AI Content Nodes
An AI cannot quote a page it cannot parse. So stop making it guess.
Your website was built for people and for Google — carousels, hero images, copy that unfolds as you scroll. When someone asks ChatGPT or Perplexity about your industry, none of that helps. The engine needs a short, sourced, self-contained answer it can lift with confidence. AI Content Nodes builds that layer for you, publishes it where the crawlers look, and then shows you which ones actually came.
The unit
One node answers one question, completely
A node is not a blog post and not a landing page. It is a small, strict record: the question, the answer in under 300 characters, the facts behind it, where those facts came from, and machine-readable markup describing all of it. Because the shape is fixed, an engine never has to work out what the page is about.
The question
Phrased the way a customer would say it out loud, and used as the page heading.
The direct answer
Under 300 characters, first thing on the page, tagged so a parser cannot miss it. This is the sentence that gets quoted.
The body
Short sections and lists that expand the answer without burying it.
The facts
Prices, hours, dimensions, turnaround times — as a definition list, not prose.
The citations
At least one claim paired with its source. Required — a node without one will not save.
The markup
A schema.org graph naming the entity and tying it back to your business, generated for you.
Why the shape is fixed
A model quoting you is taking a risk on your behalf. The easier it is to tell where your answer starts, where it ends, and what backs it up, the lower that risk gets. Rigid structure is not a limitation here — it is the entire mechanism.
These pages are noindex on purpose
Node pages tell traditional search engines not to index them, while still inviting them to follow the links. They are not competing with your website for rankings, and they will not create duplicate-content problems. They exist for the machines that read before they answer.
How it runs
Six stages, and you own the fifth
Everything up to review happens without you. Everything after it needs your approval. That boundary is fixed, and it is why nothing reaches the public web that you have not read.
Generation has a fallback
If the model is unavailable or your credits run out mid-batch, the node is still built from a deterministic template and flagged as template content, so the run never half-fails. You regenerate it later when you are ready.
Big batches go to the queue
Past twenty nodes the whole run moves to background processing and you get one notification when it settles, instead of a browser tab you cannot close.
Every version is kept
The last five snapshots of each node stay available. Roll back to any of them and the page re-renders from that content rather than restoring stale HTML.
The review gate
Five states forward, one dead end
A node walks one path to the public web, and only approved nodes can be published. Rejecting one takes it offline immediately. Published nodes cannot be deleted until you unpublish them first — the point is that nothing disappears from an engine index by accident.
The three switches that decide whether anything is public
Node pages are live only when all three line up: the feature is enabled for that business in Settings, the node has been approved and published, and — on the platform path — the business has a published microsite to host them. If your hub URL returns a 404, one of those three is off. It is almost always the first.
Where nodes come from
You have already written most of them
Before you type a single question, the blueprint reads what is already in your workspace and proposes a set of nodes. It marks each one as new or as an update to something that exists, you tick the ones you want, and only then does anything get generated.
Nothing here is generated until you select it. A preview costs nothing.
Eight kinds of answer
The type decides the markup
Picking a type is not a filing decision. Each one emits a different schema.org structure, which is how an engine knows it is looking at a price, a procedure or an opening time rather than an undifferentiated block of text.
FAQ
The questions people ask before they buy
FAQPage
Service
What you do, who it is for, what it costs
Service
Product
One product, described in facts
Product
Specification
Numbers, sizes, materials, compatibility
TechArticle
Policy
Returns, warranty, cancellation, privacy
CreativeWork
How-to
Step by step, in the right order
HowTo
Comparison
This against that, on the same axes
WebPage + ItemList
About
Who you are, where, since when
Organization
Quality, measured twice
Two scores, because they fail differently
A node can be technically perfect and still impossible to quote, or beautifully quotable and missing the markup that gets it read. Rolling those into one number would hide exactly the thing you need to fix, so they stay separate.
GEO readiness
Ten checks, equally weighted
Can a crawler take this page seriously? Every check is a yes or a no, and the score is simply how many passed.
- Exactly one H1
- A direct answer up front
- Valid structured data
- At least one citation
- Real sections, not a wall
- A visible last-updated date
- Under 30 KB
- No scripts
- Your business name present
- A way to contact you
Extractability
Five signals, weighted by impact
If a model lifts one passage out of this page and drops it into an answer, does it still make sense on its own?
Optimize draft
A deterministic pass that shortens an overlong direct answer, breaks a heavy paragraph into a list and fills in missing facts. No model, no credits, and it snapshots the node before it touches anything.
Structured data check
Validates the generated JSON-LD against what the node type promised — a how-to with no steps, or a comparison with nothing to compare, gets flagged before it goes live.
Quality-drop alerts
If a rebuild lands materially worse than the version before it, you get told rather than finding out from a chart three weeks later.
What gets published
One approval, six doorways
Publishing a node does not just create a page. It updates every file an AI crawler uses to find and enumerate your answers, because different engines discover content in different ways and you do not get to choose which one shows up.
/nodes
Answer hub
Every published node, grouped by type.
/nodes/{slug}
Node page
The answer, its facts, its sources, its JSON-LD.
/nodes/{slug}.md
Markdown twin
The same answer as plain markdown, for models that prefer it.
/nodes/sitemap.xml
Node sitemap
So a crawler can enumerate the set instead of guessing.
/llms.txt
llms.txt entry
One line per node: title, URL, and the answer in 220 characters.
/llms-full.txt
llms-full.txt
The whole knowledge base in one file, for a single fetch.
Built to be re-crawled cheaply
AI crawlers come back often. Every surface answers conditional requests, so a bot that already has the current version gets a small "nothing changed" reply instead of the whole page — which keeps them coming back rather than backing off.
Announced, not just posted
Newly published nodes can be submitted for indexing the moment they go live, so discovery does not depend on a crawler happening to wander past.
Where they live
Three ways to put these on the web
Start on the platform path today and move to your own domain when it matters. The nodes do not change — only the address does.
Platform path
Nothing to set up
Your nodes live under your Alzaro microsite. Works the moment you publish, needs no DNS, no server, no waiting.
The URL carries our domain, not yours.
Edge proxy
Your domain, your CDN
Keep your existing site exactly as it is and forward one path to us. We hand you the Cloudflare Worker and the nginx block, already filled in.
Someone has to paste a snippet into your CDN once.
Your root domain
We serve the domain
Point a verified domain at us and we serve its llms.txt, its node sitemap and every node page from the root.
The domain has to be verified first, and it serves only these files.
One rule worth knowing before you choose: AI crawlers only read llms.txt from the root of a domain. A copy in a subfolder is ignored. Both of the custom-domain options exist to satisfy exactly that constraint.
Proof, not vibes
Three stages you can actually count
Most AI-visibility tools stop at "we published something". These three numbers are measured from real requests and real answers, and the gaps between them tell you what to fix next.
Coverage
How many of your published nodes an AI crawler has actually fetched in the last 30 days.
Orphans
Published nodes nobody has come for yet, with their age — the list you work down.
Crawler intent
Split by why the bot came: training, live search, or someone asking about you right now.
Provider share
Which engines are reading you, and how often each one comes back.
Health score
Slow pages and error pages, scored — crawlers give up on both.
Spoof checks
Optional reverse-DNS on every claimed bot, so a fake GPTBot does not flatter your numbers.
How to read the gaps
Crawls but no citations usually means the answer is not quotable — check extractability and tighten the direct answer. Citations but no referral visits is normal and not a failure: plenty of people get their answer and never click. Neither number moving at all points at the surface itself, so start with whether your hub URL loads.
The four screens
Everything lives in one place
Library
Every node with its type, status, both scores and its crawl count. Filter, then act on twenty at once.
Open itGenerate
Preview blueprints from your existing content, tick what you want, generate it as drafts.
Open itSettings
The master switch, discovery files, render style, refresh window and where your nodes are served from.
Open itTraffic
Which bots came, what they read, what they skipped, and whether any of it turned into a citation.
Open itAnd the editor, where the work happens
Six tabs against a live preview of the finished page: write the content, read the quality report, move it through review, test whether an engine cites you, browse and restore earlier versions, or drop into the raw JSON if you want full control. The preview is the real rendered page, not an approximation.
Tool links open the workspace and will ask you to sign in first.
Worth knowing early
Four behaviours that surprise people
All four are deliberate, and all four are easier to meet in a feature page than in a support ticket three weeks from now.
Three switches, one surface
A node is public only when the feature is on for that business, the node is approved and published, and it is marked live. Miss one and the whole surface returns a 404.
Editing a live node takes it down
Any save sends it back to draft, which pulls it off the public surface until you approve and publish again. That is deliberate — half-edited answers should not be quotable.
Auto-refresh unpublishes
When a source fact changes, or the freshness window expires, the node rebuilds and goes offline pending your re-approval. Set the window to something you will actually revisit.
Drafts count against your quota
Your plan limit counts every node for the business, not just the live ones. Delete the drafts you abandoned.
Getting started
Ten good nodes beat eighty you never read
This is not a project. The first useful set takes an afternoon, and the ongoing work is a short review every couple of weeks.
First 30 minutes
Turn it on and see drafts
- Open Settings, pick the business, switch on public node pages.
- Choose where they live — the platform path works immediately.
- Run a blueprint preview. It reads what you already have and proposes nodes.
- Select eight or ten and generate. Long batches move to the queue and notify you.
Start with the sources you already filled in. If your business profile and microsite are written, you have enough for a first set without typing a question.
Week one
Make ten of them genuinely good
- Read each direct answer. It is capped at 300 characters, and it is the part that gets quoted.
- Check the citations. A node needs a source, and a wrong one is worse than none.
- Run the audit, then Optimize draft to fix the mechanical misses.
- Approve, publish, and load your hub URL to confirm it is really there.
Ten nodes you have actually read will out-earn eighty you generated and never opened. The engines quote specifics, and specifics are what a human has to check.
Ongoing
Work the gap, not the volume
- Check coverage and the orphan list on Traffic every couple of weeks.
- Create a tracking task so you find out whether engines quote you, not just fetch you.
- Re-approve nodes that auto-refresh took offline after your facts changed.
- Add nodes for the questions the citation gap report says you never answered.
Coverage tells you what is being read. Citations tell you what is being used. The second number is the one that moves revenue, and it only moves if you keep answering new questions.
What it costs
Only the model costs credits
| Action | Credits | Notes |
|---|---|---|
| AI node generation | 2 credits per node | Only when a model writes it. Regenerating later costs the same. |
| Citation test | 1 credit per test | Ask an engine a question and record whether it cited you. |
| Citation tracking run | 2 credits per run | The scheduled version of the same test, daily or weekly. |
| Everything else | Free | Blueprint previews, manual nodes, rendering, audits, optimize, version rollback, publishing, and every traffic report. |
Your plan also sets how many nodes each business can hold. Drafts count toward it.
What this does not do
Five things we will not claim
| Not included | Why |
|---|---|
| Publish anything on its own | A node cannot go live until you approve it. There is no auto-publish switch to forget about. |
| Guarantee a citation | You control what is extractable and verifiable. No one controls what a model decides to quote. |
| Rank you on Google | Node pages are deliberately noindex. They are an AI surface. Traditional SEO is a different tool. |
| Invent facts to fill a page | Every node needs at least one citation carrying both a claim and its source, or it will not save. |
| Work with no published site | On the platform path you need a published microsite. On your own domain you need a verified one. |
Answer ten questions properly
Turn it on, preview what your own content already suggests, and take ten nodes all the way to published. That is an afternoon, and it is enough to start seeing which engines come looking.