STANCICH.AI
Blog
SEO & AI Visibility5 min read

The AI SEO Placebo: What Agencies Sell That, by Google's and Ahrefs' Data, Does Not Work

Three tricks sold as the path to citations in ChatGPT and Google AI: llms.txt, schema 'for AI', and FAQ rich results. By Google's guidance and Ahrefs research, none of them does what it promises.

Ivan Stancich

Founder and engineer, STANCICH.AI

Key takeaways

  1. No major AI engine uses llms.txt as a signal for citations. Google confirmed this publicly, and OpenAI and Anthropic do not read the file to choose sources.
  2. Structured data (schema) is useful for classic Google rich results, not as an 'AI citation booster'. AI reads the same index as regular search.
  3. Google removed FAQ rich results for most pages in mid-2023. FAQ content still makes sense for question coverage, but not for stars and expanded answers in the results.
  4. The real levers are boring: clean indexing, a consistent Google Profile with reviews, presence in directories and registries, content with your own data.

Ever since AI search engines became a topic of conversation, a new offer has appeared: "optimization for AI", "we'll make ChatGPT cite you", "AI SEO package". Part of this offer is useful. Part of it is selling things that, by public guidance and research, do not do what they promise.

Below are the three most common ones, each with a dated reason why it does not work as a lever for AI visibility. We are not against these tools where they make sense, only against selling them as a shortcut to citations.

1. llms.txt: a file nobody reads to choose sources

What is being sold: "We add llms.txt (or ai.txt) and AI engines will read and cite your site." The file proposal is from 2024 and sounds logical, like robots.txt but for language models.

Why it does not work: no major AI engine uses this file as a signal for whom to cite. Google publicly confirmed it does not use llms.txt and that AI features in search run on the same index as regular search [1]. OpenAI and Anthropic do not read the file to choose sources in their answers. It is a proposal that has not been adopted by the very parties who would need to read it.

What this means for you: if you have llms.txt, it does no harm, it can stay. But if someone sells it to you as the reason "ChatGPT will now cite you", you are paying for a signal the recipient is not listening to.

2. Schema "for AI citations": right thing, wrong promise

What is being sold: "We add structured data (schema markup) and it will boost your citations in AI answers, by as much as 30 to 40 percent."

Why the number does not hold: structured data is genuinely useful, but for classic Google rich results (review stars, product data, breadcrumbs). It is not an "AI citation booster". Google clearly says you need no special schema to appear in AI features and that these features run on the same index and the same quality signals as regular search [1]. The specific "AI citation boost" percentages are usually something the seller cannot back up.

What this means for you: keep the schema, because it helps with classic results and must be valid. Just do not pay a premium for the promise that, because of it, "AI will cite you more often". That causal link is not proven.

3. FAQ rich results: the stars Google no longer shows

What is being sold: "We add FAQ schema and you'll get expanded answers in the results, with questions that attract more clicks."

Why it does not work: in August 2023 Google limited FAQ rich results to official government and health sites only, removing them for everyone else [3]. A normal business no longer gets these expanded FAQ blocks in the results, no matter how much FAQ schema it adds. The promise of "more space in the results via FAQs" simply has not held for most sites since 2023.

What this means for you: you can keep the FAQ content, because it covers real customer questions and reads well. Just do not expect stars and expanded answers in Google, because it no longer shows them for you.

So what actually works

The real levers for AI visibility are boring and mostly the same as for classic search:

Trick being soldStatusWhat works instead
llms.txt / ai.txtnot read for citationstechnically clean indexing, no 404s in the sitemap
schema "for AI citations, +30-40%"promise unprovenvalid schema for classic rich results, consistent Google Profile
FAQ rich resultsGoogle removed them in 2023content answering real questions, even without stars
paid placement in "AI lists"artificial, not a sourcereal presence in directories AI actually reads

In our study of 89 AI queries, businesses became visible through entirely different things: a well-kept Google Profile with enough reviews, consistent data in directories (Mojmojster, Primerjam), clean business registry records, and content with their own verifiable data. None of it is a shortcut. All of it works.

How to avoid the placebo

  1. For every "AI SEO" proposal, ask for the source: which engine exactly reads that signal, and when it confirmed it. If there is no answer, it is probably a placebo.
  2. Check the basics first: does AI mention you at all on "who do you recommend", is your Google Profile well-kept, are you in directories with real data.
  3. Put your money into things with a proven effect (reviews, consistent data, content with your own numbers), not into files and schemas that promise a shortcut.

If you want us to separate the wheat from the chaff for you

In the Visibility Teardown we measure where AI actually sees you and which levers truly move the needle in your niche, and we tell you which agency proposals are a placebo for you. It costs €290 and is fully credited if you continue working with us. No sales pressure: you can run most of the checks yourself.

Limits

Cards on the table: the claims above hold for the state of things in mid-2026 and rest on Google's public guidance and published research, not on a test of every individual tool. The technology changes; if one of these signals starts to work demonstrably, we will correct the article and date the change. Our own measurement (89 queries, 20 businesses, one day) shows patterns, not final laws.

Sources

  1. [1]
  2. [2]
  3. [3]
    Google removes FAQ and HowTo rich results

    Google Search Central · 2023-08-08

AI SEOllms.txtschemaFAQresearch
ShareLinkedInX

About the author

Ivan Stancich

Founder and engineer, STANCICH.AI

From the factory floor through Six Sigma and BI engineering to building AI systems. Builds tools and web systems for Slovenian small businesses, out of Koper.

Want to learn more?

Our team is ready to help with your digital project.