Let’s say I ask “who created Linux?” Claude correctly tells me Linus Torvalds, and links to Wikipedia.
There are probably thousands, maybe hundreds of thousands of other sites that have that some piece of information. Are LLMs supposed to link to every site?
Most sites do not have unique information at all, and even sites that do rarely contain only unique information. Implying that most sites deserve links because they were crawled seems like a statistical fallacy. You could say the same thing about the percent of sites crawled by Google versus ever showing up on first page of results.
If the goal of the web was just to transmit objective facts like 'who created Linux', this wouldn't be a problem at all. Wikipedia handles that pretty dece.
The issue arises when we move away from objective trivia and into subjective, localized buying intent, which is where the web monetizes itself.
If I ask an LLM 'who created Linux?', there is one right answer. But if I ask an LLM 'who are the best commercial roofers in Houston?', there isn't one right answer. There are dozens of highly qualified local businesses that do possess unique value, unique pricing, and unique availability.
When an LLM answers that roofing question, it typically cites 3 to 5 businesses. The other 40 legitimate roofing companies in the area are left out. My study isn't arguing that every single one of those 40 companies deserves to be in the answer; it's pointing out that those 40 companies currently have no idea they are being left out.
In the Google era, if you weren't on page 1, you could look at Search Console, see your ranking, check your backlinks, and understand why. In the AI Search era, businesses are being scraped to build these answers, but they have zero telemetry on whether they are actually making the cut. This index is just an attempt to provide that missing telemetry.
The question is very important. What if you ask/tell an AI "I need a 2 bedroom vacation rental in Park City for a trip this fall". If you only get Airbnb results you are missing a lot of the true answer.
Are we saying that it's now a problem that we're not getting scraped?
This appears to be a new generation of "SEO", marking itself as a service for getting into AI results?
This is not the future I want to be a part of.
Perhaps it's inevitable that after a break from everything being driven by money that LLMs will now be ruined by people spending $X to get into AI to make back $X+1, leading to an arms race of ever increasing X, to the detriment of users.
I get that pushback. My goal with this index/report wasn't to write a new 'AI SEO' playbook or encourage people to start gaming the system. The goal was simply to shine a bit of light on what's going on. Right now, there is a massive information asymmetry: AI companies are turning their assistants into primary search engines, but webmasters have zero visibility into whether their sites are actually being cited in those live answers.
You're absolutely right to be concerned about an arms race. If the data showed that doing X, Y, and Z guaranteed a citation, we'd be right back to the worst days of keyword stuffing.
But what this data actually shows is the opposite: even if you do everything 'right' (allow the retrieval bots, provide perfect machine-readable schema, don't block anything), you still have a 94.8% chance of never being cited. How LLMs cite is entirely opaque.
I built this to give site owners a baseline measurement of what is actually happening to their content today (as a free, anonymous view), so they can make an informed decision on whether keeping their doors open to these crawlers is actually worth it.
Odd results for me. Last month I tested five AI chat sites, with web search turned off, and four of them had a shadow of information about me as a person (what kind of books I write, what tech I use, and a random bit of other information). The linked site gave me a zero score because it was testing if the AI models recommended my site for business or sales queries.
Blocking known bot identifiers via robots.txt does nothing by the way. Too many labs are running sneaky crawlers that do not respect robots.txt. You will need to take extreme measures: blocking basically all datacenter IP ranges, VPN IPs, aggressive rate limiting, etc.
Blocking LLM crawlers has become the number one use case for our IP database customers at https://focsec.com/
It is architecturally impossible for an LLM to associate a link or citation that it crawled with a response that comes out the other end. Every link they're giving you to support their statements is tacked on because it may vaguely match the tokens it just generated. It is perfectly common to find that a "source" does not contain the statements. You cannot expect an LLM to write you a Wikipedia article, much less a legal or medical opinion supported by research.
You can do it the same way human does: recall a fact from memory, and then search to identify a citation. The citation isn't "here's why I think this" but instead "here's where you can verify this".
> It is perfectly common to find that a "source" does not contain the statements.
It is commonly this way right now, but it doesn't have to stay this way forever.
It is also common, in my usage at least, to iteratively brow-beat the bot into paring its statements down to those that which are supportable by its sources. Doing so just takes repetition, and that repetition takes time and burns more tokens.
With the present state of things, the prompts to get moving on this and to guide the ultimate response into something that is verifiably supportable by outside sources can usually be simple and largely generic.
They're easy enough prompts that a subagent can produce them.
(I've done it myself with Codex subagents and it worked very well, aside from the unsustainable burn rate that did not fit my budget.)
For standard responses you're not wrong, but increasingly there are a lot of LLMS that are essentially doing RAG against search results. For example this is I believe how Kagi works, and Google AI overviews. I have increasingly seen Claude and chat GPT also doing the same thing where instead of answering a question from the knowledge Bank they will do a web search and cite the responses that were used.
This is why it's dumb when inferencing with a model to ask "...and cite all sources". Total waste of time and likely to make the response worse.
BUT
On models with web search, they can scan the sources in context, and those are pretty good at correct citations. However it's still a "trust, but verify" situation.
This is true, but good LLMs (even local, consumer-sized ones these days) can do this much better than … whatever it is that Google’s AI overviews use.
Think about your coding harness: the model reads a bunch of files into the context, and it generally doesn’t forget/hallucinate which lines came from which file.
I'd be surprised if it were anything other than just an LLM only equivalent of Gemma some 4B param model, they can't be spending much money on it because each query essentially needs to be cheaper than the ad revenue per search.
99% of AI users these days aren't using a 'raw' LLM. They are interacting with a harness that includes tools to do search of the live (or recently crawled) web. And so the output users actually see could absolutely include correct citation of sources. So the 'architecturally impossible' bit may be technically correct but it is not practically relevant. Now whether those harnesses do a good job of orchestrating LLM output to get accurate citations (and whether they are transparent about their process) is another thing entirely. But if you're contemplating 'what LLM's can do' and aren't taking into account the harness and tooling ecosystem they are embedded in then you're missing the point.
> It is architecturally impossible for an LLM to associate a link or citation that it crawled with a response that comes out the other end.
I mean, its not. Lots of LLM's focused on finance do this already.
Bloomberg's own ASKB produces results and provides links back to the source documents or urls that it references so people who care about correctness can verify the results.
Lol, this same behavior comes up more often than you'd expect in academia too. Many many citations that nobody cross checks that also don't support the originally cited piece of information.
It's hilariously common online as well. Bullshitter makes a claim, someone calls them on the BS, they put up a link, and (lo and behold) the link when examined does nothing to defend the claim. Often the link directly contradicts the claim.
Yep..the online link posters hit close to home. Big reason I hardly even engage in discourse anymore. Got tired of verifying links that don't have the claimed content in them.
This is a weird complaint.
Let’s say I ask “who created Linux?” Claude correctly tells me Linus Torvalds, and links to Wikipedia.
There are probably thousands, maybe hundreds of thousands of other sites that have that some piece of information. Are LLMs supposed to link to every site?
Most sites do not have unique information at all, and even sites that do rarely contain only unique information. Implying that most sites deserve links because they were crawled seems like a statistical fallacy. You could say the same thing about the percent of sites crawled by Google versus ever showing up on first page of results.
If the goal of the web was just to transmit objective facts like 'who created Linux', this wouldn't be a problem at all. Wikipedia handles that pretty dece.
The issue arises when we move away from objective trivia and into subjective, localized buying intent, which is where the web monetizes itself.
If I ask an LLM 'who created Linux?', there is one right answer. But if I ask an LLM 'who are the best commercial roofers in Houston?', there isn't one right answer. There are dozens of highly qualified local businesses that do possess unique value, unique pricing, and unique availability.
When an LLM answers that roofing question, it typically cites 3 to 5 businesses. The other 40 legitimate roofing companies in the area are left out. My study isn't arguing that every single one of those 40 companies deserves to be in the answer; it's pointing out that those 40 companies currently have no idea they are being left out.
In the Google era, if you weren't on page 1, you could look at Search Console, see your ranking, check your backlinks, and understand why. In the AI Search era, businesses are being scraped to build these answers, but they have zero telemetry on whether they are actually making the cut. This index is just an attempt to provide that missing telemetry.
Makes sense! Good point.
You're effectively reminding us that brands don't get understand well how to influence ai seo
That’s why I, from now on, only allow citable information to be scraped.
The question is very important. What if you ask/tell an AI "I need a 2 bedroom vacation rental in Park City for a trip this fall". If you only get Airbnb results you are missing a lot of the true answer.
I just asked ChatGPT-5.6 that question, no source was given.
Not even Wikipedia.
I wouldn't even consider it a complaint. For a builder, I would consider it an opportunity...
Are we saying that it's now a problem that we're not getting scraped?
This appears to be a new generation of "SEO", marking itself as a service for getting into AI results?
This is not the future I want to be a part of.
Perhaps it's inevitable that after a break from everything being driven by money that LLMs will now be ruined by people spending $X to get into AI to make back $X+1, leading to an arms race of ever increasing X, to the detriment of users.
LLM's were obviously going be ruined by two factors:
1- LLM Crawler Optimization: the new SEO
2- Weightings-For-Pay: for a fee have your product or service come up more frequently in associated answers
I get that pushback. My goal with this index/report wasn't to write a new 'AI SEO' playbook or encourage people to start gaming the system. The goal was simply to shine a bit of light on what's going on. Right now, there is a massive information asymmetry: AI companies are turning their assistants into primary search engines, but webmasters have zero visibility into whether their sites are actually being cited in those live answers.
You're absolutely right to be concerned about an arms race. If the data showed that doing X, Y, and Z guaranteed a citation, we'd be right back to the worst days of keyword stuffing.
But what this data actually shows is the opposite: even if you do everything 'right' (allow the retrieval bots, provide perfect machine-readable schema, don't block anything), you still have a 94.8% chance of never being cited. How LLMs cite is entirely opaque.
I built this to give site owners a baseline measurement of what is actually happening to their content today (as a free, anonymous view), so they can make an informed decision on whether keeping their doors open to these crawlers is actually worth it.
Odd results for me. Last month I tested five AI chat sites, with web search turned off, and four of them had a shadow of information about me as a person (what kind of books I write, what tech I use, and a random bit of other information). The linked site gave me a zero score because it was testing if the AI models recommended my site for business or sales queries.
This makes sense. Why would everyone be coted for the same answer
Makes me wonder if paid Medium and paid news is input to AI training. Surely, they are.
> 94.8%
This seems consistent with Sturgeon's Law.
Blocking known bot identifiers via robots.txt does nothing by the way. Too many labs are running sneaky crawlers that do not respect robots.txt. You will need to take extreme measures: blocking basically all datacenter IP ranges, VPN IPs, aggressive rate limiting, etc.
Blocking LLM crawlers has become the number one use case for our IP database customers at https://focsec.com/
Call it web 4.0
We don't have a page rank mechanism. Most likely people are paying or threatening AI companies to boost certain sources as authoritative
It is architecturally impossible for an LLM to associate a link or citation that it crawled with a response that comes out the other end. Every link they're giving you to support their statements is tacked on because it may vaguely match the tokens it just generated. It is perfectly common to find that a "source" does not contain the statements. You cannot expect an LLM to write you a Wikipedia article, much less a legal or medical opinion supported by research.
You can do it the same way human does: recall a fact from memory, and then search to identify a citation. The citation isn't "here's why I think this" but instead "here's where you can verify this".
> It is perfectly common to find that a "source" does not contain the statements.
It is commonly this way right now, but it doesn't have to stay this way forever.
It is also common, in my usage at least, to iteratively brow-beat the bot into paring its statements down to those that which are supportable by its sources. Doing so just takes repetition, and that repetition takes time and burns more tokens.
With the present state of things, the prompts to get moving on this and to guide the ultimate response into something that is verifiably supportable by outside sources can usually be simple and largely generic.
They're easy enough prompts that a subagent can produce them.
(I've done it myself with Codex subagents and it worked very well, aside from the unsustainable burn rate that did not fit my budget.)
For standard responses you're not wrong, but increasingly there are a lot of LLMS that are essentially doing RAG against search results. For example this is I believe how Kagi works, and Google AI overviews. I have increasingly seen Claude and chat GPT also doing the same thing where instead of answering a question from the knowledge Bank they will do a web search and cite the responses that were used.
This is why it's dumb when inferencing with a model to ask "...and cite all sources". Total waste of time and likely to make the response worse.
BUT
On models with web search, they can scan the sources in context, and those are pretty good at correct citations. However it's still a "trust, but verify" situation.
This is true, but good LLMs (even local, consumer-sized ones these days) can do this much better than … whatever it is that Google’s AI overviews use.
Think about your coding harness: the model reads a bunch of files into the context, and it generally doesn’t forget/hallucinate which lines came from which file.
I'd be surprised if it were anything other than just an LLM only equivalent of Gemma some 4B param model, they can't be spending much money on it because each query essentially needs to be cheaper than the ad revenue per search.
99% of AI users these days aren't using a 'raw' LLM. They are interacting with a harness that includes tools to do search of the live (or recently crawled) web. And so the output users actually see could absolutely include correct citation of sources. So the 'architecturally impossible' bit may be technically correct but it is not practically relevant. Now whether those harnesses do a good job of orchestrating LLM output to get accurate citations (and whether they are transparent about their process) is another thing entirely. But if you're contemplating 'what LLM's can do' and aren't taking into account the harness and tooling ecosystem they are embedded in then you're missing the point.
Then the least they can do is cite ALL their sources. At least somewhere on their website.
> It is architecturally impossible for an LLM to associate a link or citation that it crawled with a response that comes out the other end.
I mean, its not. Lots of LLM's focused on finance do this already.
Bloomberg's own ASKB produces results and provides links back to the source documents or urls that it references so people who care about correctness can verify the results.
Lol, this same behavior comes up more often than you'd expect in academia too. Many many citations that nobody cross checks that also don't support the originally cited piece of information.
It's hilariously common online as well. Bullshitter makes a claim, someone calls them on the BS, they put up a link, and (lo and behold) the link when examined does nothing to defend the claim. Often the link directly contradicts the claim.
Yep..the online link posters hit close to home. Big reason I hardly even engage in discourse anymore. Got tired of verifying links that don't have the claimed content in them.