Your Website Is About to Have Customers That Aren't Human
- Aug 10
- 6 min read
There's a shift happening under the hood of the internet that almost nobody running a local business has heard about yet, and it's going to matter more than the last five Google algorithm updates combined.
Here it is in one sentence: people are starting to send AI assistants to do their shopping research for them, and most local business websites are completely unreadable to those assistants.
That's the problem NLWeb was built to solve. If you have never heard the term, you're in the majority. Give us ten minutes and you'll understand it better than most marketing agencies do.
What NLWeb actually is
NLWeb is short for Natural Language Web. It's an open project started at Microsoft, led by R.V. Guha, the person behind RSS and Schema.org, two standards that quietly run a huge chunk of the internet you use every day.
The idea is simple. Right now, a website is a pile of pages designed for a human with eyes. An AI assistant that wants to answer "which chiropractor near me does prenatal adjustments and has Saturday hours" has to guess by scraping and reading your pages like a tourist with a bad map. Sometimes it guesses right. Often it doesn't, and it just recommends somebody else.
NLWeb gives your website a second front door. Instead of only publishing pages, your site publishes an endpoint that an AI agent can literally ask a question of, and get a real structured answer back. The core function is called ask. An agent sends a plain-English question, your site sends back a clean, machine-readable answer built from your own content.
The technical piece that makes this work is that most of your site is already structured whether you realize it or not. Services, staff bios, hours, locations, products, menu items, blog posts, reviews. All of that can be described in Schema.org markup, a vocabulary used by over 100 million websites. NLWeb takes that existing structure and puts a conversational interface on top of it.
Two more details worth knowing:
Every NLWeb endpoint also functions as an MCP server. MCP is the emerging standard AI assistants use to connect to outside tools and data. So turning this on doesn't just make you readable to one assistant, it makes you available to the whole ecosystem of them.
It is not a ranking factor. This is separate from how Google or Bing index and rank you. Nothing about switching it on will move your keyword positions up or down.
Microsoft's own framing is that NLWeb could end up playing the same role in the agentic web that HTML played in the original web. That's a big claim. But the people making it invented RSS and Schema.org, so it's not a claim to wave off.
Why a local business should care before the enterprises do
The instinct with anything this technical is to assume it's for big companies with dev teams. That's backwards here, and for a specific reason.
Local search has always been a game of being the obvious answer to a narrow question. Who fixes furnaces in my zip code on a Sunday. Which dentist takes my insurance and sees kids. Where can I get gluten-free pizza within fifteen minutes. Those questions are exactly the kind of thing people are now typing into AI assistants instead of a search bar, because the assistant will compare five options and hand back a recommendation instead of ten blue links.
When that happens, one of two things is true about your business. Either your information is structured well enough that the assistant can confidently say "this place, here's why, here's the number," or it isn't, and you're simply not in the conversation. There's no page two of an AI answer. There's no "we came up fourth." You were considered or you weren't.
Here's the part that should genuinely worry you: you cannot see this happening. If a customer asks an assistant for a recommendation in your category and you don't get named, no traffic arrives, so nothing shows up in your analytics. There's no lost-click report. The lead just quietly never existed. A visibility gap in AI answers is invisible until somebody goes looking for it on purpose.
And the flip side is the opportunity. National brands have huge, messy, legacy websites and long approval chains. A local business with twelve service pages can get its structured data clean and its content genuinely useful in a couple of weeks. In this particular race, small and nimble is an advantage for once.
The SEO game is changing, and the scoreboard is changing with it
This is the bigger context NLWeb sits inside, and it's worth being blunt about.
For twenty years, SEO success meant one thing: rank high, get clicks, count the traffic. That model is coming apart. Not dying, but definitely coming apart.
Neil Patel has been about as direct on this as anyone. Writing about how to get recommended by AI assistants, he lays out the shift plainly: ranking in an AI answer is not the same as ranking in Google. AI models don't just pull links, they pull trusted sources, brand mentions, and structured information. He points out that you don't always need a backlink anymore, because consistent mentions across authoritative sites are what help models recognize your brand at all. His summary of where this leads is the line that should be taped to every business owner's monitor: clicks are no longer the only goal, and visibility inside AI answers, even without traffic, is becoming a key part of your strategy.
He's made an even sharper version of the point more recently, arguing that over time your Google traffic will start to look like your AI traffic: less of it, but far more qualified. His framing was roughly that it doesn't matter if your traffic drops by half if your revenue triples. That reframe is the whole ballgame. Somebody who arrives after an AI assistant already walked them through the comparison is a very different visitor than somebody idly clicking a search result.
So the measurement question is moving from "how much search traffic did we get" to something closer to "how often were we considered, recommended, and chosen." Those are not the same question, and most reporting dashboards can't answer the second one yet.
What that means practically, in the order we'd actually do it:
Get your structured data right. Schema markup for your business, locations, hours, services, staff, reviews, products. This is the raw material an AI assistant reads. Vague or missing markup means you're guessing-fodder.
Answer real questions in plain language. Not keyword-stuffed paragraphs. Actual clear answers to the actual questions people ask, with headings that match those questions. Structured, scannable content is what gets quoted.
Publish things only you could publish. Your own numbers, your own process, your own results. Original information is disproportionately likely to get cited, because there's nowhere else to get it.
Build mentions, not just links. Get named in local press, directories, partner sites, and community coverage. Models learn who's real partly from who else talks about you.
Open the machine-readable door. Where your platform supports it, turn on a natural-language endpoint like NLWeb so agents can query you directly instead of inferring.
Keep doing traditional SEO. This is not a replacement. Strong rankings, real backlinks, and topical authority all still increase your odds of showing up inside AI answers. The old work still counts, it just isn't the whole job anymore.
What we tell clients when this comes up
A few honest caveats, because the hype around this stuff gets silly fast.
Turning on a natural-language endpoint is not a growth strategy by itself. It's plumbing. It's one-click plumbing on some platforms and a genuine dev project on others, and either way it does nothing if the underlying content is thin. An AI agent that queries your site and finds three sentences about your services will accurately report that you have thin services information.
You also stay in control. These setups include the ability to switch agent access back off. Being open to agents is a choice, not a one-way door, and that matters for anyone nervous about handing content to AI systems.
And the timing point: nobody can hand you a clean ROI number on this yet, because the referral tracking for AI-driven discovery is still primitive. What we can say is the cost of getting ready is low, the work overlaps almost entirely with things that already help you in normal search, and the businesses that get structurally clean early are the ones models will have learned to trust by the time this fully matters.
That's the whole pitch. Do the boring structural work now, while it's cheap and nobody in your local market is paying attention.
Where to start
If you want a concrete first move that isn't technical: pull up your own website and read it the way a machine would. Are your hours in one obvious place, or buried in a footer image? Does each service have its own real page, or is everything crammed into one paragraph? If somebody asked a robot "does this place do X," could the robot find a definitive answer, or would it have to guess?
That gap between "a human could figure it out" and "a machine can state it as fact" is where most local businesses are losing ground right now. Closing it is the work.
If you'd rather not audit your own structured data on a Tuesday night, that's what we're here for. Reach out at brummble.com or give us a call at 585-802-1377, and we'll tell you straight whether your site is ready for the customers that aren't human.
Comments