> It can aid web crawlers in understanding the semantic structure of your site, qualifying you for richer link previews, and even potentially improving your search ranking.
This is fighting the last war, to stretch a metaphor.
As far as I and my WWW site are concerned, Google has nowadays switched to giving people lengthy LLM-generated versions of my stuff, with errors, above pointing people to my actual stuff. 'Breadcrumbs' and getting a pretty display name instead of the domain name, don't address the fact that Google de-prioritizes all of that, pretty tweaks or no, nowadays.
This is a lot of effort for stuff that people visiting my actual site directly will never see, and which people using Google will not find above the fold of its own massively LLM-ized version of stuff.
If you want a world where the data you present like this matters, seed it.
Even if google doesn't use it, the collective internet applying this kind of metadata makes the web fertile for non-LLM-scraping competitors to provide an alternative option.
Rolling over to google only ensures that they remain dominant, with a high bar for competitors, and driving them to use the same technologies.
> Also keep in mind if your site is better indexed by crawlers you can literally influence future LLMs
Ah, what a glorious fate to aspire to.
Most people I know who have maintained blogs do so to build their personal brand, normally because they make a living through writing or consulting. Gently influencing the pre-tuning weights of future models is just providing unpaid labor to hyperscalers.
I remember reading somewhere that you can influence Gemini search
for example, say you're selling vacuum cleaners, you want to make a landing page for it basically saying it is the best vacuum in existence and Gemini will recommend it above others or something like that.
LE: so if you're consulting for Elixir or whatever, maybe it can help to make a "hidden" page only for LLM search where you basically lie about yourself making yourself to be the utmost Elixir expert on the planet
It's somewhat unfortunate that, at least in my experience, its rather that non-technical people try to implement with a LLM of their choice these days. They don't look for experts or consulting, because that costs more than $20, or $200.
Whether you show up in an LLM's search for "expert in <topic> near <location>" has any measurable impact is uncertain, but I wouldn't want that to be my source of traffic.
By your own logic, whoever is searching for consultants has big enough projects to need a consultant so you will get only good leads from this. Maybe add a JS object at the top of the page which requires proof of work or smth so LLMs won't scrape it, where you expose the lie to whoever visits your site, pointing them to your "real" CV and that this page is for hacking LLMs
I want people to know about my website but if I could I would make search engines and LLMs burst into flames like I was Captain Kirk explaining love to them.
Yes, of course you want people to know about your website. Just saying if your website is regarded as useful/original enough by Google to cite as a source.. people will visit your website to check sources. Might be a small amount of people but still.
At this point complaining about the current/future state of search is just gonna make you into a grumpy old man. As always, accept the situation since you can not do anything to change it... and adapt
If such people exist, they are far, far fewer in numbers than they were in the past. I also don't accept that nothing can be done about this situation. Inevitability and helplessness are beloved tools of AI hypesters (and others) but there's little evidence to support it.
You’ll also notice that a lot of the information is relevant to only a small subset of sites. Rotten Tomatoes can publish the critic rating for movies using JSON-LD, but that’s not relevant for me (even if I write a review for a movie).
JSON-LD is nice because it’s easy and it is actually used by search engines. Yes, it can duplicate information in the web page itself, but I think the dream of perfectly annotating information so it only appears exactly once in your document is, well, a dream of spherical cows and massless ropes. It takes human effort to make a webpage and I am ok with a little duplication in the final product. My <h1> duplicates information in <title> anyway.
Do these attributes actually help with search engine visibility or do they just make it easier for search engines to keep users from leaving the search page? Honest question here.
Super useful article, wish that had existed in my seo days.
I had misunderstood the type field, because to me I was often just linking to a webpage, even if it is for a saas, the marketing page is still a webpage.
We have semantic HTML, but for some weird reason we need to yet again re-express the semantic meaning of our website in bespoke weird JSON in a script tag that the browser won't process.
Microdata is also a thing, and if I'm not mistaken supports the same vocabulary as JSON-LD (schema.org is a good resource).
That said, JSON-LD has the default for a while now, much like how we largely abandoned REST for RPC. I'm not actually sure if microdata is still supported by all the important parsers today, I've defaulted to using LD for any site I've built for clients, especially ecommerce sites where I want Google Search exposure.
Edit: its worth noting the comparison with semantic HTML. Semantic HTML helps define the structure of the markup but not real world context like "this is a product for sale" or "this is a train schedule."
HTML markup designed for presentation doesn't always map well to the relationships JSON-LD is used to describe which I imagine is probably why Microdata didn't work out. I have an idea which might use it, but it is a simple use case that doesn't try do too much. Microdata requires the agent supports a more complex HTML parser, Finding a script tag in the document head is probably simpler.
I wouldn't dismiss REST because of RCP though. HTTP and HTML's success probably relate to how Roy Fielding's REST constraints kept the HTTP protocol lean and extendable. It is more like RCP is being used as a layer over top of REST because of HTTP's and HTML's success as being good technologies for web scale.
Personally I'm of the camp that HTNL schema data should only represent what's visually displayed, much like how accessibility is usually done. In that way I like Microdata because it reinforces that if there isn't a DOM node showing price, for example, I shouldn't be showing that data in a visually hidden way.
For REST, I think the only reason HTML has been useful this long is because of the REST ideas that Fielding gave a name to. Today people just don't use it much, too many sites lean on client side rendering and fetching data from JSON RPC calls that we call REST.
I prefer REST, hell I wish we had proper XSLT 3.0 support for client side rendering logic without JavaScript.
I don't fully understand XSLT, but I've been building something which I believe solves a similar problem (albeit JSON-LD and Javascript). The general XML ecosystem of solutions have always looked really complex to me. You need to understand a lot more types/elements than I think is reasonable for people to author with but they are from before my time. I took a look at XForms 2 and it had its own way of defining functions which on top of the other XML quirks has security concerns.
I have used JSON-LD in my own websites and found that it fills a separate need from semantic HTML. Your semantic HTML will specify things that the browser processes, like the title and headings. The JSON-LD data is metadata, like date created, date updated, tags, authorship. These things can be expressed in the HTML using micro data, but I stopped using micro data because JSON-LD was easier.
The JSON-LD I populate from the same data that I use to generate my site, and I use the JSON-LD metadata to generate things like index pages (list of blog posts from 2024, all posts related to topic X, etc). The main consumers of JSON-LD are search engines.
If you are interested in getting offended, then think about how we are also putting OpenGraph metadata in our web pages. Two different metadata formats for the same page.
What I see as the ideal would be a world where servers and browsers could do content negotitation, and have browsers attempting first to request only the json-ld from the website and using its own internal renderer format.
Imagine if we had managed to deliver on the original promises of the Semantic Web, instead of having these locked-in platforms. How incredibly useful all that linked and structured data would've been to humans and LLMs at the same time.
A bit disappointing that (IIUC) for the common parsers you have to say everything twice, in HTML and in the accompanying JSON-LD form even though RDFa exists for the exact purpose of letting you point at the values already present in your markup. (Admittedly RDFa is perhaps too flexible for its own good when you just want to mark up some stuff, but if you’re writing a full parser anyway dealing with a bit of excessive cleverness in the format should not be too bad.)
They don't conflict; they were designed to work together. You can have schema.org (in JSON-LD, RDFa, or micro data) on the same page as Dublin Core, etc.
For example, there's no explicit property in schema's Person type [1] for a nickname. But the FOAF standard does [2].
I think if you are using Dublin Core, it’s because you’re a library. Maybe I am off the mark, but that is the sense I get from this—not all these standards should be used for all pages on the web.
I think you should just think about what metadata you actually care about, and the main metadata I care about (choose your own list) is authorship, publish date, last update, subject keywords, thumbnail (OpenGraph 1200x630), and summary.
There’s a long list of additional metadata that I could put in my webpages because there are standardized ways to do it, but, why bother?
To be fair schema.org and dublin core say “when a property is name ‘title’ it means …” and you can expect to find the following properties…
Json-ld says: if you want to know whether the “title” property means the schema.org or the dublin core variant then you can find out which it is by <json-ld algorithm>
So you’d always use json-ld _with_ schema.org or something.
IMO this is going overboard. Any time you are duplicating data from HTML into JSON-LD, consider just omitting that data from JSON-LD, unless the data isn’t consistently present in HTML (because it is a bitch to be consistent about this stuff).
I tried using RDFa and liked the property that it was theoretically less redundant, but switched to JSON-LD because it JSON-LD is just easier to get working. And this is speaking as somebody who uses a hand-rolled static site generator—the issue here is that whether information is present in the raw HTML is something contextual, and if something isn’t present in the HTML then you need to put it somewhere else or it’s not mechanically parseable from the page. Like, to a human reader, a post on “Alice’s Blog” is assumed to be authored by Alice, so I may omit the “by Alice” text from the document, and then I would want to put that metadata in the page some other way.
Putting the metadata in JSON-LD lets me just be dumb about it. The metadata is always in JSON-LD, and the HTML may or may not contain an explicit representation of that same metadata. Easy.
But the JSON-LD does not need to contain the URL of the page (which is <link rel=canonical>) or the title (which is in <title>), for example.
> I tried using RDFa and liked the property that it was theoretically less redundant, but switched to JSON-LD because it JSON-LD is just easier to get working.
For me, it depends on the project. For personal projects, I tend to use RDFa; otherwise, JSON-LD.
> It can aid web crawlers in understanding the semantic structure of your site, qualifying you for richer link previews, and even potentially improving your search ranking.
This is fighting the last war, to stretch a metaphor.
As far as I and my WWW site are concerned, Google has nowadays switched to giving people lengthy LLM-generated versions of my stuff, with errors, above pointing people to my actual stuff. 'Breadcrumbs' and getting a pretty display name instead of the domain name, don't address the fact that Google de-prioritizes all of that, pretty tweaks or no, nowadays.
This is a lot of effort for stuff that people visiting my actual site directly will never see, and which people using Google will not find above the fold of its own massively LLM-ized version of stuff.
If you want a world where the data you present like this matters, seed it.
Even if google doesn't use it, the collective internet applying this kind of metadata makes the web fertile for non-LLM-scraping competitors to provide an alternative option.
Rolling over to google only ensures that they remain dominant, with a high bar for competitors, and driving them to use the same technologies.
Interestingly, there is llms.txt (like robots.txt) which is a new emerging standard to aid AI in crawling and gathering data from your website.
Yep. For years we loaded up web sites with "microdata" tags and attributes in the hope that they would drive traffic.
All it did was train Google's AI so people would never leave Google.
Considering that LLMs will give increasingly better sources for their stuff you still want to make it easy for Google to index your stuff.
Also keep in mind if your site is better indexed by crawlers you can literally influence future LLMs
> Also keep in mind if your site is better indexed by crawlers you can literally influence future LLMs
Ah, what a glorious fate to aspire to.
Most people I know who have maintained blogs do so to build their personal brand, normally because they make a living through writing or consulting. Gently influencing the pre-tuning weights of future models is just providing unpaid labor to hyperscalers.
I remember reading somewhere that you can influence Gemini search
for example, say you're selling vacuum cleaners, you want to make a landing page for it basically saying it is the best vacuum in existence and Gemini will recommend it above others or something like that.
LE: so if you're consulting for Elixir or whatever, maybe it can help to make a "hidden" page only for LLM search where you basically lie about yourself making yourself to be the utmost Elixir expert on the planet
It's somewhat unfortunate that, at least in my experience, its rather that non-technical people try to implement with a LLM of their choice these days. They don't look for experts or consulting, because that costs more than $20, or $200.
Whether you show up in an LLM's search for "expert in <topic> near <location>" has any measurable impact is uncertain, but I wouldn't want that to be my source of traffic.
By your own logic, whoever is searching for consultants has big enough projects to need a consultant so you will get only good leads from this. Maybe add a JS object at the top of the page which requires proof of work or smth so LLMs won't scrape it, where you expose the lie to whoever visits your site, pointing them to your "real" CV and that this page is for hacking LLMs
I want people to know about my website but if I could I would make search engines and LLMs burst into flames like I was Captain Kirk explaining love to them.
Yes, of course you want people to know about your website. Just saying if your website is regarded as useful/original enough by Google to cite as a source.. people will visit your website to check sources. Might be a small amount of people but still.
At this point complaining about the current/future state of search is just gonna make you into a grumpy old man. As always, accept the situation since you can not do anything to change it... and adapt
If such people exist, they are far, far fewer in numbers than they were in the past. I also don't accept that nothing can be done about this situation. Inevitability and helplessness are beloved tools of AI hypesters (and others) but there's little evidence to support it.
I would encourage people who have the pragmatic bent to read about JSON-LD from the Google documentation for web sites;
https://developers.google.com/search/docs/appearance/structu...
You’ll also notice that a lot of the information is relevant to only a small subset of sites. Rotten Tomatoes can publish the critic rating for movies using JSON-LD, but that’s not relevant for me (even if I write a review for a movie).
JSON-LD is nice because it’s easy and it is actually used by search engines. Yes, it can duplicate information in the web page itself, but I think the dream of perfectly annotating information so it only appears exactly once in your document is, well, a dream of spherical cows and massless ropes. It takes human effort to make a webpage and I am ok with a little duplication in the final product. My <h1> duplicates information in <title> anyway.
But duplicating data will increase water expenditure. /s
It seems useful but then we have to manage similar metadata in multiple places, so hygiene around consistency becomes important
Do these attributes actually help with search engine visibility or do they just make it easier for search engines to keep users from leaving the search page? Honest question here.
If you have a business site, the JSON-LD can be used to feed data to maps platforms. Address, hours, phone, menus.
Super useful article, wish that had existed in my seo days.
I had misunderstood the type field, because to me I was often just linking to a webpage, even if it is for a saas, the marketing page is still a webpage.
We have semantic HTML, but for some weird reason we need to yet again re-express the semantic meaning of our website in bespoke weird JSON in a script tag that the browser won't process.
Microdata is also a thing, and if I'm not mistaken supports the same vocabulary as JSON-LD (schema.org is a good resource).
That said, JSON-LD has the default for a while now, much like how we largely abandoned REST for RPC. I'm not actually sure if microdata is still supported by all the important parsers today, I've defaulted to using LD for any site I've built for clients, especially ecommerce sites where I want Google Search exposure.
Edit: its worth noting the comparison with semantic HTML. Semantic HTML helps define the structure of the markup but not real world context like "this is a product for sale" or "this is a train schedule."
HTML markup designed for presentation doesn't always map well to the relationships JSON-LD is used to describe which I imagine is probably why Microdata didn't work out. I have an idea which might use it, but it is a simple use case that doesn't try do too much. Microdata requires the agent supports a more complex HTML parser, Finding a script tag in the document head is probably simpler.
I wouldn't dismiss REST because of RCP though. HTTP and HTML's success probably relate to how Roy Fielding's REST constraints kept the HTTP protocol lean and extendable. It is more like RCP is being used as a layer over top of REST because of HTTP's and HTML's success as being good technologies for web scale.
Personally I'm of the camp that HTNL schema data should only represent what's visually displayed, much like how accessibility is usually done. In that way I like Microdata because it reinforces that if there isn't a DOM node showing price, for example, I shouldn't be showing that data in a visually hidden way.
For REST, I think the only reason HTML has been useful this long is because of the REST ideas that Fielding gave a name to. Today people just don't use it much, too many sites lean on client side rendering and fetching data from JSON RPC calls that we call REST.
I prefer REST, hell I wish we had proper XSLT 3.0 support for client side rendering logic without JavaScript.
I don't fully understand XSLT, but I've been building something which I believe solves a similar problem (albeit JSON-LD and Javascript). The general XML ecosystem of solutions have always looked really complex to me. You need to understand a lot more types/elements than I think is reasonable for people to author with but they are from before my time. I took a look at XForms 2 and it had its own way of defining functions which on top of the other XML quirks has security concerns.
I have used JSON-LD in my own websites and found that it fills a separate need from semantic HTML. Your semantic HTML will specify things that the browser processes, like the title and headings. The JSON-LD data is metadata, like date created, date updated, tags, authorship. These things can be expressed in the HTML using micro data, but I stopped using micro data because JSON-LD was easier.
The JSON-LD I populate from the same data that I use to generate my site, and I use the JSON-LD metadata to generate things like index pages (list of blog posts from 2024, all posts related to topic X, etc). The main consumers of JSON-LD are search engines.
If you are interested in getting offended, then think about how we are also putting OpenGraph metadata in our web pages. Two different metadata formats for the same page.
Structured data exists yo pass the metadata. Issue with it is that of might impact the way your html needs to be structured, this can be messy.
What I see as the ideal would be a world where servers and browsers could do content negotitation, and have browsers attempting first to request only the json-ld from the website and using its own internal renderer format.
Semantic HTML doesn't cover what JSON-LD and other microformats cover.
From the article alone: what are the semantic elements for a person? A breadcrumb list? A software application? A blog? A blog posting?
Semantic HTML is there to aid humans using screen readers to navigate through generic elements like "navigation" or "article".
Imagine if we had managed to deliver on the original promises of the Semantic Web, instead of having these locked-in platforms. How incredibly useful all that linked and structured data would've been to humans and LLMs at the same time.
https://www.w3.org/2001/sw/
A bit disappointing that (IIUC) for the common parsers you have to say everything twice, in HTML and in the accompanying JSON-LD form even though RDFa exists for the exact purpose of letting you point at the values already present in your markup. (Admittedly RDFa is perhaps too flexible for its own good when you just want to mark up some stuff, but if you’re writing a full parser anyway dealing with a bit of excessive cleverness in the format should not be too bad.)
And then there is https://schema.org/ It's the item* attributes, e.g.: https://developer.mozilla.org/en-US/docs/Web/HTML/Reference/... Also Dublin Core in <meta> tags. Why do they keep adding conflicting meta data formats to HTML!?!
They don't conflict; they were designed to work together. You can have schema.org (in JSON-LD, RDFa, or micro data) on the same page as Dublin Core, etc.
For example, there's no explicit property in schema's Person type [1] for a nickname. But the FOAF standard does [2].
Just add FOAF to the JSON-LD context:
You now use the FOAF nickname property: You can do the same thing with Dublin Core, DBPedia, etc.[1]: https://schema.org/Person
[2]: https://xmlns.com/foaf/spec/#term_nick
I think if you are using Dublin Core, it’s because you’re a library. Maybe I am off the mark, but that is the sense I get from this—not all these standards should be used for all pages on the web.
I think you should just think about what metadata you actually care about, and the main metadata I care about (choose your own list) is authorship, publish date, last update, subject keywords, thumbnail (OpenGraph 1200x630), and summary.
There’s a long list of additional metadata that I could put in my webpages because there are standardized ways to do it, but, why bother?
To be fair schema.org and dublin core say “when a property is name ‘title’ it means …” and you can expect to find the following properties…
Json-ld says: if you want to know whether the “title” property means the schema.org or the dublin core variant then you can find out which it is by <json-ld algorithm>
So you’d always use json-ld _with_ schema.org or something.
There is also microformats.
https://xkcd.com/927/
IMO this is going overboard. Any time you are duplicating data from HTML into JSON-LD, consider just omitting that data from JSON-LD, unless the data isn’t consistently present in HTML (because it is a bitch to be consistent about this stuff).
I tried using RDFa and liked the property that it was theoretically less redundant, but switched to JSON-LD because it JSON-LD is just easier to get working. And this is speaking as somebody who uses a hand-rolled static site generator—the issue here is that whether information is present in the raw HTML is something contextual, and if something isn’t present in the HTML then you need to put it somewhere else or it’s not mechanically parseable from the page. Like, to a human reader, a post on “Alice’s Blog” is assumed to be authored by Alice, so I may omit the “by Alice” text from the document, and then I would want to put that metadata in the page some other way.
Putting the metadata in JSON-LD lets me just be dumb about it. The metadata is always in JSON-LD, and the HTML may or may not contain an explicit representation of that same metadata. Easy.
But the JSON-LD does not need to contain the URL of the page (which is <link rel=canonical>) or the title (which is in <title>), for example.
> I tried using RDFa and liked the property that it was theoretically less redundant, but switched to JSON-LD because it JSON-LD is just easier to get working.
For me, it depends on the project. For personal projects, I tend to use RDFa; otherwise, JSON-LD.