I came across an experiment recently that I have not been able to stop thinking about.
Somebody invented a deodorant brand. It does not exist. They spent about eleven dollars on a domain and roughly an hour building something. No fake reviews, no Reddit manipulation, none of the usual nonsense.
Three weeks later, ChatGPT was recommending it to strangers, sitting directly underneath a real brand that has spent millions building itself.
Only for a handful of very specific queries. The big category terms still belong to the incumbents, exactly as you would expect. But for a small number of long-tail questions, a made-up brand was on the shelf.
This is not a one-off
That experiment is not the only one. Ahrefs did a documented version in December, inventing a luxury paperweight company called Xarumei and putting 56 questions about it to eight different AI platforms.
Their findings are worth knowing, because they complicate the picture in a useful way.
Some models fell for it completely. Gemini, Grok, Perplexity, Copilot and Google’s AI Mode all trusted planted third-party content over the brand’s own official page. One model invented an entire Black Friday sales analysis for a company that did not exist, complete with percentages.
Others refused. ChatGPT stayed largely robust. Claude would not play at all, repeatedly stating the brand did not exist, which was technically correct.
And Ahrefs made an observation about that refusal which I think is the single most useful line in the whole study, though it is easy to skim past.
That behaviour is bad news for genuinely small, genuinely new brands.
Because a model that will not talk about a business it cannot verify does not distinguish between a business that is fake and a business that is simply new. If you started trading last year and have almost no digital footprint, some models will treat you exactly the same way they treated the paperweight company that was invented as a joke.
So here is the actual lesson
The obvious reading of these experiments is a warning: AI recommendations are not vetted, be careful what you trust.
That is true, and I will come back to it. But there is a second reading that matters more if you run a small business, and it is the one I keep landing on.
If somebody can get a fictional brand onto that shelf with eleven dollars and an hour, what is stopping you?
You have everything the fake brand had to invent.
Real jobs you have actually done. Real customers who will say so. Real photographs of real work. Real numbers about what things cost and how long they take. Real problems you have solved that nobody wrote down.
The fake brand had to conjure all of that from nothing and still got on the shelf for a handful of queries. You have it sitting in your phone and in your head.
The barrier is not credibility. The barrier is that you have not written any of it down.
Why the long tail is the way in
The other thing both experiments confirm is where the opening is.
Nobody is displacing the national brands on the obvious searches. That is not the opportunity and it never was. What is available is the enormous number of specific, narrow, oddly-phrased questions that nobody has bothered to answer properly.
Not “best plumber.” But “who fixes a leaking flat roof on a 1930s semi in Blackpool” or “what does it cost to rewire a Victorian terrace on the Fylde coast.”
Those questions have almost no competition, because writing content that specific has never seemed worth anybody’s time. That is precisely why it is worth yours.
A fake deodorant brand found a dusty corner of a shelf nobody was standing in front of. Your version of that corner exists and is probably emptier.
But do the part that makes you real
Here is where I would depart from the experiment, because the fake brand cannot do the next bit and you can.
Some models will not recommend a business they cannot verify exists. So the job is not only publishing content. It is establishing that you are a real, operating business that other people know about.
Which in practice means the unglamorous list:
A Google Business Profile, complete and active. This is the single strongest signal that you are an actual business in an actual place.
Reviews from real customers, recent ones, replied to. Nothing else corroborates you as effectively.
Consistent details everywhere. Same name, same address, same phone number across every place you appear. Contradiction is what makes a machine hesitate.
Mentions on things you do not own. Directories, suppliers, trade bodies, local press, anywhere your name legitimately appears. The fake brand’s weakness was that everything about it traced back to a single domain. That is also the weakness of most new small businesses, mine included.
That last one is the gap I am working on myself, so I am not lecturing from a comfortable position.
And the warning half, because it is real
I said I would come back to it.
These experiments are entertaining. What is happening alongside them is not.
Cloned retail sites have been turning up in ChatGPT’s shopping results. Fraudulent copies of a well-known British retailer appeared complete with discounted pricing. There are documented cases of people losing money buying from a fake shop that an AI recommended to them.
The mechanism is exactly what the fake brand experiments demonstrate, with worse intentions behind it. Nobody is checking. And an AI recommendation carries an implicit stamp of authority that a search result never did, because the machine hands your own marketing copy back in its own calm, vetted-sounding voice.
Which means two things.
For you as a buyer: do not treat an AI recommendation as a verification. If you have never heard of the business, check it exists somewhere other than its own website before you give it money.
For you as a business: this is why the open door will not stay open. Every one of those fraud stories increases the pressure on the platforms to tighten up, and they will. The businesses that walked through with genuine substance behind them will be fine. The ones that got there with eleven dollars and a domain will not.
I have watched that pattern for fifteen years in search. It is completely reliable, and the only variable is timing.
What I would actually do
Go and be the boring, real, verifiable version of that experiment.
Write the specific things. Not the category page everybody has, the narrow question nobody has answered. Put your real numbers in it.
Get your profile properly filled in and your reviews flowing, so that when a model checks whether you exist, the answer is obviously yes.
And get your name to appear somewhere that is not your own website, because a business that only exists on its own domain looks, to a machine, uncomfortably like a business that was invented three weeks ago.
The shelf is open. It is open because nobody is checking yet, which is a slightly uncomfortable reason. But a real business with real work behind it should find it considerably easier to get on than somebody making a brand up, and unlike them, you get to stay there.
Want to know whether an AI thinks your business is real? Get your free audit, or email hello@inkon.co.uk. I read every message myself.