A client called me last month asking why their backlink count had gone up 12% but their leads were flat as a pancake. I fired up ChatGPT and asked it to recommend the best vendor in their category – and what did I find? Their competitor was getting cited three times in a row. A competitor who has half their backlinks and a quarter of their domain authority.

That’s the new reality, my friends. The SEO inputs we used to obsess over are still important, but they’re not the inputs that decide whether you show up in an AI answer anymore. AI search runs on different signals than Google’s blue links – and most teams are still optimising for the wrong scoreboard.

So what SEO inputs actually drive AI search visibility – and which ones can you safely ignore?

The old inputs are still good, but not as good as they used to be

Traditional SEO isn’t dead, I’m not saying that. Most of what gets you ranked on Google still helps you get cited in AI answers. But the weights have changed, and a few things you used to take for granted are doing less work than you think.

Domain authority is now a filter, not a ranking lever (sigh)

AI engines use authority signals to decide which sources are even worth reading at all. Once you clear that bar, what you’ve actually published is what matters most – not some fancy domain authority score. A decent content piece from a site with a lower authority score can outperform some high-authority site with shallow content almost every time when it comes to citation counts.

Backlinks still count, but brand mentions are worth way more

In 2026, having 6.5 times more mentions from third-party sources than your own domain is the real deal. It’s not about “links pointing at me” anymore – it’s about people talking about you. A Reddit thread, a G2 review, a Forbes mention, a YouTube creator name-dropping you – those are the modern backlinks for AI search.

Exact-match keywords don’t cover half the ground

Semrush found that 65 to 85% of ChatGPT prompts don’t even have a matching keyword in their database. People ask AI different questions than they type into Google. They use full sentences. They give context. Your keyword strategy only covers a tiny sliver of how people are actually asking for what you sell.

The inputs that actually matter for AI search visibility

Here’s where I’d put my budget if I were starting from scratch.

1. Mentions on the actual sources AI engines trust

According to Profound’s citation analysis, Wikipedia makes up a whopping 7.8% of all ChatGPT citations. Reddit is 1.8%. Forbes and G2 sit at about 1.1% each. Add YouTube transcripts, LinkedIn posts, and category-specific review sites, and you’ve mapped most of where these models pull from when they synthesise answers.

The new question isn’t “how do I rank for X?” It’s “where does my brand show up across the sources AI models trust?” That changes how I think about PR budget, community engagement, and review-site strategy. Being consistent across Reddit, G2, Wikipedia (where eligible), and a handful of trusted publications does more for AI visibility than another guest post on some low-relevance blog.

2. Content that actually adds value and is easy to extract

The strongest correlation Semrush found for AI citations wasn’t traffic or backlinks. It was content depth and readability. AI engines need to extract self-contained chunks of text that directly answer specific questions. Thin, vague or buried content doesn’t get pulled.

In practice, that means direct answers to specific questions near the top of the page, clear headers that mirror real user phrasing, and structures like FAQs, lists and tables that are easy to extract. If you read your page out loud and can’t find a single sentence that cleanly answers a likely user question, you’re not getting cited.

3. Freshness – especially in fast-moving categories

ChatGPT, Perplexity and Gemini all weigh recency. Perplexity is the most aggressive about it. ChatGPT pulls heavily from pages indexed in the last 90 days when its search mode is on. Gemini sits in between.

Content that was great two years ago and hasn’t been touched since is basically invisible to a lot of AI workflows now. Refreshing cornerstone content quarterly, even if it’s just updating stats and adding a new example or two, buys you a real visibility lift for almost zero production cost.

4. Brand Search Volume: A Wake-up Call

To be honest – I was a bit surprised by this one. Brand popularity, as measured by branded search volume on Google, has a surprisingly high correlation with AI chatbot mentions – especially when it comes to ChatGPT.

Now, once you think about it, it all starts to make sense. These models are basically trained on co-occurrence and authority signals from the data they’re fed. So, brands that get a lot of searches are also going to get talked about, linked to and generally get a lot of representation in the training corpus. It’s a bit of a feedback loop, really – brand demand and AI visibility end up feeding off each other.

So whether it’s podcast appearances, conference talks, original research that’s getting shared around or PR that’s driving up branded search – it all does double duty now. It not only builds awareness in the present but also helps to reinforce your AI visibility for the next time the model is re-trained.

What to Stop Spending On

There are some inputs that used to be super high-leverage but are now pretty low-priority for AI search.

• Building generic backlinks from low-relevance sites – barely has any impact on AI citation patterns.
• Keyword stuffing or exact-match optimisation – AI engines are smart enough to interpret intent, not just strings of text.
• llms.txt files – pretty much every analysis I’ve seen says LLMs just aren’t using them in any meaningful way.
• Mass content production that doesn’t deliver any real depth – volume without substance gets totally ignored by extraction-based models.

The Shift in a Nutshell

The old SEO paradigm asked: can Google find my page? But AI search is asking a very different question: has the model been told my brand is the answer?

That’s a totally different optimisation target. The first one is all about technical hygiene, getting links and keyword targeting right. The second one is all about sustained presence across the sources that AI models trust, actually putting some real depth in the content you publish and being consistent in how your brand shows up across the web.

If your traditional SEO program is still mostly about getting links and targeting keywords, you’ll probably find that you’re quietly losing ground in AI search – even if your traditional rankings are still holding steady. The rules have changed and the work has to follow.