I rebuilt my website
three times in one week.
My wife's version won.

I bought faaizjaved.com on a Tuesday. By Sunday I had built it three times, thrown two of them away, and learned more about AI tools from the failures than from anything that worked.

Here is the honest version, because the useful part is the bit most people leave out.

Attempt one: the site that looked like a machine made it

I started the way everyone starts. I described what I wanted, and the machine gave me a website. It had a centred hero. It had three service cards with a thin coloured rail on top. It had a strip of numbers counting up, and a serif headline on a warm cream background.

It was competent. It was also instantly recognisable as a thing that was generated rather than designed, and my first reaction when I opened it was not "this is me", it was "I have seen this site fifty times".

That is the part nobody warns you about. AI does not produce random output, it produces the average of everything it has seen. Ask for a personal site and you get the centre of the distribution: tidy, safe, and interchangeable with every other site made the same way that week.

Attempt two: a good idea, still not the answer

So we went the other way and built something with a real point of view. A weekly dispatch. Editorial type, a status ledger of everything I run, travel entries with stamps, reviews printed as receipts. A Telegram bot that collected what I built each week, wrote the update, and waited for me to approve it before anything went out.

I liked it. I still like it. But it answered the wrong question. It was built to grow an audience, and what I actually needed was for a founder in Texas or Manchester to land on the page and understand in ten seconds that I can fix their problem.

Attempt three: my wife sent a plan

She had gone away and written the whole thing out. Six pages. Positioning as a partner rather than an agency. Services grouped by the problem the client has, in the client's own words, not by the tools I use. Case studies that end in results instead of adjectives. A project form that asks what you need and what your budget is, instead of the usual name, email, message.

Then she sent a design to match it.

It was better than both of mine. Not because it was fancier, but because it was aimed at somebody specific. Mine were aimed at "an audience". Hers was aimed at a person with a budget and a problem, which turns out to be the only audience a services site actually has.

So we shipped hers, and I spent the rest of the week filling it with real things.

What actually took the time

Not the code. The code was the fast part. What took the time was the truth.

Deciding what I am allowed to publish. I am COO at an e-commerce group. I have built systems for them that I am proud of. None of that belongs in my portfolio without the founder's agreement, so the blog engine I built lives on my site as a capability with the client unnamed, and their product images stay off it entirely. Same rule for an old client app: described, not named.

Cutting the work down. My portfolio had 47 pieces. That is not a portfolio, that is a hard drive. One character I had generated appeared nineteen times, because I was proud of it, which is the worst reason to publish anything. It is now four pieces, and the gallery is better for it.

Killing the tells. Em dashes everywhere. Captions that described what was in the photograph instead of saying something. Headlines that shouted. Every one of those is a small signal that a machine wrote the page and nobody read it afterwards. I read it afterwards.

What I would tell anyone building with these tools

Direct, do not request. The difference between my first site and the third one is not the model. It is that somebody decided what it should be before anyone started typing.

Your defaults are the enemy. If a thing appears without you choosing it, question it. That goes for layouts, colours, stock phrasing and captions.

Consistency is the real skill. Anyone can generate one good image. Generating the same character across twelve shots, in different light, at different hours, and still recognisably the same person, is the job. That is what I put at the top of my gallery.

Publish what is true. Every number on my site is one I can defend. Every project is one I actually own. The client work that is not mine to show is described, not displayed. A portfolio that impresses a prospect and then collapses under one question is worse than no portfolio.

Show the work, not the prompt. Nobody is hiring you because you can write a prompt. They are hiring you because you knew which of the forty results was the right one.

Where it landed

faaizjaved.com is live. It has the services, the case studies, a studio gallery of 28 pieces with the tool credited on each one, and a form that asks the useful questions.

It took a week, three attempts, and being told by my wife that her version was better. She was right.

If you have a problem that fits any of this, the form is at faaizjaved.com/contact. Tell me what is broken, not what you want built. That is a better place to start anyway.