Bias used to be easier to spot. Cola enthusiasts of a certain age will remember the Pepsi Challenge. Representatives from PepsiCo would visit shopping malls across America with a table and two plain white cups. In one cup – Pepsi. In the other – Coca Cola.
Shoppers would then be encouraged to taste both and reveal which soft drink they preferred. Americans, as it turned out, preferred Pepsi.
This test – the “sip test” – was flawed. When you give someone a really small amount of two beverages, they normally prefer the sweeter one. The same is true for wines.
Most people didn’t realise this at the time – but still, they took the adverts with a grain of salt. After all, this was a Pepsi advert saying Pepsi is good. And consumers aren’t stupid.
If you wanted to do the Pepsi Challenge today – you wouldn’t put the results in an advert. You’d plaster the results anywhere you could – across owned content, Reddit, Youtube and LinkedIn. Then, when a consumer asked ChatGPT about which soda to drink – they’d be told: “well, according to blind taste tests, it’s Pepsi you want.”
Bias used to come labelled
Since the dawn of advertising, brands have been writing their own references.
This continued into the internet era. “Brand A vs Brand B” is an age old SEO tactic. So is writing a listicle and putting yourself on top.
As long as you’re not publishing false claims, there’s nothing wrong with this in my opinion. Like the Pepsi adverts – when consumers and buyers see something in a certain context, they apply a healthy dose of cynicism.
Chatbots and zero-click search have quietly stripped that context away. The self-serving content is still out there – it’s just no longer arriving with a name attached. It gets hoovered up, synthesised, and read back to a buyer in a flat, neutral, authoritative voice.
The grain of salt goes missing. A buyer asks ChatGPT “any reason not to go with this provider?”, gets an answer that sounds impartial, and never learns that a chunk of it was lifted from a rival’s marketing page.
We wanted to know how often that actually happens, so we ran the numbers across three sectors. We’ve called the project Gullible Engine Optimisation.
What we found
We recently conducted some research to look at two things related to ChatGPT and co. Firstly, how easy is it to trick ChatGPT and co into saying something that isn’t true? Answer: very easy. It took just one blog post to convince leading chatbots that my business partner is the best dressed man in PR.
But that’s a bit silly. So we wanted to look at more serious stuff – i.e., whether brands were doing this, too. It turns out, they are. In one test, we found that 70% of sources driving answers to buyer-style prompts were simply brands recommending themselves with listicles. In simple terms that means publishing a top ten list and putting yourself at number one.
Elsewhere, we found other examples of this. Brands publishing hit pieces on competitors, and comparisons pages where – shock horror – the brand publishing the comparison came out top.
In the olden days of the internet – where customers actually went beyond the search results page – there was nothing wrong with this. If Pepsi.com says Pepsi tastes better than Coca-Cola – well, so what?
The problem is this content is now driving answers on a supposedly neutral website.
Are my competitors shaping my brand?
Based on this research, and data I’ve seen elsewhere in countless audits I’ve run for brands, there’s a very high probability that when a prospective customer asks ChatGPT about your brand, one of your rivals is influencing the result.
This influence might be fairly light, or it might be substantial. I’ve seen many examples of both.
This is, of course, a problem. According to Forrester, 94% of business buyers now use AI in their purchasing process. That number is only moving in one direction, and so is the weight buyers place on what these tools tell them.
Here’s the good news: this is fixable. Not easily, and not by waiting for the engines to grow a conscience, but by going back to the fundamentals of marketing and communications.
Start with the data. Most brands have no idea what the engines currently say about them, or which competitor is filling the silence where their own answer should be. Find that out, then close the gaps: put the truth about yourself on the record, in the places and formats these engines actually trust, before someone with a stake in the answer does it for you.
Maybe the engines will learn to doubt in time. But that’s not something to wait for. For now they’ll keep hoovering stuff up and repeating it to your prospects and customers. If you want to take control of your algorithmic reputation – listen, learn and correct, correct, correct.
