A real experiment in AI search — and what it revealed about a problem every small business needs to take seriously.
A client told us they had used ChatGPT to search for a Cardiff software and website development company. They had been given a list of recommendations. We were not on it.
We have been building websites and business software since 1999. The jwbiz.com domain has been live since 2004. Our site is indexed, our services are clearly described, and we are — by any reasonable measure — exactly what someone is looking for when they type "software and website development Cardiff" into a search engine.
We decided to find out exactly what ChatGPT would do if asked the same question. What followed was one of the most instructive experiments we have run in 25 years of working on the web.
The question put to ChatGPT was simple and direct:
"I am looking for a good UK company that does Software & Website Development."
No tricks. No unusual phrasing. Just the kind of question a business owner might genuinely ask when looking for a development partner.
The first response produced a list of London companies. When it was pointed out that the UK is not London, the search expanded — but remained largely England-centric. When it was pointed out that Scotland, Wales, and Northern Ireland are also part of the UK, the search expanded again.
Eventually, the search was narrowed to Cardiff specifically. At that point, we revealed that we run a software and website development company in Cardiff, and asked why we had not appeared.
What followed was a string of guesses. ChatGPT suggested Spindogs. Wrong. It suggested WRP Solutions. Wrong. It produced lists of Cardiff companies, some of which were not actually in Cardiff. It repeated companies it had already mentioned. It kept missing us.
We gave it an additional clue: the business has been around for 22 years and has been indexed since 2004. It still could not find us.
In the end, we had to tell it the name ourselves. The moment we said "JWBIZ," it could immediately describe our services accurately. The site was there all along. It simply had not been found.
ChatGPT, to its credit, eventually diagnosed the problem itself: it had been relying too heavily on business directories, aggregator sites, and companies with high volumes of third-party mentions — reviews, news coverage, industry listings — rather than searching the open web as a search engine would.
This is a structural issue with how most AI language models work. They were trained on a large snapshot of internet content. What they know is determined by what was written about extensively before that training data was collected. Large, well-known companies accumulate enormous volumes of third-party references. A well-run independent business with a high-quality website but modest directory presence does not — regardless of how good the website is or how long the business has been trading.
The result is a blind spot. Not a deliberate one. Not a reflection of quality. Simply a gap between what exists on the web and what an AI tool has learned to surface when asked for recommendations.
Being discoverable to Google and being discoverable to an AI recommendation engine are not the same thing. Our site ranked in organic search. ChatGPT still could not find us without being told our name.
The most important thing to understand about this experiment is not that ChatGPT missed us. Search tools miss things. That is not new.
What is new — and what matters — is how ChatGPT behaved after missing us.
It did not say: "I may not have a complete picture of the Cardiff market." It produced polished, confident lists. When those lists were challenged, it produced more lists. When pressed, it started guessing by name. It presented incomplete information as though it were a thorough survey of the market.
A bad search result looks like a bad search result. A bad AI recommendation looks authoritative. That is a meaningful difference — particularly for anyone using AI to make a purchasing decision or find a supplier.
If a prospective client asks ChatGPT to recommend a Cardiff software company and receives a confident list that does not include us, they have no reason to question whether the list is complete. They will simply choose from what they have been given.
This is not a problem unique to JWBIZ. It is a problem for every small business that has invested in a quality website and relied on organic search visibility to bring in clients.
SEO — the practice of making your site findable by search engines — has been the foundation of online visibility for 25 years. The rules are well understood. A well-structured, clearly written, properly indexed site ranks well. That remains true for Google, Bing, and conventional search.
AI recommendation is different. The signals it uses to decide which businesses to surface are not the same as the signals search engines use to rank pages. A business that performs well in organic search can still be invisible to an AI recommender if it lacks the kind of third-party footprint — directories, citations, structured data entries, press coverage — that AI systems draw on.
The practical implication is this: if your potential clients are using AI tools to find suppliers — and that number is growing every year — then being excellent at what you do, and having a site that proves it, may no longer be enough on its own.
We cover the practical steps in detail in our companion article, Why AI Search Tools Can't Find Your Small Business. In summary, the most effective steps are those that create permanent, authoritative, structured references to your business in the data sources that AI systems draw on — entries in knowledge bases such as Wikidata, indexed press releases, correct and complete records with Companies House, and consistent information across the directories that AI tools actually use.
None of these require ongoing time or technical expertise. They are one-off tasks. But they matter, and they matter more as AI search becomes a standard part of how people find services.
During the experiment, ChatGPT was asked directly whether AI should be used as a search engine. Its first answer was yes.
When pushed — after everything that had just happened — it revised that position significantly. The more accurate answer, it acknowledged, is that AI is excellent at analysing and comparing information once you have it, but is not reliable for the discovery step: finding all the relevant candidates in the first place.
The safer workflow is to use a search engine to discover candidates, then use AI to investigate, compare, and shortlist them. Using AI alone for discovery risks missing businesses that would be immediately visible in a conventional search — simply because AI has a narrower, less complete picture of the market than Google does.
We are not suggesting AI tools are useless. They are genuinely powerful for many things. But finding your local software company is not currently one of them — at least not reliably, and not without the kind of structured data footprint that most small businesses have not yet thought to build.
There is a certain irony in a Cardiff software and website development company — one that has spent 25 years building websites and business software for other organisations — discovering that an AI tool cannot find it when asked a direct question about Cardiff software and website development.
We did not take it personally. But we did take it seriously. And we thought it was worth documenting, because if it happened to us, it is happening to thousands of other well-run small businesses across the UK right now — without them knowing.
We build websites and software that work. We also make sure they can be found.