Does llms.txt matter? We tracked 10 sites to find out

The Brewing Controversy Over AI Indexing Standards

The advent of generative AI and large language models (LLMs) has fundamentally challenged traditional web optimization methodologies. As users increasingly turn to conversational interfaces like ChatGPT, Claude, Perplexity, and Gemini for answers, digital publishers are scrambling to ensure their content is discoverable and accurately utilized by these powerful AI agents. Central to this transition is the controversial file known as llms.txt.

The debate around llms.txt has quickly become one of the most polarized topics in web optimization. Proponents view it as foundational infrastructure—a necessary standard, akin to the venerable robots.txt or sitemap.xml—designed to guide AI crawlers toward the most valuable and extractable content. They argue that it is crucial for navigating the next generation of discovery.

Conversely, many seasoned SEO veterans dismiss the file as speculative infrastructure or “theater.” While numerous platform tools flag a missing llms.txt file as a critical site issue, anecdotal evidence and early server logs have suggested that mainstream AI crawlers rarely, if ever, request or parse them. To move past speculation and establish a data-driven conclusion, we conducted a focused tracking study across 10 diverse websites.

Google’s Ambiguous Relationship with `llms.txt`

The ambiguity surrounding the file intensified when Google, the creator of the sitemap standard and a leading force in AI, appeared to adopt it—and then quickly retreated. In December, the company added llms.txt files across several developer and documentation sites. For many digital publishers, the signal was clear: if the company guiding search standards was implementing it, then llms.txt must be an essential component of future AI strategy.

However, this perceived validation was short-lived. Google pulled the file from its primary Search developer documentation within 24 hours of its initial appearance. This swift reversal created significant confusion within the technical SEO community.

When questioned about the files that remained on other Google properties, John Mueller, a prominent figure in Google’s Search Relations team, offered critical clarification. Mueller explained that the initial change was the result of a sitewide Content Management System (CMS) update that many internal content teams were unaware of. Regarding the remaining files, he stated they were not “findable by default because they’re not at the top-level” and suggested that “it’s safe to assume they’re there for other purposes,” implicitly meaning they were not intended for standard external AI discovery or indexing.

Google’s mixed signals highlighted a crucial point: intent matters. If the file is not placed at the root level and is not actively supported by the largest LLM providers, its utility for external discovery is severely limited.

The Methodology: Tracking 10 Sites for Real Data

To move beyond the ongoing debates and anecdotal evidence, we initiated a controlled study designed to isolate the impact of llms.txt adoption on real-world performance metrics. Our goal was simple: to acquire data, not merely participate in the discussion.

We tracked the adoption and performance of llms.txt across 10 distinct websites representing diverse verticals:

Finance (Neobank)
B2B SaaS (Workflow Automation and HR Tech/Marketing Analytics)
E-commerce (Pet Supplies, Home Goods, Fashion)
Insurance
Pet Care

For each site, we analyzed performance over a 180-day window: 90 days before the file implementation and 90 days after. This pre-post analysis allowed us to establish a clear baseline and measure changes attributed to the file.

The key performance indicators (KPIs) we tracked included:

AI crawl frequency (via server logs, looking for known AI agent strings).
Direct referral traffic volume originating from major conversational AI platforms (ChatGPT, Claude, Perplexity, and Gemini).
Concurrent site changes (to identify confounding variables such as large content pushes, PR campaigns, or technical SEO fixes).

The Study Results: Little Correlation Found

The overall results demonstrated a stark reality: llms.txt, in isolation, had virtually no measurable impact on AI discovery or traffic for the vast majority of sites.

Two of the 10 sites saw measurable AI traffic increases of 12.5% and 25%, respectively. However, detailed analysis showed that llms.txt was not the causal driver of this growth.
Eight sites experienced no measurable change in AI traffic or crawl frequency.
One site declined by 19.7% during the tracking period.

The 2 ‘Success’ Stories Weren’t About the File

While two sites showed encouraging traffic spikes from LLM referrals in the post-implementation period, a deeper investigation revealed that the gains were driven by sophisticated content strategy and technical hygiene, not the documentation file itself.

The Neobank: 25% Growth Driven by Utility and Authority

This digital banking platform implemented llms.txt early in the third quarter of 2025. Ninety days later, AI traffic referrals had climbed by 25%—a phenomenal result on the surface. However, this growth occurred concurrently with a massive effort focused on content utility and external validation:

Major PR Campaign: The company executed a strategic PR campaign centered on its new banking license, resulting in high-authority coverage in major national publications, including Bloomberg. This external visibility significantly boosted the site’s authority and trustworthiness signals, which are key inputs for all LLMs.
Content Structure Overhaul: Product pages were comprehensively restructured to include readily extractable comparison tables detailing vital financial metrics such as interest rates, fees, and minimum account balances.
Targeted FAQ Expansion: The content team launched 12 new, highly specific FAQ pages, strategically optimized for rapid extraction by AI models looking for direct answers.
Resource Center Relaunch: A rebuilt resource center introduced new, authoritative content explaining complex banking concepts and financial information.
Technical Remediation: Critical technical SEO issues, particularly concerning header structures and crawl accessibility, were identified and fixed during this same window.

When a company generates high-profile press coverage, optimizes content for structured data extraction, and simultaneously fixes months-old technical barriers, it is impossible to attribute the resulting 25% growth solely, or even primarily, to the introduction of a new documentation file.

The B2B SaaS Platform: 12.5% Growth Powered by Functional Assets

The workflow automation company experienced a 12.5% jump in AI traffic just two weeks after implementing llms.txt. This timing seemed initially to present a compelling correlation.

However, the company’s internal content roadmap provided the real explanation. Three weeks prior to the file deployment, the platform published 27 highly functional, downloadable AI templates. These assets covered popular, high-value frameworks, including financial modeling tools, workflow planners, and project management blueprints.

Crucially, these were functional tools, not passive content marketing articles. The sheer utility of these assets drove significant engagement. During the same 90-day tracking period, Google organic traffic to the template section rose by 18% and continued to climb.

The increase in AI referral traffic mirrored the organic search success. Both search engines and conversational AI models surfaced these templates because they solved real, immediate user problems and represented an entirely new, high-value site section—not because their metadata was formally listed in an llms.txt file. The content’s function drove the discovery; the documentation was merely an afterthought.

The 8 Sites Where Nothing Happened After Uploading `llms.txt`

The vast majority of sites in the study—eight out of ten—saw zero measurable change in AI crawl frequency or referral traffic. This grouping included sites across high-traffic sectors like e-commerce (pet supplies, fashion, home goods), financial services, and B2B SaaS (HR tech, marketing analytics).

These seven sites all adopted the file and used it to document their best, most relevant existing content: API documentation, detailed product pages, buying guides, and case studies. Ninety days after implementation, traffic remained flat, and crawl frequency patterns were identical to the pre-implementation period. This strong finding suggests that if content is already indexed and discoverable through standard web pathways, llms.txt does not provide an additional boost to AI discovery.

The single declining site, an insurance provider that implemented the file in early September, saw a 19.7% traffic drop across all measured channels. This decline was likely related to market seasonality or specific competitive shifts, not the file itself. The key takeaway here is that llms.txt neither prevented the decline nor created any protective advantage against general market volatility.

The core insight from the majority group is this: Sites that launched new, functional, and highly optimized content saw gains, regardless of llms.txt adoption. Sites that simply documented their existing content using the file saw no gains whatsoever.

Analyzing the Disconnect: Where Are the LLM Standards?

The primary reason llms.txt failed to move the needle for eight out of ten sites is straightforward: there is currently no widespread, official commitment from the major LLM providers to parse or use the standard for content discovery.

As of this writing, neither OpenAI, Anthropic, Meta, nor Google has officially committed to treating llms.txt as an active indexing signal. If the primary consumers of the data (the AI models) are not looking for the documentation, the documentation holds no value for content discovery.

Google’s John Mueller reinforced this reality, stating plainly: “None of the AI services have said they’re using llms.txt, and you can tell when you look at your server logs that they don’t even check for it.”

The file may exist, and advocacy for its adoption may be strong, but the necessary platform implementation by the AI giants simply has not materialized yet. For digital publishers, investing resources into a standard that is not yet being honored is, at best, a low-priority exercise.

The Token Efficiency Argument and Its Limits

Despite the lack of indexing commitment, there is one niche argument where llms.txt offers potential value: token efficiency and clean data parsing.

When an AI agent needs to interact deeply with complex documentation—such as API instructions or user manuals—parsing raw HTML, which is often riddled with navigation elements, advertisements, JavaScript, and complex styling, is computationally expensive and “token inefficient.” Tokens are the currency of LLMs, and maximizing the amount of relevant information per token is crucial for cost and speed.

This is where the standardized format of llms.txt shines. By linking directly to clean, structured markdown documentation (often stored in a separate directory), publishers can provide AI agents with pure data, saving time and tokens when processing complex information.

The software deployment company Vercel, for example, has reported that 10% of their signups originate from ChatGPT referrals. Their llms.txt implementation includes contextual API descriptions that help AI agents efficiently fetch highly technical documentation.

This efficiency matters almost exclusively for developer tools, SaaS platforms with robust APIs, and products where the audience uses AI coding assistants like GitHub Copilot or Cursor to interact directly with the platform. In these cases, clean, direct data improves integration and reduces friction.

However, for general digital publishing—such as e-commerce selling pet supplies, insurance sites explaining policy coverage, or B2B SaaS targeting non-technical business buyers—token efficiency does not translate into a significant increase in discovery or referral traffic.

`llms.txt` is a Sitemap, Not a Strategy

The most useful analogy for understanding the true role of llms.txt is to compare it to a sitemap. Sitemaps are valuable infrastructure. They provide search engines with a clear, hierarchical map of all content, improving the efficiency of discovery and indexing.

Yet, no experienced SEO specialist credits massive traffic growth to the act of adding a sitemap. The sitemap documents what exists; the underlying quality and usefulness of the content drive actual discovery and search visibility.

llms.txt functions similarly. It may streamline the process for AI models to parse your site efficiently, assuming those models are actively looking for the file. But it holds no power to make existing content more useful, more authoritative, or more likely to be selected as the best answer to a user query.

Our analysis confirms that the sites that experienced real growth did so because they implemented core, foundational SEO and content strategies:

They created functional assets (like downloadable templates, structured data tables, and functional comparisons) that provided direct utility to the user.
They earned external visibility and trust through strategic press outreach and high-quality backlinks.
They removed technical barriers to crawling and indexing that had previously blocked content access.
They published content optimized for direct extraction, ensuring that FAQs and comparison points were clearly labeled and structured.

In the two success cases, llms.txt simply documented these successful efforts; it did not drive them.

What Actually Works for AI Discovery

The performance metrics of the successful neobank and SaaS platform offer a powerful roadmap for maximizing content utility in the age of AI. Digital publishers must prioritize the following fundamental strategies:

1. Create Functional, Extractable Assets

AI models prioritize content that directly solves a user problem. The B2B SaaS platform’s downloadable templates were assets of immediate, functional value. Focus resources on creating calculators, comparison charts, interactive tools, or guides that users can immediately deploy. These functional elements are often surfaced first by conversational interfaces because they represent definitive answers, not subjective commentary.

2. Structure Content for Immediate Extraction

The neobank’s shift to rebuilding product pages with clean comparison tables—detailing complex data like interest rates and fees—is a perfect example of extraction optimization. AI models excel at pulling definitive data points. Ensure key data, comparisons, prices, policy details, and steps are placed within clear headings, lists, or structured data (Schema) formats. If an AI model has to interpret a lengthy prose paragraph to find a number, it will often select content that presents that number directly.

3. Fix Technical Barriers First

Technical accessibility remains paramount. The neobank’s decision to fix header structure issues and crawl errors proved that if AI models cannot technically access or efficiently render your content, no amount of sophisticated documentation (like llms.txt) will help. Prioritize site health, rendering efficiency, and addressing indexing coverage issues before tackling experimental indexing standards.

4. Earn External Validation

Authority matters deeply to LLMs. The coverage gained by the neobank in Bloomberg and other major publications drove not only referral traffic but also branded searches. High-quality, trusted external validation signals to both traditional search algorithms and advanced LLMs that the content is authoritative, significantly increasing its likelihood of being selected and cited in generative answers.

5. Optimize for User Intent

Both successful sites targeted highly specific user queries: “best project management templates” or “how do [brand] interest rates compare?” Content that directly maps to defined user intent will always outperform content that is merely well-documented. AI models are engineered to satisfy the user request, making content utility the ultimate performance metric.

Should You Implement an `llms.txt` File?

For the majority of content-driven websites, particularly those in e-commerce, insurance, or general B2B marketing, the recommendation is clear: treat llms.txt like a sitemap—as useful, non-urgent infrastructure.

Implementing the file is good practice and involves minimal effort, so it won’t hurt your site. However, it should not be considered a growth lever or a priority tactic for traffic generation.

If your organization is a developer tool, an API provider, or heavily reliant on technical documentation where AI coding assistants are a primary distribution channel, then token efficiency becomes a relevant factor, and implementation is worthwhile. Your specific audience is already interacting with agents that benefit from clean data feeds.

For everyone else, the time and development resources spent implementing llms.txt are almost always better allocated to proven, high-ROI activities:

Restructuring product pages with extractable data points.
Publishing functional, downloadable assets.
Fixing outstanding technical SEO and indexing issues.
Developing highly structured FAQ and comparison content.
Executing strategies to earn high-quality press coverage and validation.

These proven tactics have already demonstrated real ROI in the landscape of AI discovery. llms.txt, based on the results of our 10-site study, has not—at least, not yet.

The lesson learned from tracking these 10 sites is that, in the rapidly changing digital ecosystem, it is easy to reach for familiar control mechanisms. llms.txt offers that comfort—it is concrete, actionable, and shaped like the established web standards we know. But looking like effective infrastructure is not the same as functioning as effective infrastructure.

Focus on the fundamentals that drive utility and accessibility, regardless of the constantly evolving platforms and formats:

Create truly useful content.
Structure it perfectly for machine extraction.
Ensure it is technically accessible to all crawlers.
Earn the external validation that builds authority.

Platforms and preferred documentation formats may shift, but the fundamentals of quality and utility remain the constant drivers of digital success.