The Bike Shed Company

Where it started: March 2026

The Bike Shed Company makes bike sheds and garden storage. It was my second content client, through Rob Saunders at Splendid Web, who looks after its website. James Fulford, one of its two senior managers, was the person we worked with. The service was the same as for Five Education: articles each month, researched and checked, to help the business show up in Google and in AI assistants. It started at the end of March.

April: the articles James shaped

The first pieces were on premium garden storage. On 3 April James sent detailed feedback by email about the bespoke side of the business and the craftsmen behind it, and that became an article on how a bespoke shed is designed. We talked it through on a call on 9 April, and the first article went live.

That part worked well. James knew his customers, and the articles were better for his steer. But every steer was his time, and he had very little of it.

May and June: the slow bits

Each month we sent a short questionnaire by email and waited for answers. Access to the site's Search Console data, which shows how pages do in Google, never came through: Rob couldn't add users to it and James didn't have the time. So we worked from Google Analytics alone.

We also sent James a monthly report: what was published, how it did, and what we proposed next. The June report went to him on 3 July with the July questionnaire attached for his sign-off. Each one asked a little more of his time.

In June we found a fault of our own. Our articles linked to the site's www address while the site itself lived at the plain address, so its content system wouldn't connect the links. We fixed it on 16 and 17 June. James approved the June articles on 18 June ('Please put them live'), and Rob published one on bike storage for small front gardens that day and one on caring for a wooden shed on 22 June.

July and August: the numbers

The articles ranked as well as you'd expect. But The Bike Shed Company already has a large, busy website. In August its articles brought 0.37% of the site's traffic. Five Education's brought 6.2%. When we checked AI answers in August, The Bike Shed Company had 0.9% of the mentions in its category, sixth of the six brands we tracked. The business was already visible to AI assistants, so a few more articles couldn't show a clear change.

We measured AI answers the same way for every client: 20 test questions a customer might ask, put to the AI assistants each month, counting how often the business is named. For Five Education that count went from nothing to four. For The Bike Shed Company it barely moved.

12 and 13 August: James asks to stop

On 12 August James emailed: "I don't think I have the bandwidth to properly client this service given the small impact it's having. I'm afraid unless the picture has changed very considerably I want to stop."

I replied with the July figures, which had improved, and asked him to decide either way. He read them and came back the next morning: "I think the service is not a good match for our business or my capabilities to client and I'd like to stop. Thanks very much for your efforts."

His reasons were clear and fair. The articles ranked fine but were about 0.1% of the site's clicks. The AI benefit couldn't be proved, so the service read as a defensive move. He found the articles long and boring for a human reader. And the real cost was his time as one of two senior managers, which mattered more than the fee. We cancelled the subscription and refunded the last payment, and I thanked him for feedback that was more useful than he probably knew.

What I changed because of it

I took three things from it. The first is who the service is for. Small, specialist sites, where a few good articles visibly move the numbers and the value shows early. Five Education's same-day articles earned 195 clicks against none at The Bike Shed Company, on very different sites.

The second is time. The service now asks for nothing each month beyond a quick look: everything we need is gathered once, at the start. The third is the writing itself. Every article has two jobs, being found and being a good read for the client's own customer, and the second one now comes first. James's line about AI and people sits at the top of how we judge a draft.

If you're weighing up a content service

Ask what share of your traffic a few new articles could realistically become. On a big site the honest answer may be a rounding error, and that's worth knowing before you pay for it.

Ask how much of your time it needs each month and how the supplier will prove the value, especially any claim about AI assistants. If they can't show you a before and after, you're buying a hope. And read one of their articles as your own customer would. If it bores you, it'll bore them.

Where it is now

The engagement ended in August, on good terms. In October I asked James whether I could tell this story honestly, and he said yes. I'm grateful, because the client who leaves teaches you more than the one who stays.

In their words

“The stuff that makes them good for AI, makes them less good for a human.”

James Fulford, The Bike Shed Company

Questions people ask

Why did The Bike Shed Company stop the service?

James Fulford said the impact was small for the time it asked of him. The articles ranked fine, but they were a tiny share of a large site's clicks, the gain in AI answers couldn't be shown, and he found the articles long for a human reader.

Who is AI content writing a good fit for?

Small, specialist sites where a few good articles visibly move the numbers, and owners who want it done for them. In August 2026 Five Education's articles took 6.2% of its site's traffic against 0.37% at The Bike Shed Company.

Can an article be good for AI search and for people at the same time?

It has to be. James's point was that what made the articles good for AI made them less good for a person. Since August every article is written to be a good read for the client's own customer first.