Stop chasing Reddit; five lessons for getting GEO right

For a while there, Reddit looked like it might be the cheat code for GEO… AI likes human conversation, Reddit contains an extraordinary amount of human conversation, so get your brand discussed there and, hey presto, ChatGPT might start recommending you.

Then something happened. In August, Reddit’s share of tracked ChatGPT Search citations fell from an average of 3.83% to 0.52% in a matter of days… an 86.4% collapse. At roughly the same time, Promptwatch spotted what appeared to be a significant change in the way ChatGPT searched the web, including a huge increase in searches targeted at specific domains.

Nobody outside OpenAI can definitively say why Reddit citations fell, and correlation isn’t causation… but that’s actually part of the lesson here. Because if your entire GEO strategy can be broken by a change you don’t understand, can’t predict and have absolutely no control over, you probably need a better GEO strategy.

So here’s what marketers should actually take from it.

Don’t build a Reddit GEO strategy; build a GEO strategy in which Reddit has a role
Reddit matters because it’s full of enthusiasts, experts, obsessives, customers and people with extraordinarily strong opinions about everything from running shoes to air fryers, but it is still only one source.

And while Reddit’s ChatGPT citations fell dramatically, the decline across Google’s AI surfaces was considerably smaller. Different machines already look in different places, those places will keep changing, and trying to optimise for whichever corner of the internet the machines happen to favour this month feels like a game we’re unlikely to win.

So build authority across your own content, journalism, reviews, experts, creators, communities and Reddit… because you don’t know where the machine will look next.

Don’t confuse citation with influence
A Reddit link appearing underneath an AI answer is wonderfully easy to count, which means we’ll inevitably become terribly excited about counting it, but that’s not really the outcome we’re after.

The useful question is whether, when somebody asks an AI a commercially important question about your category, your brand is mentioned, understood and recommended… and whether what the machine says about you is actually what you want it to say.

Reddit can be part of the evidence shaping that answer without Reddit.com appearing neatly underneath it.

So measure the answer, not just the footnotes.

Use Reddit to find the questions… not manufacture the answers
This is where Reddit becomes really useful because people ask questions there they would never put into your customer survey… “Is X actually worth the money?”, “What does nobody tell you before buying Y?”, “What’s a cheaper alternative to Z?”

That’s marketing gold because it tells you how people really talk about your category, what annoys them, what they compare and what they actually care about, which in turn tells you the kinds of questions they’re likely to ask AI.

So listen to Reddit, learn from it and use those insights to make sure useful answers exist wherever a machine might go looking for them.

Make your own website worth finding
Promptwatch also found that ChatGPT’s use of searches targeted at specific domains jumped dramatically at roughly the same time as the Reddit change. Again, we shouldn’t pretend we know precisely why. But it does raise an important question for marketers.

If a machine comes directly to your website looking for an answer, can it actually find one?

Not “Welcome to Brand X, where innovation meets passion”… I’m not making this up.

That means clear product information, comparisons, pricing, FAQs, specifications, evidence and case studies are all the boring useful stuff we’ve spent years hiding three clicks below the lovely brand film.

Start thinking about reputation, not just optimisation
This is where I think GEO gets really interesting, because it’s increasingly starting to look less like a technical optimisation discipline and more like reputation management.

AI isn’t only looking at what you say about yourself, it’s looking at what everybody else says too. Your website might say you’re brilliant value while customers say you’re expensive; you might describe yourself as innovative while journalists consistently describe three competitors as the innovators in your category.

That’s not really a keyword problem… It’s an evidence problem.

Which means the marketer’s job increasingly becomes understanding the story the wider internet tells about the brand, then making sure there’s enough genuine, independent evidence to support the associations you want humans and machines to make.

Reddit absolutely has a role in that; listen to it, learn from it, participate openly when you’ve genuinely got something useful to add and, most importantly, give customers experiences worth talking about in the first place.

Just don’t build your GEO strategy around persuading one website to tell one machine you’re brilliant… because tomorrow, the machine might decide to look somewhere else.

Jon Williams is CEO and founder of The Liberty Guild