What it is
NLWeb is a project Microsoft open-sourced in 2025: it uses your site’s existing Schema.org structured data plus RSS to turn the site into an endpoint you can query in natural language. And each NLWeb instance is also an MCP server — wiring your content directly into the agent ecosystem.
Why it’s the same road as GEO
GEO has always been about making AI treat you as a trustworthy, well-structured, easy-to-consume source. NLWeb pushes that to its limit — instead of passively waiting for AI to crawl and guess your structure, you actively expose your organized data as a queryable interface. The work you put into Schema.org and entity consistency becomes an asset here directly.
The honest state of it
It’s an open-source experiment, not a standard, adoption is early, and it’s certainly not “install it and mainstream AI cites you.” Its value is directional: when agents start querying sites directly, the sites with structured data and a conversational interface get wired in first.
What it means for you
You don’t need to stand up NLWeb now. But it confirms the same thing again: site health (clean structured data + consistent entities) is the foundation, and new interfaces are just outlets that cash it in. Without the foundation, any interface you bolt on is hollow.
honest question for the author, does this change month to month? like do you get picked up and then quietly dropped when the model updates, or is it sticky once youre in
Both happen. Tactics-driven gains get wiped when a model updates — that's the churn you're describing. But citations that come from genuinely being a clear, trustworthy source tend to be sticky across updates. So we tell people: don't chase this month's behaviour, build the part that survives the update.
honestly half of this reads like the early seo blogs from 15 yrs ago. 'do good content, be the authority, get mentioned in trustworthy places' ... yeah we know, the hard part was always the how and that's exactly the part everyone skips
good read but the part about structured content felt a little thin. would love an actual before/after of a page that started getting picked up
the part about schema markup is slightly off. the engines aren't 'reading' json-ld the way you imply, most of them rely on the rendered text + retrieval from an index. structured data helps disambiguate entities but it's not the primary signal. worth clarifying so people don't go spend a week on schema thinking it's the magic switch
good stuff. i run a small agency and we've been quietly doing this kind of work for clients for about a year, happy to compare notes with anyone here who's experimenting, no pitch just nerding out
doesnt work for me
wait so do i need to pay for chatgpt to get my shop to show up in it?? sorry if dumb question
not a dumb q — no, paying for chatgpt does nothing for that. it's about your site/info being clear and trustworthy enough that the model picks you when someone asks. paying just gets YOU the fancier model, doesn't make it mention you.
saved this. been trying to figure out why we show up on google fine but the AI answers never mention us. makes more sense now
ok but how do you even measure if youre getting cited? like is there a tool or are people just typing questions into chatgpt all day and screenshotting
Honestly there's no single magic tool — what matters is the method: test across multiple mainstream AI engines, with several real-world phrasings, repeatedly over time. One-shot from one engine is mostly noise. We run it as continuous monitoring because doing it by hand weekly will burn you out, but the principle's the same if you DIY.
wait so do i need to pay for chatgpt to get my shop to show up in it?? sorry if dumb question im not techy, i just run a small bakery and my niece said i should look into this
finally someone says it. been telling clients for a year that getting mentioned by the ai is not the same as ranking #1 and they look at me like im speaking martian
ok but how is this any different from seo with extra steps? feels like every few years someone renames the same thing and sells it back to us
Question — does this matter at all if you're b2b and your buyers aren't asking chatgpt about niche industrial parts? feels very consumer-brand coded. genuinely asking before i spend budget here
Genuine answer: B2B is affected too, just later in the funnel. Your buyer may still email you — but they often ask an AI a round of questions first and walk in with a shortlist already formed. You're not chasing consumer-style volume; you're making sure that when they do their homework, you exist in the answer.