone who's run the numbers would take.

One honest caveat, because I don't want you walking away thinking "publish a stat, get cited." Kevin Indig has a sharp piece showing what you need to do to get this data actually cited. You need more than the stat itself; you need a clean method, real sourcing, and the honesty to name what the numbers don't prove. The data's the easy part. The rigor is the moat.

Owned vs. earned content for AI search (and where to start)

I'm not going to tell you earned media doesn't matter for AI search. It really, really does. But owned content is still important for all the reasons I’ve covered here.

Picture your authority in two layers. There's what you control directly: your site, your blog, your LinkedIn, your YouTube. And there's what you don't: press, third-party mentions, someone citing you in their own article. The owned layer is the easier one to create and edit (you own it), and it's where your expertise enters the public record to begin with. The earned layer is slower and harder to get.

There's a finding buried in that DeltaV data that reframed this for me. They scored how much each type of source gets trusted once AI actually pulls it up, and brand-owned domains converted better than editorial sites, reference sites, UGC, and even institutional sources. Their conclusion: AI doesn't distrust brand websites at all. Brands just don't get retrieved because often their content is regurgitated from other sources.

That's a completely different problem than the one most people think they have. Your site isn't disqualified. It's not the source, usually because there's nothing on it a model couldn't get somewhere upstream.

So in YMYL, you start with the owned layer. It's your highest-leverage move, partly because it's the one you can act on *now*, and partly because the trust bar in these categories is so high that you need the expert-backed foundation in place before anyone external is going to corroborate you anyway.

Build the expert-authored hub, with the material only you have. Then push it out to the surfaces AI already trusts, which is what gets you retrieved. This is the Authority Engineering approach.

Which brings me back to that wellness brand ranking for "benefits of NAD+." One set of content, built around one credentialed practitioner, did both jobs: organic traffic climbed 154% in 90 days, and the AI citations started showing up almost right away.

Your takeaway

Take a product-focused, resource-driven question a buyer actually asks you, the one that comes up on basically every sales call. Drop it into ChatGPT and Google's AI Overview and read who gets cited, and how they answer it.

Then take that exact answer to your in-house expert and ask: what's missing here? What would you tell an actual customer/patient/client in this situation?

Whatever they say back is the content worth publishing, the specific, only-you-would-know-it stuff no model can generate on its own.

The brands showing up in AI answers across health and finance didn't outspend Mayo or out-authority NerdWallet. They got specific, on their own turf, with a real person's expertise behind the words. If that person is already on your team, you're closer than you think.

Finding the expert and turning what they know into content AI actually cites, is the work I do with fractional clients in health, wellness, and finance. If your best expertise is stuck in someone's head instead of out on the web where it counts, let's change that.

Have an expert on your team you're not putting to work? Let’s talk.

FAQs