Knowledge Hub in Alzaro

AI can write in any voice. Getting it to write in yours — and only claim what you can prove — is the actual work.

Knowledge Hub is where your business stops being something you explain over and over and becomes something the platform knows: who you are, how you sound, and which claims have evidence behind them. You set it up once. Every AI feature reads it from then on.

Why this exists

Three things go wrong when an AI does not know you

Everything reads like everyone

Ask any AI to write about a plumbing company and you get the same paragraph every plumbing company gets. Nothing in it is wrong. Nothing in it is you either.

It invents things confidently

Award-winning. Trusted by thousands. Guaranteed results. An AI will write those about you without being asked, and it has no idea whether any of them are true.

The real answers live in one head

Your licence number, your actual service radius, the case study you always tell. None of it is written down anywhere a machine can reach, so every tool starts from zero.

None of these are solved by writing a better prompt each time. They are solved once, by giving the machine a file it can read before it writes.

How it is organised

Five stages, in order

Each stage hands its result to the next. You cannot define a voice before you know which facts are true, and nothing activates downstream until it has been verified.

01

Collect

Where does what we know actually live?

Your website, your price list, your certificates, your growth report. Each one becomes an evidence source with a name attached, so every later claim can point back at something.

02

Verify

Which of these are true enough to publish?

The Hub proposes facts. You confirm them one at a time. Nothing you have not confirmed ever reaches a customer, and that single gate is the whole point of the module.

03

Define

Who are we, and how do we sound?

One sentence for what you are, one for why you are different, plus the tone words you want and the phrases you never want to see. This is what makes the output sound like you.

04

Activate

Where does all of this show up?

Nowhere you have to visit. Every AI feature on the platform reads this profile before it writes, and machine-readable versions get published for search engines and AI assistants to fetch.

05

Maintain

What happens when the truth changes?

Prices move, certificates lapse, a page gets rewritten. Facts expire on a schedule, conflicts open themselves, and anything that drifts turns into a task with your name on it.

The pages you will use

Eight tabs, and what each one is for

01 · Collect

Business picker

Which business am I working on?

One separate knowledge file per business, never merged. Each card shows verified facts, how many are waiting on you, and how complete the file is across five dimensions.

Evidence

Where did that claim come from?

Five ways in: a typed note, a website address we read for you, an uploaded PDF or Word file, your Growth Blueprint report, or the copy from your own microsite.

02 · Verify

Facts

What is waiting on my judgement?

A review queue sorted so the most confident suggestions come first. Approve one at a time or in bulk, retire the ones that stopped being true, and settle disagreements between two sources.

03 · Define

Profile

Can an AI say who we are in one sentence?

Five short fields: what you are, why you are different, who you serve, where you serve them, what you sell. Paste your website address and the Hub drafts all five for you to accept field by field.

Voice

Does this sound like us?

A brand archetype, the tone words you want, the phrases you never want printed, writing samples the Hub can reverse-engineer, and four hard claim limits it will not cross.

04 · Activate

Outputs

Can a machine read our business?

Six machine-readable artifacts built strictly from verified facts. Generate, read the warnings about what is still missing, then publish.

05 · Maintain

Actions

What needs me this week?

Tasks the platform opened by itself: an unbacked claim found on your live microsite, a fact that just expired, a published artifact that went out of date.

Activity

Who changed what, and when?

An append-only log of every verification, retirement, conflict and publish. Useful when an agency or a teammate manages the file alongside you.

Open Knowledge Hub

It opens the workspace and will ask you to sign in first.

The mechanism that matters

One gate stands between a suggestion and your customers

Everything arrives as a suggestion — including everything the AI extracted for you. A suggestion is invisible to the rest of the platform. It becomes usable at the exact moment a person approves it, and not before.

Your business profile Website we read Documents you upload Growth Blueprint report Notes you type Evidence You approve Facts tab Verified Reaches every AI feature Not yet verified Stays in your queue, goes nowhere

The AI proposes, it does not decide

Extraction saves you the typing. It never grants itself approval, and for the riskiest categories it is not even allowed to.

Two sources disagreeing is a question, not a coin flip

When a new value contradicts one you already approved, both are shown side by side and you pick the winner. Nothing gets silently overwritten.

Removing the evidence removes the standing

Delete a source and the facts it supported drop back to needing review. The value is kept so you do not lose work — the approval is not.

Where the bar is higher

Nine categories need two keys, not one

These are the claims that get a business in trouble — the ones a regulator, a competitor or an unhappy customer will ask you to substantiate. So they cannot be approved automatically, and they cannot appear in anything public unless the evidence behind them is something you have marked as publicly citable.

Awards Certifications Revenue Customer counts Case results Legal Medical Financial Pricing guarantees

Key one: a person approved it

Bulk approval skips these on purpose, and the automatic extractor is refused outright. Someone has to look at the certificate.

Key two: the evidence can be cited

Every source has a publicly citable switch, and it starts off. A private contract can prove something to you without becoming something we are willing to point at in public.

Four lines the writing never crosses

Set on the Voice page, applied everywhere afterwards.

No awards we cannot evidence
No client numbers we cannot evidence
No guaranteed outcomes
No pricing promises we have not checked

The payoff

Fill it in once, and ten places change

There is no button to push this anywhere. The features below already read your file before they generate, so the improvement shows up the next time you use any of them.

Social captions Local campaign plans Review replies Blog articles YouTube to Blog SEO content fixes Your microsite Landing page copy AI answer nodes Copilot answers Knowledge Hub One file per business

It writes from your file

Before an AI feature writes a word, it is handed your definition, your positioning, your tone words, your banned phrases and your verified facts. You can read that exact instruction on the Voice page — nothing is hidden from you.

It flags what you cannot back up

The same file is used in reverse. If a draft claims an award, a client count or a guarantee with no verified fact behind it, the claim gets marked in the editor — and on a live microsite or an automatic post, it opens a task instead.

And six files written for machines, not people

Search engines and AI assistants do not read your homepage the way a customer does. The Outputs tab publishes a plain summary for AI crawlers, structured data about your organisation and location, a question-and-answer file, a brand summary and an about paragraph — all assembled strictly from facts you approved. Publish them and your microsite and answer pages serve them automatically.

What happens over time

A fact has a life, not just a value

The reason most brand guidelines rot is that nothing forces anyone to look at them again. Here, the facts most likely to go stale come back to you on their own.

Suggested Needs review Verified Retired you approve you retire it it expires, or its evidence is removed Conflicted you choose the winner

How long an approval lasts

When the clock runs out the fact returns to your queue and a task opens. It is not deleted and nothing breaks — you are just asked to confirm it is still true.

Kind of fact Comes back after Reasoning
Prices 90 days The number most likely to be quietly wrong.
Pricing guarantees 90 days A promise you have stopped honouring is worse than no promise.
Contact details 180 days Numbers, addresses and hours drift without anyone noticing.
Certifications 365 days Most renew annually, so we ask annually.
Awards 365 days Last year is fine to mention. Five years ago needs a date on it.
Everything else Never expires Your founding year is not going to change. We do not invent chores.

Knowing when you are done

Five dimensions, and a number for each

A knowledge file is never finished, which makes it easy to stop halfway and never come back. So each dimension carries a target — the number of approved facts at which an AI has enough to write about that part of your business without reaching.

Founder and authority
10 approved facts

Who runs this, how long, what qualifies them.

Products and services
12 approved facts

What you sell, how it works, what it costs.

Customer problems
10 approved facts

What people actually ask before they buy.

Proof and case studies
8 approved facts

Results, credentials, things you can show.

Local trust
6 approved facts

Where you are, who vouches for you locally.

Around 46 approved facts covers a business properly. Most owners get there across a few sittings, not one.

Your progress against these five shows on the Knowledge Hub home page and on your main dashboard, so a half-finished file is visible rather than forgotten.

Getting started

Two days of setup, then a weekly glance

You do not have to finish this before the rest of the platform is useful. Every fact you approve improves the next thing you generate, starting with the first one.

Day 1

Let it read your website first

  • Open Knowledge Hub and pick a business
  • Paste your website address on the Profile page and press Fetch
  • Accept the drafted fields one at a time, or fix them as you go

Editing a form that is already sixty percent filled is a different job from staring at a blank one. Your contact details are already in there before you arrive — they were captured the moment the business was created.

Day 2

Add what the website does not say

  • Upload your price list, certificates and any FAQ document
  • Import your Growth Blueprint report if you have run one
  • Type in the two or three things you always say on the phone

A website is marketing language. The facts worth quoting are usually sitting in an internal document nobody has ever shown a machine.

Week 1

Empty the queue, then set the voice

  • Work down the review queue and approve what is true
  • Settle any conflicts by choosing which value wins
  • Generate a voice baseline, edit the tone and banned phrases, save

Until you approve them, none of these facts leave the Hub. An unworked queue is the single most common reason the rest of the platform still sounds generic.

Ongoing

Publish, then let it keep watch

  • Generate the six artifacts and publish the ones that look right
  • Check the Actions tab once a week
  • Re-approve facts as they come up for renewal

This is the part that keeps paying. You are not maintaining a document — you are answering the occasional question the platform raises when something it depends on has gone stale.

Being straight with you

What this is not

Verified means a person verified it

We are not checking your claims against the outside world. We are checking that what the platform says about you traces back to evidence you supplied and approved. That is a narrower promise than fact-checking, and it is the one we can actually keep.

The published files are assembled, not written

The six machine-readable outputs are built directly from approved facts rather than generated by an AI. That is deliberate — it means they can never contain something you did not approve. It also means that with a thin file they come out short, and the page will tell you exactly what is missing.

Limits worth knowing before you start

LimitWhat it means for you
Flagged claims are not blocked When a draft makes a claim you have not backed up, it gets marked and a task opens. Generation still finishes. A tool that refuses to let you publish is a tool people route around.
Websites that need JavaScript to show their text We read the page as it is delivered. If your content only appears after scripts run, the read comes back empty — upload a document or type the notes instead.
Local and internal addresses For security, we refuse to fetch addresses that resolve inside a private network. Staging sites on internal domains have to be imported as files.
Nothing is shared between businesses Each business keeps its own file. If you run several similar locations, the facts are not copied across — which is safer, but it is real work each time.
The reading is a first draft Extraction from a website or document gets you started; it does not get you finished. Expect to correct wording and fill gaps, because it can only find what the source actually said.

Start with your website and ten minutes

Pick a business, paste your address, and let it draft the file. Approving the first handful of facts is enough to feel the difference in whatever you generate next.