SEO Is Dead. Here's How to Optimize Your Site for LLMs Instead.
Let me be upfront: SEO isn't completely dead. You still need it. But if you're building a personal site, a portfolio, a blog, or a small SaaS — and your content strategy is still centered around Google rankings — you're optimizing for the wrong audience.
The audience has shifted. And it's shifted fast.
What Actually Changed
For the past 20 years, the funnel was simple: user has a question → types it into Google → clicks a link → reads your page. You optimized for that click. Title tags, meta descriptions, backlinks, keyword density — all of it existed to serve that one moment.
That funnel is breaking.
Users now increasingly ask ChatGPT, Claude, Perplexity, or Gemini instead of Googling. They don't get a list of links — they get an answer. And buried in that answer, sometimes, is a mention of your site. Or your name. Or your project.
You didn't get the click. But you got the reference.
The question now is: how do you get referenced?
How LLMs Actually "Know" Your Site
This is where most people get it wrong. They think LLMs work like search engines — crawl, index, rank. They don't.
LLMs know your content in a few different ways:
1. Training data cutoffs
GPT-4, Claude, Gemini — they were trained on a snapshot of the web. If your site existed and had meaningful content before the cutoff, there's a chance it's in the weights. This is not something you can directly optimize for in real time.
2. RAG-based systems (Retrieval-Augmented Generation)
Perplexity, ChatGPT with Search, Bing AI — these crawl the live web and pull context in real time. This is where current, structured, crawlable content matters enormously. This is your real target.
3. Citations and reputation signals
If other credible sources (GitHub READMEs, blog posts, forum threads, documentation) mention your content by name or URL — LLMs will reference you more. It's not backlinks for PageRank. It's mentions for plausibility.
The Structural Difference Between SEO and LLM Optimization
SEO rewards keywords and authority signals.
LLM optimization rewards clarity, structure, and factual density.
A page stuffed with keywords to rank for "best redis caching tutorial" might still rank on Google. But when Claude reads that page to answer a user's question, it doesn't care about keyword density. It cares about whether your explanation is actually clear and correct.
LLMs are, in a weird sense, better readers than search engine crawlers. They understand context. They prefer concise, structured, accurate prose over keyword-saturated fluff.
This is good news for people who write well and bad news for SEO farms.
Concrete Things That Actually Help
1. Write in Declarative, Parseable Prose
LLMs extract meaning sentence by sentence. If your content is buried in long rambling paragraphs, it gets diluted or misrepresented.
Write like you're answering a question directly. Lead with the point, then explain it.
Bad:
Redis, which is an in-memory data structure store that many developers use in production environments, can be quite useful when you need to cache things...
Better:
Redis stores data in memory. It's fast because there's no disk I/O. Use it when you need sub-millisecond reads and can tolerate data loss on restart.
The second version is extractable. An LLM can pull that out and quote it accurately.
2. Use Semantic HTML Properly
This applies to both traditional crawlers and LLM-augmented search. Your page structure signals what matters.
<article>
<h1>Why Redis Beats Memcached for Session Storage</h1>
<section>
<h2>Persistence</h2>
<p>Redis supports RDB snapshots and AOF logging...</p>
</section>
<section>
<h2>Data Structures</h2>
<p>Memcached only supports strings. Redis supports...</p>
</section>
</article>
Compare that to a wall of <div> soup where your content is structurally indistinguishable from your navigation and footer. Crawlers — and LLMs — have to guess what's content and what's chrome.
Don't make them guess.
3. Add a /llms.txt File
This is new and not yet universally adopted, but it's gaining traction. Similar to robots.txt for crawlers, llms.txt is a plain-text or Markdown file at your root that gives LLMs a structured summary of your site:
# TIROK — Alireza Ghotbi
Backend developer focused on Go, Python, and self-hosted infrastructure.
## Content
- /blog/ — Technical blog posts on Go, FastAPI, Redis, Docker, and Linux
- /projects/ — Open source projects including calcirok, music.tirok.ir
- /about/ — Background and stack
## Key Posts
- /blog/redis-vs-postgres-caching — When to use Redis vs Postgres for caching
- /blog/fastapi-telegram-integration — Building a FastAPI backend for Telegram bots
## Contact
dev@tirok.ir
There's no official standard yet, but Perplexity and some other RAG-based systems are starting to look for it. It costs you maybe 20 minutes to write one. Do it.
4. Make Your Content Citable
LLMs cite things they can attribute. That means:
- Have a clear author — not "admin" or "staff". Your real name and some verifiable identity (GitHub, LinkedIn, portfolio).
- Date your posts — undated content feels untrustworthy to both LLMs and humans.
- Use canonical URLs — if the same content lives at multiple paths, pick one and canonicalize it. LLMs don't want to cite an ambiguous source.
In your HTML <head>:
<meta name="author" content="Alireza Ghotbi">
<meta property="article:published_time" content="2026-11-08T00:00:00Z">
<link rel="canonical" href="https://portfolio.tirok.ir/blog/your-post-slug">
Structured data (schema.org JSON-LD) also helps RAG systems understand your content's metadata without having to parse prose:
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "BlogPosting",
"headline": "Why Redis Beats Memcached for Session Storage",
"author": {
"@type": "Person",
"name": "Alireza Ghotbi",
"url": "https://portfolio.tirok.ir"
},
"datePublished": "2026-11-08",
"url": "https://portfolio.tirok.ir/blog/redis-vs-memcached"
}
</script>
This is not new — it's been best-practice for years. But its value just went up considerably.
5. Write Content That Answers Questions, Not Performs Keywords
This is the hardest shift if you've spent years doing SEO.
Old mindset: "I want to rank for 'FastAPI Redis caching'."
New mindset: "Someone will ask an LLM 'how do I cache responses in FastAPI'. I want my post to be the best answer to that question."
The difference is subtle but it changes your whole structure. Instead of building a page about a keyword, you're building a resource that directly answers a question.
FAQ sections, explicit problem/solution framing, numbered steps for procedural content — these all help LLMs extract and represent your content accurately.
6. Don't Block Crawlers (Including AI Crawlers)
Check your robots.txt. A lot of people reflexively block everything that isn't Googlebot. That includes Perplexity's crawler, OpenAI's crawler, and others that feed into RAG systems.
# Don't do this if you want LLM visibility
User-agent: GPTBot
Disallow: /
User-agent: PerplexityBot
Disallow: /
If you're blocking AI crawlers because you're worried about content scraping — that's a real concern. But understand the tradeoff: you're making yourself invisible to the systems your potential readers are increasingly using.
A middle path: allow crawling, but add llms.txt to shape how your content is summarized.
What Doesn't Transfer from SEO
Some old SEO tactics are not just irrelevant now — they actively hurt LLM visibility.
Keyword stuffing — LLMs can tell when prose is unnatural. It degrades extractability.
Thin content pages — A page that exists solely to target a long-tail keyword with 200 words of fluff has nothing for an LLM to cite. It'll be ignored.
Backlink schemes — Backlinks built artificial authority for PageRank. LLMs don't use PageRank. They use the content itself plus signals from genuine mentions in real discussions (Reddit, GitHub, HN, etc.).
Meta keyword tags — Already dead for Google. Doubly irrelevant now.
The Honest Tradeoff
Here's what I won't sugarcoat: optimizing for LLMs right now is partly a bet on the future. The standards aren't settled. llms.txt might go nowhere. AI search might consolidate differently than expected.
But the underlying principle — write clearly, structure well, be attributable, be accurate — is not a bet. That's just good content. It worked before SEO existed. It works now. It'll work after the next paradigm shift too.
The difference is that the LLM era punishes content that optimizes for the algorithm at the expense of the human reader. For once, those two things are aligned.
Write for the human. The machine will follow.
If you're running your own blog or portfolio, this is worth implementing today. The technical lift is small — a few meta tags, a llms.txt, semantic HTML, and a writing habit adjustment. The payoff compounds over time.
# Comments (0)
No comments yet. Be the first.