Technical fixes mostly decide whether AI assistants can use your website as a source. They do much less to decide whether assistants recommend you. Across 916 websites audited with VisibAI, sites with structured data were cited as a source twice as often as sites without it (30% against 15%), while their average mention rate in unbranded answers barely moved (6.2% against 6.0%). An llms.txt file made no measurable difference to either.
TL;DR
- Sites with structured data were cited as a source in 30% of cases, against 15% without.
- Sites with sameAs links in their schema were cited in 34% of cases, against 23% without.
- Sites whose content depended on JavaScript to render were cited in only 6% of cases, against 28%.
- Sites with an llms.txt file were cited in 28% of cases, against 27% without: no real difference.
- Being recommended in answers was only weakly linked to any technical signal. That depends mostly on what other sources say about you.
Why separate "cited" from "recommended"?
Because they are different outcomes, and they respond to different work.
- Recommended (mentioned) means the assistant names your business when a buyer asks a question that does not mention you, such as "best accounting software for freelancers". This is what creates new customers.
- Cited means the assistant links to your website as one of its sources. This sends traffic and lends authority, and it can happen even when the answer is about your category in general.
Most technical advice treats these as one thing. The data shows they behave differently.
How did we analyse the 916 websites?
Each VisibAI audit fetches the audited website and records its technical signals: robots.txt rules for AI crawlers, structured data, FAQ content, an llms.txt file, JavaScript rendering problems and more. It then asks AI assistants a set of buyer-style questions and records, for each answer, whether the business was mentioned and whether its website was cited. Questions that name the business directly are excluded, because those always mention it.
We took the most recent completed audit for each distinct website, audited between February 22 and October 12, 2026, and kept the 916 sites whose homepage could be fetched. The sample is self-selected and leans towards B2B and software. These are associations, not proof of cause: a business with good schema often also invests in content and PR.
Which technical signals help a site get cited?
| Signal | Share of sites | Cited at least once (with) | Cited at least once (without) |
|---|---|---|---|
| Any structured data | 78.8% | 30% | 15% |
| sameAs links in schema | 40.8% | 34% | 23% |
| FAQ-style content | 47.1% | 31% | 24% |
| llms.txt file | 43.8% | 28% | 27% |
| AI crawler blocked in robots.txt | 6.9% | 22% | 27% |
| Content depends on JavaScript | 3.7% | 6% | 28% |
Structured data stands out. It tells an assistant in machine-readable form who you are, what you offer and where, and sites with it were twice as likely to appear as a cited source. sameAs links, which connect your site to your profiles on LinkedIn, review platforms and elsewhere, showed a similar effect.
The sharpest gap was JavaScript rendering. Sites whose main content only appeared after scripts ran were cited in 6% of cases, against 28% for the rest. The group is small (34 sites), but the direction is clear: if the text is not in the HTML, many AI crawlers never see it.
Does llms.txt make a difference?
Not in this data. Sites with an llms.txt file were cited in 28% of cases, sites without in 27%, and their average mention rates were similar (5.8% against 6.4%). We publish a free llms.txt generator and still think the file is cheap and harmless to add, but it is not a lever on its own. If you have limited time, structured data and server-rendered content come first.
Why didn't blocking AI crawlers hurt more?
Sites that blocked at least one AI crawler were cited slightly less often (22% against 27%), which is what you would expect. Their mention rate, though, was not lower. The likely reason is that many of these sites belong to larger, well-known brands. Their names appear across reviews, news and directories, so assistants recommend them from those sources even without reading the brand's own site.
That does not make blocking safe for a smaller business. If few other sources describe you, your own site is the main evidence an assistant has. Check what you are blocking, usually by accident through a security plugin or hosting default, with the free AI Crawler Checker.
So what actually gets a business recommended?
Coverage beyond your own site. Technical signals had only a small link to being mentioned at all: 40% of sites with structured data were mentioned in at least one unbranded answer, against 31% without, and the average mention rate was almost identical. The businesses that assistants recommend are the ones that review platforms, directories, communities and editorial sites describe clearly and consistently. Our citation data from 961 sites shows which of those sources ChatGPT and Perplexity favour.
Think of it as two layers. Technical work makes your site usable as evidence. Off-site presence gives the assistant a reason to name you in the first place. You need both, and most businesses that stall have done only the first.
What should you fix first?
- Make sure your content is in the HTML. If your site relies on client-side rendering, ask your developer about server-side rendering or prerendering.
- Add structured data with Organization or LocalBusiness schema and sameAs links to your real profiles. Our schema guide lists the types that matter.
- Check AI crawler access and remove accidental blocks.
- Write FAQ content that answers real buyer questions in the first sentence, then mark it up.
- Then shift effort off-site: reviews, directories, communities and editorial coverage in your category.
Frequently asked questions
Does schema markup improve AI visibility?
It mainly improves your chance of being cited as a source. In our data, sites with structured data were cited in 30% of cases against 15% without. Its link to being recommended was much weaker, so schema works best combined with third-party coverage.
Does llms.txt help with ChatGPT or Perplexity?
We found no measurable difference. Sites with an llms.txt file were cited as a source in 28% of cases, against 27% without, and had similar mention rates. It is quick to add, but it should not be your first priority.
Should I block AI crawlers?
For most small and mid-sized businesses, no. Sites blocking AI crawlers were cited less often as a source. Large brands can be recommended from other sources even when blocked, but a smaller business usually depends on its own site as the main evidence.
What is the difference between being cited and being mentioned by AI?
Being mentioned means the assistant names your business in its answer. Being cited means it links to your website as a source. A business can be mentioned without being cited, and a site can be cited in a general answer without the business being recommended.
What matters most for getting recommended by AI?
Clear, consistent coverage on the sources assistants trust in your category: review platforms, directories, community discussions and editorial articles. Technical fixes make your site readable, but recommendations come mostly from what others say about you.
See both layers for your own site
A VisibAI audit checks the technical signals on your site and measures, separately, how often AI assistants mention you and how often they cite you. Run one free and you will see which of the two layers is holding you back.