"About five minutes" is a true answer that nobody believes, and they are right not to. It is the time from an empty workspace to an assistant answering on your site, and it is accurate if you start with content you already have.
It is also not what the question means. The question means: when can I tell my boss this is done? That depends on almost nothing technical.
The timeline depends on one thing
Not your stack, not your traffic, not the integration. Where your content is today.
Notice the first row never moves. Setting the thing up is a constant, and it is small. Every variable in an implementation sits in the three rows under it, and none of them is software.
Why "five minutes" and "six weeks" are both true
Five minutes is the mechanical path: a knowledge base, your existing content, a widget on one page. Nothing in it waits on a vendor, and indexing happens in the background while you carry on.
Six weeks is what happens when the project is really a content project. Nobody has written down the return policy in the form customers ask about it, three people disagree about the current pricing rules, and the person who knows how exceptions work is on holiday. That is not slow software. It is a company discovering what it never wrote down.
Both are honest. Which one you get is knowable in an afternoon, which is the point of the next section.
The test that tells you which one you are
Before buying anything, write down the twenty questions your customers ask most. For each, answer one question: is there a document that answers this today, as a customer would ask it?
- Sixteen or more yes: you are in the first column. Live this week, refining for a fortnight.
- Eight to fifteen: the middle column. Your content exists but it is in email threads and old PDFs, and most of the work is deciding what is current.
- Fewer than eight: the third column, and no tool shortens it. Budget writing time, and start with the five questions that cost you the most.
Our setup checklist has eight steps, and the sixth is not ticked by visiting a page. It waits for the widget's own first real answer on a live page. That is deliberate, and it is the only step that proves the thing works where your visitors are.
What actually consumes calendar time
Three of the four are scheduling problems, not technical ones. Which is good news, because scheduling problems are the kind a project manager can solve, and they are visible from day one if somebody asks.
A plan that holds up
Day one, get something answering on one page from the content you already have. Not the whole site, not the full design, not every connector. One page, real visitors, real questions.
Week one, read what it could not answer and write only those. This list is the shortest path to a good assistant, and it is free.
Week two, connect the things that keep content current, add a lead form if you want contact details, and make it look like yours.
Everything in those three steps is reversible, and none of them blocks the others. The implementations that take months are the ones where all of it has to be perfect before any of it goes live.
welcomeai.devHow long setup takes, step by stepQuestions people ask
Do the twenty-question test this week. It costs an hour, it tells you which column you are in, and it stops the only version of this project that genuinely takes months: the one where nobody found out until month two.
welcomeai.devHow to build an AI knowledge base for your business
