llms.text for affiliates: worth the effort or just a myth?

Everyone following SEO discussions in 2025 heard this pitch repeatedly – publish an llms.txt file at your site’s root, point it at your best content, and AI tools will start citing you instead of your competitors. It was free, it looked harmless, and it gave teams something concrete to do about a problem nobody fully understood yet. Nobody ever produced a case study showing that the file changed a single metric, which somehow didn't stop anyone from recommending it. 

Google put an end to the speculation in June 2026. The generative AI optimisation guide got a new "Mythbusting" section, and the wording left no room for interpretation: llms.txt files don't help or hurt your visibility in Search, because Google Search doesn't use them as a visibility or ranking signal. There was no mention of future support and no suggestion that this might change.

Around the same time, two studies came out asking a better question than the one Google answered. Google was never the only destination that mattered here. Affiliates want to know whether ChatGPT or Perplexity reads the file, since those tools are eating into comparison traffic. Both studies looked at exactly that, and the answers are worth a proper read.

What Google said, and why

In its June 2026 update, Google explained that llms.txt has nothing to feed into. AI Overviews and AI Mode run on the same index and the same ranking systems as regular Search. They retrieve from pages Google has already crawled, then write an answer from what comes back. There's no separate AI pipeline waiting for a text file to activate it. A crawlable page that Google understands is already in the running for an AI answer, and a Markdown summary at your root does nothing to change that.

There's a decent punchline attached to all this. In December 2025, an llms.txt turned up on Google's own developer documentation site, generated by an internal system that nobody had thought to switch off. It disappeared within a few hours, and the URL now returns a 404. Gary Illyes confirmed at a Search Central Live event that Google doesn't support the format and isn't planning to. John Mueller compared it to the keywords meta tag, making the point that a site describing itself is a claim the engine still has to verify against the page. So it reads the page.

The numbers behind "won't help or hurt"

By mid-2026, most people had accepted that Google ignores the file. The more useful question for affiliates is whether ChatGPT, Perplexity or any other answer engine uses it, since those increasingly handle comparison and "best of" queries in iGaming.

Ahrefs analysed roughly 137,000 domains, about 38,000 of which published an llms.txt file. During May 2026, 97% of those files received no requests. Among the requests that did arrive, most came from non-AI bots, including SEO crawlers and profiling tools. AI retrieval bots, including OAI-SearchBot, PerplexityBot and Claude’s search crawler, accounted for roughly one percent of all requests combined.

GEO strategist Eugen Ullrich published a sharper version of the same question in July 2026. Instead of measuring whether files get fetched generally, he took the source domains used by ChatGPT, Perplexity and other models for one finance client's prompt set over a month, tracked through Peec.ai, then checked which of those cited domains had an llms.txt. Of 2,041 domains, 25% did. Domains with a file were cited slightly less often, with a 0.26 average citation rate against 0.29 for domains without one, and were fetched less frequently per domain. The correlation between having the file and getting cited landed between -0.02 and -0.05, which is statistically indistinguishable from zero.

Neither study proves the file causes harm. What they show is narrower and more useful: no mechanism exists, official or observed, by which publishing an llms.txt increases how often a model finds or quotes your site.

Why most affiliates don’t qualify where the file genuinely helps

The studies tested llms.txt against AI search visibility, which was never what the format was for. Jeremy Howard designed it in September 2024 to solve a specific problem: helping an AI coding assistant like Cursor or Claude Code orient itself quickly in a set of API docs, instead of parsing a JavaScript-heavy site built for humans. Developers actually use it that way, and it works. Stripe, Cloudflare and Anthropic maintain lean llms.txt files for exactly this reason, because developers deliberately point their tools at them.

An affiliate review site publishing odds comparisons, bonus breakdowns and operator reviews has no equivalent audience. Nobody opens an AI coding assistant and asks it to integrate against a "Best Live Casino Bonuses" page. The one real exception in this vertical is odds-feed and affiliate-tracking API providers, where third parties genuinely build software against your endpoints. If that describes your business, a tight llms.txt pointing at your API docs is worth twenty minutes. If your business is content and comparison tables, it isn't.

The problem sitting underneath all of this

Otterly's AI citations report 2026 went through over a million tracked citations and found something far more actionable than any llms.txt study. About 73% of sites had technical barriers affecting AI crawler access. For those sites, content quality never came into play because the crawler never got past the door. Robots.txt rules written for training bots catch retrieval bots by mistake. CDN and WAF settings challenge user agents they don't recognise. Content that only appears after JavaScript runs shows up as a blank page.

iGaming has a version of this that most industries never deal with. Geo-blocking and age verification aren't marketing choices here; they're licence conditions. But an AI crawler hits your site from a data centre IP that doesn't match any licensed state, and a "confirm you're 21+ and in New Jersey" screen hides every page behind it from a bot that can't click through. Your reviews can be flawless, and the crawler still sees a compliance wall, exactly like a user on a VPN would.

Where to spend the time instead

Unless you run a public API, take llms.txt off the list and work through the things that have a mechanism behind them.

• Check robots.txt against named AI user agents. GPTBot, ClaudeBot, PerplexityBot, OAI-SearchBot and Google-Extended are separate decisions, and teams constantly collapse training-bot blocks and retrieval-bot blocks into one rule.
• Find out what your CDN does with AI crawlers by default. Several providers ship with AI blocking enabled, so robots.txt says yes while the edge says no and nobody notices for months.
• Test your age gate and geo-block the way a crawler experiences them. If the interstitial loads before your odds table, that table doesn't exist as far as retrieval goes.
• Confirm reviews and comparison tables appear in the raw HTML. Several crawlers don't run a full browser, so anything rendered client-side reaches them empty.

One more thing worth saying. If you're tempted to generate Markdown copies of every review and link them from an llms.txt just to cover yourself, skip it. You'd create another version of each review to maintain, with potential crawling and canonicalisation complications for no established AI citation benefit.

llms.txt isn't a myth so much as a good answer to a question affiliates weren't asking. The format works fine for the audience it was designed around, and most of us aren't in it. The hours you save by ignoring it are better spent finding out whether your own compliance setup is quietly keeping you out of the answers you're trying to appear in.