# LLM SEO: what it is and why traditional SEO isn't enough

**Author:** Piotr Litwa - GTM & Analytics Specialist
**Published:** 2026-07-10
**URL:** https://piotrlitwa.com/articles/en/llm-seo.html
**Language:** en
**Keywords:** ["generative engine optimization", "geo seo", "ai search optimization", "get cited by chatgpt"]

---

LLM SEO is the practice of getting your content quoted inside AI answers rather than merely ranked on a results page. It goes by other names, generative engine optimization and AI search optimization among them, and they all describe the same shift.

The change is structural. Google shows ten links and lets the user choose. ChatGPT, Perplexity, and AI Overviews read the sources, synthesise an answer, and cite two or three. **Position one becomes worthless if the model doesn't pick you as one of the two.**

Here is what almost nobody says out loud: **most of what works for LLM SEO is what already worked for good SEO.** Clear structure, direct answers, real expertise, factual precision. The genuinely new parts are narrower than the industry pretends, and the tactics being sold as revolutionary are mostly repackaged fundamentals with a higher price tag.

This article covers what actually changed, what didn't, and how to measure a channel that deliberately sends you no referrer.

> **Key Takeaways**
> - LLM SEO optimises for being *quoted*, not ranked. Being the top link matters less than being the paragraph the model lifts.
> - The single biggest lever is answering the question directly in the first two sentences. Models extract early and rarely read to the end.
> - Models quote facts, numbers, and specifics. They almost never quote adjectives, so "industry-leading solution" is invisible to them.
> - Most AI traffic sends no referrer and lands in Direct. You will systematically undercount it unless you look for it deliberately.
> - Traditional SEO still does most of the work: the models are largely reading pages that already rank.

## What actually changed

Three things, and they matter in this order.

**The unit of value moved from the page to the paragraph.** Google ranks documents. A language model assembles an answer from fragments. Your beautifully structured 3,000-word guide might contribute one sentence, or nothing, depending on whether any single passage cleanly answers the question that was asked.

This is why the burying of answers is now actively expensive. A 400-word introduction that "sets the scene" before getting to the point used to cost you a few bounces. Now it costs you the citation, because the model found a cleaner answer in the first paragraph of a worse article.

**Citation replaced clicks.** When Perplexity answers with your content and links you as source 3, the user may never visit. You get attribution and brand exposure, and you may get no traffic at all. That's a genuinely different value exchange, and if your only success metric is sessions, you will conclude this channel doesn't work while your brand is being recommended daily.

**The competitive set changed.** You are no longer competing for a position against nine other links. You are competing to be one of maybe three sources a model considers worth quoting. There's no page two to hide on.

**Want to see what AI models can actually read on your site?** [AI Visibility](https://aiv.piotrlitwa.com/?lang=en) crawls your pages and shows what an LLM sees: missing markdown, missing structured data, unreadable pages. Free plan, one site, no credit card.

## What didn't change (and this is most of it)

The industry has an incentive to tell you everything is different. It isn't.

**Crawlability still decides everything.** If `GPTBot` and `ClaudeBot` can't fetch your page, nothing else on this list matters. Check your `robots.txt` before you buy any tool. I have seen companies pay for AI visibility consulting while actively blocking the crawlers in question.

**Authority still matters, and it still comes from the same places.** Models were trained on, and retrieve from, pages that rank. Backlinks, topical depth, and being cited by others are still the substrate. There's no shortcut where a thin site with no authority gets quoted because it wrote a good `llms.txt`.

**Content quality still wins.** Genuine expertise, specific claims, and being right are what makes a passage quotable. This isn't new advice. It's the advice that has always been correct and was always the hardest to follow.

The honest summary: **LLM SEO is 70% good SEO, 20% structural discipline, and 10% genuinely new.** Anyone selling you a wholly new discipline is selling the 10% at the price of the whole.

## LLM SEO tactics that actually move the needle

In order of return on effort.

### 1. Answer in the first two sentences

This is the highest-return change on the list and it costs nothing.

Whatever question the page addresses, answer it immediately, in plain sentences, before any narrative. Models extract from the top. A page that opens with "In today's rapidly evolving digital landscape..." has told the model nothing, and the model will find its answer elsewhere.

Every article on this site starts with the answer. That isn't a style preference. It's the mechanic.

### 2. Write facts, not adjectives

Models quote things that can be verified: numbers, prices, versions, dates, named products, specific mechanisms.

- "Our industry-leading monitoring solution" is unquotable. It says nothing checkable.
- "€150/month, weekly automated GTM health checks, monthly written report, cancel anytime" is quotable, because every clause is a fact.

Go through your key pages and count the checkable claims. If a page has none, no model will ever cite it, no matter how well it ranks.

### 3. One idea per section

Give each H2 a single clear job. A model retrieving a section on cross-domain tracking wants a section that is about cross-domain tracking, not a section about tracking generally that mentions it in passing.

This also makes the page better for humans, which is a recurring theme and not a coincidence.

### 4. Structured data and markdown twins

Schema markup (Article, FAQPage, HowTo) gives models an unambiguous parse of what your page contains. Markdown versions of your pages go further and remove the parsing problem entirely.

This is the mechanical half of the work, and I covered it in detail in the [llms.txt guide](https://piotrlitwa.com/articles/en/llms-txt-generator.html). It's genuinely useful and it isn't where most of the value is. Do not let a vendor convince you that a file fixes weak content.

### 5. FAQ sections written as real questions

Write FAQ questions the way people actually type them into ChatGPT, not the way SEO tools suggest. "Is GA4 legal in Europe?" is how a person asks. "GA4 Europe Compliance" is how a keyword tool talks, and nobody has ever typed it into a chat window.

## Measuring LLM SEO when the channel hides from you

This is where most LLM SEO programmes quietly fail, and it's worth being precise.

**AI traffic mostly arrives with no referrer.** A user reads an answer in ChatGPT, clicks your citation, and lands on your site. Depending on the client, you may get a referrer, or you may get nothing at all, in which case the session lands in [Direct](https://piotrlitwa.com/articles/en/direct-none-traffic-ga4.html) and is invisible as an AI referral.

Three things you can actually do:

**Check your server logs for AI crawlers.** `GPTBot`, `OAI-SearchBot`, `ChatGPT-User`, `ClaudeBot`, `Claude-User`, `PerplexityBot`. Live-fetch agents like `ChatGPT-User` and `Claude-User` are the interesting ones: they mean a model went to your page *because a user asked something*. That's demand, and it's measurable today.

**Filter for known referrers where they exist.** `chat.openai.com`, `perplexity.ai`, `claude.ai`, `gemini.google.com`. This undercounts, badly, but the trend line is still informative.

**Just ask the models.** Take the twenty questions your customers actually ask, put them into ChatGPT, Claude, Perplexity, and Gemini, and record who gets cited. It's manual, it's unglamorous, and it's the only method that directly measures the thing you care about. Repeat monthly, watch the trend.

That last one is what I do. It's a spreadsheet, not a platform, and it beats any dashboard because it measures the outcome instead of a proxy.

## What I would actually do first

If you have limited time, in this order:

1. **Check `robots.txt`** and confirm you aren't blocking AI crawlers you want to be read by.
2. **Rewrite the first two sentences** of your ten most important pages so they answer the question immediately.
3. **Audit for facts.** If a key page contains no checkable claims, it can't be quoted. Add numbers.
4. **Run the twenty-question test** and find out where you actually stand before optimising blindly.
5. **Then, and only then**, do the structural work: schema, markdown, `llms.txt`.

Notice the file everyone talks about is last. That ordering is deliberate, and it's the opposite of how this is usually sold.

## Frequently asked questions

**What is LLM SEO?**
Optimising your content to be quoted inside AI-generated answers rather than only ranked in a list of links. It's also called generative engine optimization or AI search optimization. The goal is being the passage a model lifts, not the top result.

**Is LLM SEO different from traditional SEO?**
Mostly no. Crawlability, authority, and content quality still do most of the work. What genuinely changes is that the unit of value moves from the page to the paragraph, and that citation can replace the click entirely.

**How do I get cited by ChatGPT or Perplexity?**
Answer the question directly in the opening sentences, write checkable facts rather than adjectives, keep one idea per section, and make sure the AI crawlers can actually reach the page. Authority still matters: models mostly retrieve from pages that already rank.

**Does llms.txt help with LLM SEO?**
Possibly, and nobody can prove it yet. No model provider has publicly committed to reading it. Markdown versions of your pages are the part of that work with reliable value, because they remove the parsing problem regardless of any standard.

**How do I measure AI traffic in GA4?**
Poorly, if you rely on referrers alone, because much of it arrives with none and lands in Direct. Check server logs for AI crawler user agents, filter for the referrers that do exist, and run manual prompt tests against the questions your customers ask.

**Is LLM SEO worth investing in yet?**
The fundamentals are, because they are the same fundamentals that improve normal SEO. The speculative parts, including `llms.txt` and AI-specific tooling, are cheap enough to be worth doing and too unproven to build a strategy around.

## Next steps

LLM SEO is real, and most of the advice being sold about it isn't new. The models are already reading pages that rank, extracting passages that answer questions directly, and quoting claims that can be checked.

So start with the boring, high-return work. Make sure the crawlers can reach you. Put the answer in the first two sentences. Replace adjectives with numbers. Those three things will do more for your AI visibility than any file you can generate.

Then measure it honestly, which means asking the models directly rather than waiting for a dashboard to tell you something it structurally can't see.

If you want to know what AI models can actually read on your site right now, [AI Visibility](https://aiv.piotrlitwa.com/?lang=en) crawls your pages and shows you the gaps: missing markdown, missing structured data, pages a model can't parse. Free plan, one site, 200 URLs, no credit card. And if you want a person to look at it and tell you what to fix first, that is a [scoping call](https://piotrlitwa.com/services.html#custom).

---

**Sources and further reading**
- [OpenAI crawler documentation](https://platform.openai.com/docs/bots) (OpenAI)
- [Anthropic crawler documentation](https://support.anthropic.com/en/articles/8896518-does-anthropic-crawl-data-from-the-web) (Anthropic)
- [Google AI features and your site](https://developers.google.com/search/docs/appearance/ai-features) (Google)

---

*Written by [Piotr Litwa](https://piotrlitwa.com/about.html), independent GTM & Analytics specialist.*