Every marketing team I talk to now has the same question on its roadmap. How do we get ChatGPT to recommend us?
It is a fair question. It is also the wrong place to start.
Algorithms do not happen to us. We train them. Every pause on a post, every click, every quick scroll away tells a feed what to show next.
Search engines have used aggregate click behaviour as a relevance signal for years. AI systems are doing a version of the same thing, reading the trail humans leave and trying to reconstruct what those humans trust.
Which means the algorithm is always the copy. The original is human behaviour.
The signals that matter never reach a dashboard
Think about what actually happens when someone trusts your work. They forward your article to a colleague in Slack, save your research into NotebookLM, or put your chart in a board presentation.
Or they write an unprompted review because the experience was good enough to talk about.
Some of those actions leave signals machines can observe. Many never leave a WhatsApp group or a meeting room. All of them mean the same thing: a person decided this was worth passing on.
The most interesting referral traffic in any analytics account is not from Google. It is from Teams, Google Docs, internal wikis, calendars and learning platforms. Somebody had to decide your content belonged there. That is human trust, not algorithmic attention, and it is the raw material every AI system is trying to approximate.
I wrote about the difference between AI mentions and AI trust a few months ago. The same distinction applies one level up. Visibility is a result. Trust is the cause.
AI answers look for receipts
AI systems increasingly look past what a company says about itself and toward the evidence around those claims. Reviews, employee comments, press coverage, forum threads, comparison articles written by people with no reason to flatter you.
That changes the job. If an AI answer keeps repeating a criticism of your company, the instinct is to find a way to suppress it. That rarely works and it is the wrong instinct anyway.
The better move is the boring one. Check whether the criticism is still true, and if it is, fix the underlying problem.
If it is not, publish the evidence that it has changed, in places people and machines both read. Old receipts get replaced by new ones, not by clever optimisation.
I have seen this pattern at difrnt. more than once, with companies asking us to "fix" their AI answer. In most cases the answer was accurate, summarising a real complaint or a real service problem from a few years back.
The marketing fix took a week. The operational fix took a quarter, and it was the only one that moved the answer.
The receipts live in other departments
This has an organisational consequence most marketing teams have not absorbed. The evidence AI systems read is mostly produced outside marketing.
Customer support writes the experiences that turn into reviews. HR shapes what employees say on public platforms. Product decides whether the thing actually does what the homepage claims. Finance decides whether the invoice matches the quote.
If AI visibility sits only on the marketing roadmap, it will be managed as a content problem. It is closer to a reputation problem, and reputation is produced by every department that touches a customer. The companies that treat it that way will get further with half the effort.
Optimising the copy versus the original
There is a whole industry forming around influencing AI answers directly. Some of it is legitimate: clean structured data, clear pages, consistent facts across the web. Some of it is buying citations, which is the link farm with a new logo.
The distinction is simple. Legitimate work makes it easier for a machine to read what is already true about you. Manipulation tries to make the machine believe something humans do not.
Manipulation has a short half-life. Every recommendation system in history, from PageRank to the Facebook feed, eventually learned to discount signals that did not match human behaviour. There is no reason to expect AI answers to be different, and every reason to expect them to learn faster.
Meanwhile the brands that earn real recommendations compound. A product customers voluntarily recommend generates new receipts every week. A useful piece of research gets cited by journalists, who get cited by analysts, who get read by models. None of that needs a new tactic every time an algorithm changes.
Start from the human question
So reframe the roadmap item. Instead of asking how to get AI to recommend you, ask why a person would.
Why would a customer recommend you to a peer, unprompted? Why would a journalist cite your data? Why would someone save your guide instead of skimming it? Why would people keep describing you as good at the thing you claim to be good at?
If you cannot answer those questions with evidence, no amount of optimisation will hold. If you can, the machines will eventually notice, because noticing that is their entire job.
You still need to measure what AI answers say about you, because that is where you find the old receipts. That is exactly what GEOflux.ai maps: where your brand appears in AI answers, what drives it, and what to fix. But the fix itself almost always lives outside marketing.
Algorithms are intermediaries. Moody, fickle and constantly changing. The humans behind them are far more consistent. Build for them.