NarsilCreative

Articles · SEO

Does llms.txt Work? 5 AI SEO Tactics Google Says to Skip

By Nate Morris · 5 min read

TL;DR

No: llms.txt does nothing for Google Search, and that's Google's own statement. Its July 2026 guidance names five 'AI SEO' tactics that don't work for Google: llms.txt and special AI markup, chunking content, rewriting pages 'for AI,' chasing inauthentic mentions, and over-focusing on structured data. Each has a fundamentals-shaped replacement, and the list doubles as an audit of your current agency's invoice.

No: llms.txt does nothing for Google Search, and that isn't our opinion, it's Google's. The Search Central guidance on generative AI features (updated July 2026) puts llms.txt first on a list of five "AI SEO" tactics it explicitly says don't work for Google. The same document says what does work, and none of it requires a new file, a new format, or a new line item on your invoice.

Here are the five, each with Google's position and the replacement that earns the effort instead. I'll say the against-interest part up front: this list shrinks what an AI SEO agency can legitimately sell you. We publish it anyway, because the acronym was never the product; the execution is.

1. llms.txt and special AI markup

The pitch: a file that tells AI systems what your site contains, like robots.txt for LLMs. The reality, per Google: Search doesn't use it, and it neither helps nor hurts. AI Overviews and AI Mode ground their answers in the same index and ranking systems as regular Search, so a side-channel file has nothing to plug into.

We don't maintain one on narsilcreative.com. If some non-Google system you care about reads the file, keeping one costs nothing; just make that a deliberate choice, not a deliverable you paid for as a Google play.

Do instead: make the pages themselves unambiguous. Clear titles, direct first-sentence answers, semantic HTML. The systems that matter read your actual site.

2. "Chunking" your content into fragments

The pitch: AI retrieves small passages, so break your pages into bite-sized blocks so the machine can lift them. Google's position: unnecessary; its systems understand multi-topic pages fine.

Do instead: structure for readers, which is what retrieval rewards anyway. Real headings that match real questions, an answer in the first sentence under each, one idea per section. That's not chunking; that's writing. (Google's query fan-out means a good multi-topic page can surface for sub-queries you never targeted; fragmenting it just spreads thin signals across thin pages.)

3. Rewriting your content "for AI"

The pitch: rephrase pages into machine-friendly wording, stuff synonyms, chase long-tail keyword variants so models recognize you. Google's position: its systems handle synonyms and meaning already; the paranoia is wasted.

One distinction worth keeping sharp, because the phrases sound alike: rewriting for AI (contorting prose so a model likes it) is the waste Google named. Editing AI-drafted content for humans (the de-slop pass, the em dash purge, the read-it-aloud test) is just editing, and it's more valuable than ever. One serves an imagined algorithm; the other serves the reader both audiences reward.

Do instead: spend the rewrite budget making content non-commodity. Google's own contrast is instructive: "7 Tips for First-Time Homebuyers" is a commodity anyone can generate; the story of why you waived the inspection and what it cost is content only you can write. That's what gets cited.

4. Chasing inauthentic mentions

The pitch: AI systems weigh brand mentions, so seed your name into forums, comment sections, and low-rent listicles at scale. Google's position: its spam policies apply to AI features too; manufactured mentions hit the same walls as manufactured links.

Do instead: earn the real thing. Genuine mentions (press, podcasts, reviews, community answers written by someone who means them) correlate with AI visibility precisely because they're hard to fake. This is the Temper phase of our method for a reason: verified authentic reviews, real press, real appearances. Slow is the feature.

5. Over-focusing on structured data

The pitch: more schema, more types, more markup, because AI needs structured data to understand you. Google's position: structured data isn't required for generative AI features. It still matters for rich results and product/commerce surfaces, so this one is a de-prioritization, not a ban.

Do instead: keep schema where it demonstrably serves a surface (products, business info, articles), and put the reclaimed hours into the content and site structure that feed AI answers. Keeping that core business data consistent everywhere machines read it is its own discipline, and we've covered it on its own.

Why these five keep selling anyway

It's worth naming the incentive, because the list will regrow under new names. All five tactics share two properties that make them irresistible line items: they're cheap to produce, and their effect is unfalsifiable. An llms.txt file takes minutes to generate. A "content chunking optimization" is a find-and-replace with a methodology name. Nobody can prove the mention-seeding didn't help, because rankings move for a hundred reasons and the invoice only claims one. Fundamentals have the opposite profile: expensive to execute, easy to audit. A rewritten money page either answers the query in the first sentence or it doesn't. Guess which category the acronym economy prefers to sell.

One more warning from the same Google document, pointed the other direction: its scaled-content abuse policy explicitly covers churning out pages to chase AI fan-out queries. The mirror image of the five fake tactics is one real one taken to abuse: mass-generating thin pages for every sub-query a model might spawn. The document that freed you from llms.txt also closed that door.

The forwardable version

The giveaway, formatted to paste into an email to whoever runs your SEO:

Per Google Search Central, "Optimizing your website for generative AI
features" (updated July 2026), Google Search does NOT use or need:
  1. llms.txt / special AI markup files
  2. Content "chunking" for AI retrieval
  3. Rewriting existing pages "for AI"
  4. Inauthentic brand mentions at scale
  5. Structured data beyond its normal rich-result uses

Which of our current deliverables map to items 1-5, and what would
you reallocate that effort to?

That last question is the audit. A good agency answers it happily, shows you the reallocation, and points at the same fundamentals Google does. A defensive answer full of proprietary-methodology fog tells you where the margin lives.

The bigger picture sits in the hub piece of this series: the very document that dismisses these hacks also tells you agents are coming to your website and to prepare your infrastructure. Both halves are true, and the full AI SEO checklist covers the floor that's left when the theater is stripped out. We run that floor, and the agent-readiness layer above it, as the Reforge Method; the five items above are the parts of the industry's pitch we'd rather retire than invoice.

Originally published August 17, 2026. Last updated August 19, 2026.

FAQ

Does llms.txt work for Google Search?
No. Google's Search Central guidance on generative AI features states that Google Search doesn't use llms.txt or special AI markup files; they neither help nor hurt. AI Overviews and AI Mode ground themselves in the same index and ranking systems as regular Search, so the file has nothing to plug into.
Should I delete my llms.txt file?
No need. It's harmless, it costs nothing to leave in place, and non-Google systems can choose to read it. The point isn't that the file is toxic; it's that paying for it as a Google visibility tactic buys nothing. Keep it or skip it deliberately, not because an invoice said it moved rankings.
So is GEO just a scam?
No, and that's not what Google said either. The same document that dismisses the hacks describes agentic experiences, UCP, and AI features grounded in Search, and says optimizing for them is still SEO. The real work (non-commodity content, clean structure, crawlability, semantic HTML) is exactly what earns AI citations. The scam version is the acronym sold with hacks attached; the legitimate version is fundamentals executed to a higher standard.
What makes Narsil Creative different?
We're an AI SEO agency that publishes its pricing ($1,500 and $4,000 per month), its method (the Reforge Method), and pieces like this one that shrink our own pitch: some of what the industry sells under AI SEO is, per Google, unnecessary. What's left is execution quality, and that's what we charge for.

See what AI says about your business

Ask ChatGPT to recommend a business like yours. If you're not in the answer, that's the problem we fix. Start with a free AI Visibility Audit. We'll show you where you appear, where you don't, and what to reforge first.

Related articles