AI SEO · LLM SEO
LLM SEO that makes your site legible to machines.
A language model can only work with the page it actually received. LLM SEO is the engineering that gets your content past the crawler, into the model, and through retrieval with its meaning intact.
Which AI crawlers are allowed through?
- A GPTBot Allow
- A ClaudeBot Allow
- A PerplexityBot Allow
- D Google-Extended Disallow
- D CCBot Disallow
Get seen on the engines your buyers actually ask
Our LLM SEO strategy
-
01
Open the right doors
Decide which AI crawlers to allow, and what allowing each one actually buys you.
-
02
Render without JavaScript
Serve a model the same content a browser sees, not an empty shell.
-
03
Survive the chunking
Structure pages so meaning holds when they are split into retrieval passages.
-
04
Keep one version of the truth
Remove contradictions so a model never has to guess which of your pages is right.
LLM SEO is the technical practice of making a website retrievable and correctly interpreted by large language models and the crawlers that feed them, covering crawler access, server-rendered HTML, page structure that survives being split into retrieval passages, and facts that stay consistent across every page.
Most AI visibility problems are plumbing problems
Before a model can decide whether to recommend you, four mechanical things have to go right. When a brand is missing from AI answers, one of them has usually failed silently.
The crawler never got in
A robots.txt line written in 2023, a bot-management rule on your CDN, or a firewall that treats unfamiliar agents as attacks. The block is usually accidental, and nothing in your analytics reports it.
The page arrived empty
Many AI crawlers read the raw HTML response and stop there. If your copy is assembled by JavaScript in the browser, what the model stored may be a shell with a loading spinner in it.
The passage lost its meaning
Retrieval cuts pages into short chunks. A claim in paragraph nine that depends on a heading in paragraph two becomes an orphan sentence with no subject once the two are separated.
Five agents worth an explicit decision
These are real user agents with published documentation, and they do different jobs. Allowing all of them is a choice. So is the blanket disallow somebody added last year without telling marketing.
| Agent | What it feeds | What blocking costs you |
|---|---|---|
| GPTBot | OpenAI’s crawler for web content used in model training. ChatGPT search and user-triggered fetches run under separate agents. | Your pages stop entering future OpenAI training data. Search inclusion is governed by the other agents, so blocking this one alone does not remove you from ChatGPT search. |
| ClaudeBot | Anthropic’s crawler for web content used in training Claude. Live fetches and search use separately named agents. | Claude has less of your own material to draw on when it answers from what it learned rather than from a page it fetches during the conversation. |
| PerplexityBot | Perplexity’s search index, which is what the product queries when it answers and links out to sources. | You can fall out of the pool Perplexity retrieves from and links to. On that platform, being in the index is the route to being cited. |
| Google-Extended | A robots.txt token rather than a crawler. It controls whether content Googlebot already fetched may be used to train and ground Gemini. | Nothing about your Google Search indexing or ranking changes. What you give up is presence in the Gemini surfaces the token governs. |
| CCBot | Common Crawl, a public web archive that many model builders and researchers use as a training source. | Future crawls stop while existing snapshots stay where they are. One line here quietly affects many downstream datasets at once. |
Who LLM SEO is for
This work pays for itself fastest where the technical gap is wide and the content is already good enough to win if a model could read it.
Sites built on client-side frameworks
React, Vue and similar stacks can deliver an almost empty document to any client that does not run scripts. Your visitors see a full page. A crawler that skips JavaScript sees markup with nothing in it, and that is the version the model keeps.
Brands models describe with old facts
Retired pricing, a former office, a product name you changed. Training data preserved a version of you that no longer exists, and correcting the live page is only the first half of the fix.
Large catalogues and documentation sets
Thousands of pages means thousands of chances for two of them to state different things about the same product. Models resolve contradictions by picking one version, and they do not ask you which.
Teams whose security stack blocks by default
Bot management is doing its job when it challenges unknown agents. The problem is that AI crawlers arrive looking exactly like the traffic those rules were built to stop, so the allowlist has to be deliberate.
How our LLM SEO process works
Six steps, run in order, because each one is only worth doing once the step before it stopped failing. Every step leaves evidence you can check yourself.
-
1
Crawler access audit
We read your server logs for each AI agent, then trace every rule that could be stopping them: robots.txt, CDN bot rules, firewall policy and rate limits.
-
2
Render parity check
We fetch your key pages the way a non-executing crawler does and diff that against the rendered DOM. The gap between the two is the content models never see.
-
3
Chunk simulation
We split your pages the way retrieval systems do and read each passage on its own. Anything that stops making sense in isolation gets restructured until it stands up alone.
-
4
Canonical fact sweep
We collect every version of your core facts across the site, pick the true one, and rewrite the rest to match so no model has to choose between two of your own pages.
-
5
Machine-readable surface
Clean semantic HTML, accurate sitemaps, structured data that agrees with the visible copy, and an llms.txt file pointing to the pages you most want read.
-
6
Recrawl and monitoring
We watch AI agent hits in your logs, confirm the corrected pages were refetched, and keep checking that a deploy or a firewall change has not quietly closed the door again.
What you actually receive
Files, configurations and documents your engineers can act on without us in the room. This is the foundation that Answer Engine Optimization and your organic SEO both stand on.
The crawler access policy
- A rewritten robots.txt with every AI agent named explicitly
- The reasoning behind each allow and each disallow, in writing
- CDN and firewall rule changes your team can review before shipping
The rendering report
- Raw HTML against rendered DOM for every template you run
- The exact content missing from the non-executed response
- A prioritised fix list for server rendering or prerendering
The retrieval chunk map
- Your key pages broken into the passages retrieval would produce
- Passages flagged where meaning collapses without their context
- Rewritten headings and blocks that survive being read alone
The canonical fact sheet
- One agreed statement of each fact about your business
- Every contradiction found on site, with the page it lives on
- An llms.txt file and a plan for keeping it current
“We started with a free strategy that showed us specific gaps in our campaigns. Today, LineUp is our regular partner, delivering a measurable return on every invested zloty.”
What clients say on Google
Unedited screenshots from our Google Business Profile. Swipe, or use the arrows, to read them all.
Read these reviews on Google. Reviews written in Polish or Turkish are shown in Google’s own translation.
Questions we get asked
What is LLM SEO?
LLM SEO is the technical work of making a website retrievable and correctly interpreted by large language models and the crawlers that supply them. It covers crawler access, server-rendered HTML, page structure that survives chunking, and facts that stay consistent everywhere they appear.
Should we block GPTBot and ClaudeBot?
That depends on what you sell. Blocking those agents keeps your pages out of future training corpora, and it also removes the material models draw on when they describe your category. Most brands that want to be recommended allow them and control the message instead.
Do language models run our JavaScript?
Often they do not. Many AI crawlers fetch raw HTML and never execute client-side scripts, so anything your framework renders in the browser can be invisible to them. Server-side rendering or prerendering puts the same content in the initial response.
What is llms.txt and does it work?
llms.txt is a proposed markdown file at your root that points models to your most useful pages in plain text. No major vendor has committed to reading it. We ship it because it costs little, and we never let it replace crawlable HTML.
Why do models still quote our old pricing?
Training data is a snapshot. A model that learned a fact two years ago repeats it until something newer and better corroborated overwrites it. Fixing the live page is step one. Getting the correction restated across the sources models retrieve is what actually moves the answer.
How is LLM SEO different from AEO and GEO?
LLM SEO is the infrastructure layer: access, rendering, structure and fact consistency. Answer Engine Optimization works on the answer a model gives to one question, and Generative Engine Optimization works on the longer synthesised passages. Both depend on the plumbing this page covers.
Find out what the models can actually read.
We will fetch your site the way an AI crawler does, check which agents are getting through, and send you the list of content that never made it into the response.
Last reviewed 28 July 2026.