AI search feels random, here's what the data actually shows

Hey, it's Amit here again, with an edition of The Link Letter.

If you’re reading this from the US, there’s a decent chance you’ve just finished nursing a Super Bowl hangover right now.

Either the good kind… or the “let’s not talk about it” kind!

Either way, this is a good week to talk about something that’s stressing a lot of marketers out, and probably p**sing off a lot of consultants trying to justify their retainers.

AI search.

Rand Fishkin just published a couple of weeks ago, one of the most useful studies we’ve seen so far on how AI tools recommend brands and products.

The headline finding is uncomfortable, but important:

AI tools are wildly inconsistent when giving recommendations.

Ask ChatGPT, Claude, or Google’s AI the same question 100 times and you’ll get:

In most cases, getting back the exact same list of brands from your response, will happen less than 1% of the time.

So if you’ve ever seen your brand appear in an AI answer one week and vanish the next, you’re not imagining things.

That’s not a bug, that’s just how these systems work.

Here’s the part that really matters.

While the order of recommendations is chaotic, the brands that show up repeatedly are not.

Across dozens or hundreds of prompts, certain brands keep appearing. Others barely surface at all.

That tells us something important.

AI systems aren’t ranking brands the way Google ranks pages. They’re sampling from a consideration set.

Your goal isn’t to be “#1 in ChatGPT”.

It’s to increase the probability that your brand is even in the mix.

This is where a lot of teams go wrong.

They see volatility and respond by chasing:

But in a probabilistic system, shortcuts decay quickly.

What holds up is off-page authority.

AI answers are consistently pulled from:

In other words, the same off-page ecosystem SEO has relied on for years. The difference is that it now influences more surfaces at once.

Rand said it himself in the story - “stop throwing money at AI tracking products that don’t provide stats-backed, publicly-reviewable research”

And judging by the cost of most of them and the level of transparency behind them… I’d say you don’t need any of these prompt tracking tools. 

Ahrefs have the “Brand Radar”  feature, which is pretty decent if you have a well known brand, and you are tracking the AI visibility “trend” of your brand name. 

It’s good because they have such a large keyword database, and that’s being used to fuel their prompt inputs. 

However, I see this being more useful to a large brand that wants an overall idea of their brand growth, in AI systems. A “barometer” of AI mention growth, of sorts.

Now, aside from that - if you really want to track your AI visibility. Do it on the cheap. 

For example:

You could probably hire a dev to make the above (or vibe code it yourself in Claude Code) - and that’ll work out cheaper, considering how low cost DataforSEO’s credits are.

We already know what influences AI answers.

The hard part isn’t theory anymore. It’s execution.

Increasing your chances of showing up means:

None of this is difficult in isolation.

Doing it well, repeatedly, without cutting corners is where most teams struggle.

That’s what GEO and AEO really are. Not new tactics. Just a higher bar.

This is why I am writing this newsletter.

And why everything we publish at TLG focuses on off-page work that compounds over time.

Not hype. Not dashboards pretending AI is predictable.

Just building the kind of authority that keeps your brand in the conversation, even when the answers change.

Are you currently tracking AI “rankings”, or just whether your brand shows up at all?

Hit reply. I read every response.

 Amit Raj - The Links Guy

P.S. AI answers may be random. Brands that earn real authority usually aren’t.