An honest look at the website work marketing leaders obsess over, what it actually does for you in AI search, and the far more consequential work it quietly distracts you from.

This question has come up in nearly every conversation I’ve had with a marketing leader over the past few months, so it felt like time to give it the answer it deserves in one place. It usually arrives already second-hand—a CEO relaying something a board member asked, a VP of marketing paraphrasing a Slack thread—and it goes something like: “Why aren’t we showing up when people search for us in AI?” 

The instinct, almost every single time, is to go straight to the website—to ask whether the site is even optimized for AI search in the first place. That instinct is right, which is the confusing part. The website is the piece you can actually get your hands on and can directly control, it’s the thing you fix first, and if it’s a mess the engine can’t even see you well enough to evaluate anything else. But the speed and certainty with which you can finish it is precisely what makes it so easy to check that box, move on to the next task within your power to control and skip the rest, and for more important, work that influences how AI engines describe and evaluate your brand.

The instinct to focus on your website first isn’t random. For two decades, the website was the lever. Search engine optimization ran on keyword matching, which meant you could tune your site to the way Google read it—the right words in the right places, structured the right way—and watch the rankings respond. Cause, effect, control, all in your hands. 

Generative search doesn’t work like that. It runs on reasoning, which is by definition messier, less predictable and far less directly manipulable than a keyword index ever was, and that’s deeply uncomfortable for anyone who built a career on a channel they could steer. The early data is already complicating the old instincts: When Seer Interactive ran 10,000 questions through ChatGPT to see what drove brand mentions, the backlinks marketers spend years chasing turned out to have little to no impact, while the broader strength of a brand’s presence across the web mattered far more. So the old playbook gets carried over and pointed at the new platform, and the website—the one thing that still feels controllable—becomes the thing marketers glob onto. 

It doesn’t help that the website is also the kind of investment that fits neatly into the way businesses are wired to think: in quarters and fiscal years, in QBRs and earnings calls, in things you can point to as done. You can finish a website. That’s exactly why it’s so easy to mistake finishing it for finishing the job. 

The question, “Is our site optimized for AI search?” obviously has a handful of technical pieces sitting underneath it, the crawler access and structured data and llms.txt files you’ve probably seen flying around your feed (a quick, jargon-free glossary is below, because none of it is as complicated as it sounds). They’re worth understanding, and they’re worth doing. They are not, however, where the battle gets won.

Here’s what sits underneath that question, in plain English:

• Crawler access: Whether you’ve let the AI bots through the front door at all; block them and you’re invisible by default.
• Structured data (schema): Labels in your site’s code that tell an engine what it’s looking at, so it reads “this is the author, this is the price, this is a verified review” instead of guessing.
• Content structure: Clear headlines and organization so an engine can lift a clean answer out of your page rather than fighting through marketing copy.
• llms.txt: Proposed plain-text file that hands an AI a map of your best content; barely used by engines today, but the standard is quietly being set.

We’re communications strategists, and our value has always been in the strategy and execution, not the technical plumbing. But the shift from digital marketing to AI search has dragged this technical layer into our world in the same way a film director may never touch his cinematographer’s camera but had better understand exactly what the lens can and can’t do. So we went and did the homework. What the homework turned up is that most of what marketers are losing sleep over down at this level barely registers.

Take llms.txt, since it’s generating the most noise of the bunch. One analysis of more than 500 million AI bot visits found a grand total of 408 that actually touched an llms.txt file. Google has said on the record that it doesn’t use the thing at all, and yet the major SEO plugins have quietly started auto-generating it across millions of sites anyway, which tells you everything about how a standard gets set in this space, with the tooling racing ahead of any evidence that it works. Put it in place, sure, because the floor needs to be there. Just don’t confuse the lobby for the penthouse suite.

All of the technical optimization in the world, no matter how clean you can make it, won’t make an AI engine “trust” you, which is the part that decides everything. This is the crucible I keep coming back to with clients: AI search is the place your brand authority gets tested in real time, melted down to whatever a model can actually verify about you and handed back to a buyer in a sentence or two, and a spotless website only buys you a ticket into the fire. 

What you’re made of once you’re in there gets decided by a completely different set of signals. Whether credible third parties cite you; whether your executives are visible and genuinely attached to the topics that define your category; whether you’ve earned the media, the reviews, the recognition and the independent research that a model keeps finding corroborated across a dozen sources it already trusts.

One analysis after another bears this out with searing clarity: When Ahrefs studied 75,000 brands, external brand mentions turned out to be the single strongest signal correlated with showing up in an AI answer, three times more predictive than backlinks and well ahead of anything happening on your own site. A controlled Stacker and Scrunch study found that distributing the same story through third-party publications rather than your own domain produced a median 239% lift in AI citations. That tracks with everything we see in our own work: The engine can parse a perfectly tagged page and still have no earthly reason to say your name.

That deeper work is the work that compounds, the work that ladders up from a business goal to a brand authority strategy to the earned coverage and executive visibility that feed it, the work that pulls on your PR, your comms and your sales teams all at once. It operates on a fundamentally different clock. The website you can influence directly—you own it, you change it, it’s done. This other work you can only influence indirectly, because it lives in other people’s coverage, other people’s reviews, other people’s sense of whether your executives are worth listening to, all of it filtered through engines that are themselves being consulted by an audience adopting AI at a growing rate across every generation. You don’t flip that on. You build toward it over 60 days, 90 days, a year, two years—a runway, not a weekend—which is precisely the timeline that sits least comfortably in a business wired to quarterly reports. You can finish the website fast and tell the board you’ve handled AI. You’ll have done no such thing. You’ll have built the ground floor and left the building. 

So pay the price of admission, get the technical house in order, and then spend your real energy on the work that changes whether you get chosen—because the brands AI puts forward aren’t the ones with the tidiest schema, they’re the ones that spent years building an authority a model can find everywhere it looks, and that authority eventually shows up on the only scoreboard your CEO is actually emailing you about: whether the next buyer already had you on the shortlist before they ever raised a hand.