Getting Cited by AI: What Actually Works — and Why llms.txt Doesn’t
Quick Answer: 97% of llms.txt files received zero requests in May 2026, according to an Ahrefs study of 137,000 domains. Schema markup fared no better — across 1,885 pages adding schema, AI citation changes were statistically indistinguishable from zero (Ahrefs, 2026). The tactics the industry is selling don’t match the data. What actually drives AI citations is unique, answer-first content that doesn’t exist elsewhere, backed by original numbers and delivered through clean, structured data paths.
In this article
- Does publishing an llms.txt file get you cited by AI?
- How many sites bothered to publish llms.txt?
- Do the AI bots that generate citations even fetch llms.txt?
- Does schema markup drive AI citations?
- Does Google use llms.txt for AI Mode?
- What actually drives AI citations then?
- Is there any correlation between llms.txt and citations at scale?
- What should a store do instead of chasing llms.txt?
Does publishing an llms.txt file get you cited by AI?
No — an Ahrefs study of 137,000 sites found 97% of llms.txt files received zero requests in May 2026 (Ahrefs, 2026). The file exists on paper as a proposed standard, but no major AI platform has committed to reading it, and the usage data confirms it’s being ignored at scale.
The premise behind llms.txt is reasonable on the surface: give AI crawlers a machine-readable summary of your site so they can index it efficiently. The problem is that the AI engines generating citations — ChatGPT, Perplexity, Google AI Mode — don’t use it. They pull from their existing web indexes, retrieval-augmented generation pipelines, and search APIs. A text file sitting in your root directory doesn’t enter that pipeline.
Here’s the thing: the gap between the advice and the data is enormous. Consultants and tool vendors are selling llms.txt implementation as a core AEO tactic. The largest study to date says it does nothing measurable. That’s not a minor discrepancy — it’s a complete disconnect between what’s being sold and what works.
97% of llms.txt files received zero requests in May 2026 — the file the industry is selling as essential is being universally ignored by the platforms that matter (Ahrefs, 2026).
How many sites bothered to publish llms.txt?
28% of the 137,000 domains published an llms.txt file despite no major AI platform committing to read it (Ahrefs, 2026). Adoption outpaced demand by a staggering margin — more than one in four sites invested effort in a file that almost nobody fetches.
That 28% figure tells you something about how AEO advice spreads. A proposed standard with no platform adoption got implemented on over a quarter of a large, representative sample of domains. The advice to “add llms.txt” propagated through blog posts, agency recommendations, and SEO tool checklists faster than anyone checked whether it produced results.
The adoption curve followed the usual pattern: early-mover blogs wrote about it, tools added automated generation features, agencies added it to their deliverables, and clients paid for implementation. At no point did the feedback loop include “did this actually get read by an AI engine?” The answer, for 97% of those files, was no.
Related: The Data Says AI Crawlers Ignore Your llms.txt
Do the AI bots that generate citations even fetch llms.txt?
Barely — AI retrieval bots like ChatGPT’s and Perplexity’s accounted for just 1.1% of the requests that reached llms.txt files (Ahrefs, 2026). The vast majority of the 3% that got any traffic at all came from SEO crawlers and monitoring tools, not from the engines that actually generate citations.
This is the number that should end the debate. Even among the tiny fraction of llms.txt files that received any request at all, the traffic wasn’t coming from citation engines. It was coming from bots that crawl everything — Googlebot variants, Ahrefs’ own crawler, and similar infrastructure. The bots that power the AI answers people actually see were almost entirely absent.
Translation: the 3% that got traffic weren’t getting cited. They were getting crawled by the same bots that crawl every file in your root directory. The llms.txt file wasn’t being treated as special input — it was being treated as just another page to index.
Does schema markup drive AI citations?
Not meaningfully — across 1,885 pages adding schema, Ahrefs found AI Mode and ChatGPT citation changes statistically indistinguishable from zero (Ahrefs, 2026). Schema helps search engines understand your page structure, but it does not appear to influence whether AI engines cite you.
This finding challenges the second pillar of mainstream AEO advice. If llms.txt is the “make yourself visible” lever and schema is the “make yourself understandable” lever, the data says neither lever is connected to the machine. AI citation engines are making decisions based on content quality, uniqueness, and relevance — not on whether your JSON-LD is correctly nested.
That doesn’t mean schema is worthless. It still serves its original purpose: helping Google’s traditional search engine render rich snippets, understand product data, and improve structured search results. The mistake is assuming that what works for traditional search automatically transfers to AI citation. It doesn’t. These are different systems making different decisions.
Related: Does schema markup improve AI search citations?
Across 1,885 pages adding schema markup, AI Mode and ChatGPT citation changes were statistically indistinguishable from zero (Ahrefs, 2026).
Does Google use llms.txt for AI Mode?
No — Google has stated it ignores llms.txt, and the data confirms 97% of such files never received a single request (Ahrefs, 2026). Google AI Mode pulls from its existing search index, not from a separate AI-specific file.
This matters because Google AI Mode is the highest-traffic AI citation surface for most sites. If your AEO strategy is built around llms.txt and Google explicitly doesn’t use it, you’ve optimised for a channel that doesn’t exist. The citations Google AI Mode serves come from the same indexed content that powers regular search results — filtered through a different presentation layer, but sourced from the same corpus.
The implication is clear: if you want Google AI Mode to cite you, optimise the same things you’d optimise for search visibility — unique information, clear answers, authoritative sourcing. The AI layer is a different interface to the same index, not a different index with different inputs.
What actually drives AI citations then?
Unique, useful information that doesn’t exist elsewhere on your site, delivered answer-first, is what gets cited — the format of the file doesn’t matter; the uniqueness of the information does (Search Engine Land, 2026). AI engines cite content that directly answers a question with a specific, verifiable claim.
The pattern across every study points to the same variables. Content gets cited when it contains a specific data point no one else has, when it answers the question in the first sentence rather than burying it in paragraph four, and when other sites reference and link to it as a source. These are content-quality signals, not technical-implementation signals.
For WooCommerce store owners, this means your competitive advantage isn’t a file or a markup format. It’s the data only you have: your own conversion rates, your own A/B test results, your own customer behaviour patterns. Original numbers from your own stack are inherently unique — no competitor can replicate them, and AI engines that value unique sourcing will favour them.
| Tactic | Industry Advice | Data Says | Source |
|---|---|---|---|
| llms.txt | “Essential for AI visibility” | 97% got zero requests | Ahrefs, 2026 |
| Schema markup | “Helps AI understand your content” | Citation change indistinguishable from zero | Ahrefs, 2026 |
| llms.txt in prediction models | “Correlates with citations” | Removing it improved model accuracy | Search Engine Land, 2026 |
| Unique answer-first content | “Table stakes” | Primary citation driver across all studies | Ahrefs / SEL, 2026 |
LIFETIME LOCK on Harvest Annual Plan
Push your WordPress event data server-side seamlessly into BigQuery so AI can analyse it.
Is there any correlation between llms.txt and citations at scale?
No — an analysis of 300,000 domains found that removing llms.txt from a citation-prediction model actually improved its accuracy (Search Engine Land, 2026). The file’s presence had zero or slightly negative predictive value for whether a site gets cited by AI.
This is the strongest piece of evidence against llms.txt as a citation lever. It’s not just that the file doesn’t help — it’s that including it in a model designed to predict citations makes the model worse. The signal is noise. Sites that publish llms.txt aren’t more likely to be cited; if anything, the correlation runs the other way, likely because the sites investing in llms.txt are the same ones following generic AEO playbooks rather than building genuinely unique content.
The question isn’t whether to publish llms.txt. The question is whether the time and attention spent on implementation tricks would be better spent creating content that actually earns citations. At 300,000 domains, the data is clear enough to answer that.
Related: What product data do AI search engines cite?
What should a store do instead of chasing llms.txt?
Publish genuinely original, structured answers with verifiable stats and build brand mentions across the web — these correlate with AI visibility far more than any file format (Ahrefs, 2026). Invest in clean first-party data infrastructure so your content carries numbers no one else has.
The actionable takeaway is a shift in focus from format to substance. Stop asking “have I implemented the right files?” and start asking “does my content contain information an AI engine can’t get anywhere else?” Original data is the moat. A WooCommerce store sitting on its own conversion data, customer behaviour patterns, and A/B test results has information no competitor can replicate — and that’s exactly what citation engines favour.
Tools like Transmute Engine build that foundation by routing clean, structured first-party data server-side — giving you accurate numbers to publish rather than relying on the 60–70% of conversion data that survives client-side tracking. You can’t publish original stats if your tracking stack loses a third of your data before you see it.
Let that sink in. The entire llms.txt and schema-for-AI conversation is about format. The data says the lever is content. The question isn’t X. The question is: do you have original data worth citing?
LIFETIME LOCK on Harvest Annual Plan
Push your WordPress event data server-side seamlessly into BigQuery so AI can analyse it.
Key Takeaways
- llms.txt is dead on arrival: 97% of files got zero requests; AI retrieval bots account for just 1.1% of the traffic that does arrive.
- Schema doesn’t move the needle: across 1,885 pages, adding schema produced citation changes indistinguishable from zero.
- Google ignores it explicitly: Google AI Mode doesn’t read llms.txt — it pulls from its existing search index.
- The lever is content, not format: unique, answer-first information with original data is what gets cited.
- Original data is the moat: numbers only you have can’t be replicated — invest in the data infrastructure that produces them.
No. An Ahrefs study of 137,000 sites found 97% of llms.txt files received zero requests in May 2026. The file exists on paper as a standard, but no major AI platform has committed to reading it.
28% of the 137,000 domains in the Ahrefs study published an llms.txt file despite no major AI platform committing to read it. Adoption outpaced demand by a wide margin.
Barely. AI retrieval bots like ChatGPT’s and Perplexity’s accounted for just 1.1% of the requests that reached llms.txt files. The vast majority of the 3% that got any traffic came from SEO crawlers, not citation engines.
Not meaningfully. Across 1,885 pages adding schema, Ahrefs found AI Mode and ChatGPT citation changes statistically indistinguishable from zero. Schema helps search engines understand page structure, but it does not appear to influence whether AI engines cite you.
No. Google has stated it ignores llms.txt, and the data confirms it. 97% of such files never received a single request. Google AI Mode pulls from its existing index, not from a separate AI-specific file.
Unique, useful information that doesn’t exist elsewhere, delivered answer-first with original data and clear sourcing. AI engines cite content that directly answers a question with a specific, verifiable claim. The format of the file is irrelevant.
No. An analysis of 300,000 domains found that removing llms.txt from a citation-prediction model actually improved its accuracy, meaning the file’s presence had zero or slightly negative predictive value.
Publish genuinely original, structured answers with verifiable stats. Build brand mentions across the web. Invest in clean first-party data infrastructure so your content carries real numbers no one else has, rather than chasing file-format tricks that the data says don’t work.