New Study Shows llms.txt Has Virtually Zero Impact on Answer Engine Optimization

The pitch usually goes something like this: unless your website carries a special file for AI crawlers, assistants like ChatGPT will never recommend your firm. Some companies have gone a step further and started selling a footer link that says "Hey AI, learn about us," which is supposed to steer those crawlers toward a page built just for them. Underneath all of it sits the same warning. Ignore this, and the AI hands your next client to a competitor.

I bring this up because Custom Legal Marketing got tired of hearing it and went looking for evidence. On November 9 their team ran 12,800 lawyer recommendation prompts through ChatGPT, Claude, Gemini, and Grok, and then they did the part nobody else bothers to do. They crawled every law firm website the platforms named and checked what was actually there. The report on llms.txt viability was released this week, and the topline number is a bad one for the pitch crowd. Of the firms AI recommended, 63.7% have no llms.txt file whatsoever.

How the study was put together

What I like about the approach is that it avoids the trap most AEO advice falls into, where somebody floats a theory and it gets repeated at conferences until everyone treats it as settled fact. CLM skipped the theory stage. Using their Sequoia platform, they asked the AI platforms to recommend lawyers across 30 practice areas in 100 U.S. cities, and the platforms came back with 55,512 website citations spread over 7,502 domains. Once the researchers cleared out the directories, bar associations, courts, dead URLs, parked domains, and anything else that wasn't a real working law firm site, they were left with 3,212 firm websites to examine. Each one got a visit and two simple questions. Is there an llms.txt file here, and is there a link anywhere on the page that's aimed at AI systems?

So rather than a survey where firms describe their own marketing, which is notoriously unreliable, this is an audit of the sites the AI platforms already trust enough to name in an answer.

Most of the recommended firms never bothered

Out of 3,212 recommended firm websites, 2,045 had nothing at the llms.txt address. That did not stop a single one of them from showing up in at least one AI answer, which is sort of the whole point of this exercise. Another 21 firms had an empty placeholder file with no heading and no links, and I would argue that's worse than having nothing, because it suggests someone started the job and gave up.

Even the 1,146 files that contained real content aren't quite what they seem. Half of them, 574 in total, carried the signature of an SEO plugin that generates the file automatically. Yoast SEO was responsible for 67.6% of the plugin-made files, followed by Rank Math at 17.2% and All in One SEO at 14.3%. What that tells me is that a large share of llms.txt "adoption" was never a deliberate decision. Somebody turned on a setting in a WordPress panel, and there's a decent chance they never thought about it again.

The adoption numbers also vary wildly depending on practice area. Injury firms, which spend more on marketing than anyone else in legal, had the file 48.2% of the time, while business and corporate firms sat at 17.9%. Remember that gap, because it becomes important when we get to the one finding that superficially favors the file.

The "Hey AI" link is rarer, and far more templated

The on-page version of this tactic, the little footer invitation telling AI to come learn about the firm, appeared on only 122 of the 3,212 sites, or 3.8%. The phrasing is strikingly consistent. Seventy-five of those 122 firms used some variation of "Hey AI, learn about us," and 58 of them used that exact wording. Forty of the dedicated AI pages live at the same URL path, /llm-info/, on completely unrelated firms. And where the footer displayed a website credit, at least 17 of those firms credited the same marketing agency.

When dozens of unrelated firms share identical wording at an identical address, you are looking at something a vendor rolled out, not a grassroots movement. Templates aren't a sin on their own. But they do force the question of whether the thing being templated actually works.

The part firms are paying for

This is where the study gets genuinely uncomfortable for anyone selling llms.txt as a service, so let me lay out the comparison between the 1,146 firms with the file and the 2,045 without it.

When it comes to position, nothing happened at all. The researchers found 2,388 answers that included firms from both groups, and in those head-to-head answers the firms with the file held the better average position 43.0% of the time and the worse one 44.8% of the time. If that sounds like randomness, that's because it is. The group I paid closest attention to was the 93 firms that all four platforms agreed on, since clearing four different AI systems is about the strongest endorsement in the dataset. Those consensus firms were the least likely of anyone to have the file. Three out of four had none.

In the interest of honesty, there is one gap worth discussing. Firms with the file averaged 5.04 recommendations compared to 4.19 for firms without it. Before anyone cites that number in a sales deck, though, consider that the median firm in both groups was recommended exactly twice, and that knowing whether a firm has the file explains less than half of one percent of the variation in recommendation counts. When the researchers compared firms only against others in their own practice group, the gap survived in just two of six groups, injury and criminal defense, and for business firms it actually ran backwards. That pattern is exactly what you would expect if the file were simply a marker of firms that spend heavily on marketing in general, rather than something the AI platforms respond to.

The footer link produced the same result at a smaller scale: 122 firms, nearly identical positions, and no detectable overall difference in how often they were recommended (p = 0.41).

None of this should come as a surprise

Back in April 2025, Google's John Mueller said that no AI service had claimed to use llms.txt, and that based on server logs, none of them even request it. He likened it to the keywords meta tag, and anyone who worked in SEO during the early 2000s understands how cutting that comparison is. The keywords tag died because it was a self-description that spammers stuffed until search engines stopped trusting it. A file where you get to explain, in your own words, why your site matters to AI has the same structural weakness from day one.

None of this means llms.txt is harmful. It is a text file, and a plugin will generate one for free. What deserves scrutiny is the invoice attached to it. If a vendor is charging your firm a recurring fee for the file, or presenting it as the centerpiece of an AI visibility strategy, this study hands you 2,045 reasons to push back.

What actually earns the recommendation

Nearly two-thirds of the recommended firms earned their spots with no file and no footer link, so clearly something else did the work for them. The likely explanation is the unglamorous one. What the firm's own pages say, whether independent third-party sources confirm it, and whether the firm's details stay consistent across the web are the things that seem to matter. That is the working core of law firm AEO, and none of it can be compressed into a text file sitting at the root of your domain.

AEO is still a young industry, and young industries tend to produce tactics that spread much faster than the evidence behind them. It is entirely possible the platforms formally adopt llms.txt at some point and it becomes standard practice. I wouldn't rule it out. But as of November 2026, the firms that AI recommends most often are overwhelmingly the ones that never set one up.

If you want the complete methodology, the statistical breakdowns, and the disclaimers, the full study from Custom Legal Marketing is worth reading before your next vendor renewal comes up.