The SEO Playbook Doesn't Work Here. What Does?

If you've spent years building an SEO practice, you have good instincts. You understand how to earn visibility in a system where the rules, while complicated, are at least knowable. You can check your rankings. You can measure your traffic. You can trace a line from effort to outcome.

None of that prepares you for what's happening now.

AI-assisted discovery runs on different mechanics, rewards different signals, and produces outcomes you cannot see unless you go looking for them. The skills transfer. The playbook doesn't.

This post is for the people who already understand search deeply and want an honest accounting of what changes when the answers come from a language model instead of an index.

---

The first thing to understand: there are no rankings

In traditional search, you occupy a position. You're #3 for a query, or #11, or #47. That position is stable enough to track, optimize against, and report on. The entire SEO industry is built on this stability.

AI recommendations work nothing like this.

SparkToro ran 2,961 prompts across ChatGPT, Claude, and Google AI with 600 volunteers over two months. They wanted to know how often the same prompt produces the same list of brand recommendations. The answer: less than 1% of the time. The probability of getting the same list in the same order drops below 0.1%. Ask ChatGPT to recommend a product a thousand times and you will almost never see an identical response twice.

This doesn't mean AI visibility can't be measured. It means position is the wrong unit. What matters is frequency: how often does your brand appear across many prompts and many runs? Think of it as share of voice across a probabilistic system rather than a rank in a deterministic one.

Any vendor selling you "AI rankings" is selling you noise. Visibility percentage across large sample sizes is the metric that holds up. Individual position does not.

---

Backlinks don't carry the weight you think

If there's one thing SEO practitioners believe in their bones, it's the power of backlinks. And for Google's traditional algorithm, that belief is well-founded. Links remain a core ranking signal.

For LLMs, the picture is different. Ahrefs found a weak correlation between backlink profiles and ChatGPT citations. The strongest predictor of whether a brand gets mentioned by AI models turns out to be brand search volume, with a 0.334 correlation that outweighs traditional link metrics.

That's a meaningful shift. It suggests that what matters to LLMs is not how many other sites point to you, but how many people already know about you and search for you by name. Brand recognition, not link equity, is the closer analog to domain authority in the AI context.

This doesn't mean links are worthless. Sites with strong backlink profiles do get cited more often by Google's AI Overviews, which still lean heavily on traditional ranking signals. But outside of Google's own AI features, the relationship weakens considerably. ChatGPT Search primarily cites pages ranked position 21 and below about 90% of the time. The models are reading different parts of the web than the first page of Google.

---

What LLMs actually look for

If backlinks aren't the primary signal, what is? The research points to a different set of factors.

Content depth and readability. When it comes to securing AI citations, content depth and readability matter most, while traditional SEO metrics like traffic and backlinks have limited impact. LLMs favor comprehensive, clearly written content that covers a topic thoroughly. They're evaluating the quality of what you've written, not just the signals surrounding it. Extractable claims. 44% of all LLM citations come from the first 30% of a page's content. The models are looking for clear, direct assertions they can pull into a response. Vague copy doesn't get cited. Content that leads with a definitive answer and structures its claims in short, parseable sections does. Princeton's GEO research found that adding statistics can increase AI visibility by 22%, and including quotations from credible sources can boost it by 37%. Freshness. AI crawlers disproportionately access recent content. 65% of AI bot traffic targets pages published or updated within the last year. There's emerging evidence of a three-month citation cliff, where content older than 90 days begins losing AI visibility regardless of its quality. The implication is that publishing and forgetting is a worse strategy for AI than it ever was for Google. Entity authority across platforms. LLMs don't just read your website. They synthesize information about your brand from everywhere: review platforms, social media, forums, news coverage. Domains with profiles on Trustpilot, G2, Capterra, and similar platforms have 3x higher chances of being cited by ChatGPT than those without. Domains with significant presence on Reddit and Quora have roughly 4x higher citation rates. Your AI visibility is the sum of everything the web says about you, not just what you publish yourself. Cross-platform presence. Only 11% of domains are cited by both ChatGPT and Perplexity. Google's AI Overviews and AI Mode cite different sources 87% of the time for the same query. Each platform has its own retrieval preferences, which means optimizing for one doesn't guarantee visibility in another. A brand that dominates in ChatGPT responses may be invisible in Perplexity, and vice versa.

---

The distribution game has changed

One of the most underappreciated findings in recent AI visibility research: distributing content across a wide range of publications can increase AI citations by up to 325% compared to publishing only on your own site.

This makes intuitive sense when you understand how LLMs work. The models build confidence in a claim or recommendation by seeing it corroborated across multiple independent sources. If only your website says you're the best at something, the model treats that as self-promotion. If your website says it, and an industry publication says it, and a review site says it, and a forum thread says it, the model starts to believe it.

For SEO practitioners, this shifts the emphasis from earning links (where the value is the link itself) to earning mentions (where the value is the claim being made about your brand in a credible context). Guest posts, analyst coverage, earned media, and review platform presence all contribute to what might be called "corroborated authority," the thing that makes a model confident enough to recommend you.

---

What doesn't work

Some popular SEO tactics translate poorly or not at all.

Keyword optimization in the traditional sense. LLMs don't match keywords. They interpret intent and context. Stuffing a page with exact-match phrases doesn't improve AI citation rates and may hurt readability, which does matter. Thin content at scale. Publishing hundreds of pages with surface-level information to cover keyword territory is actively counterproductive for AI visibility. The models evaluate content quality when choosing what to cite. One comprehensive, expert-led article outperforms dozens of generic ones. Faceless content. Named authors with real credentials significantly outperform anonymous content. LLMs look for what the industry calls E-E-A-T signals: experience, expertise, authoritativeness, and trustworthiness. If a model can't find a real person with verifiable knowledge behind a piece of content, it's less likely to cite it. Optimizing for one platform. Given the minimal overlap between what different AI platforms cite, a strategy focused solely on ChatGPT or solely on Google's AI features will leave gaps. The brands showing up consistently are the ones visible across the whole ecosystem.

---

What to actually do

Here's where this gets practical.

Audit your current AI visibility before optimizing anything. You need a baseline. Query the major models with the questions your customers actually ask and document what comes back. This is the equivalent of running your first rank tracker report in 2005: you can't improve what you haven't measured. (This is what Renown is built for, but even doing it manually is better than guessing.) Restructure content for extractability. Lead every page with a direct, definitive answer. Front-load the specific claims you want models to pick up. Structure with clear headings, short paragraphs, and factual assertions that stand alone when pulled out of context. Make the AI's job easy. Invest in original data and research. Content that includes first-party statistics, original surveys, proprietary benchmarks, or unique case studies is disproportionately cited by AI models. If you can generate data that nobody else has, you become a primary source that models return to. Build your brand's presence across platforms. Ensure your product or service has accurate, detailed profiles on relevant review sites, directories, and community platforms. Respond to reviews. Participate in relevant forums. This isn't about link building. It's about creating the kind of distributed, corroborated presence that gives models confidence in recommending you. Refresh content on a regular cycle. The freshness signal is real and the cliff is steep. Content that was performing well in AI citations three months ago may have already begun fading. Build a refresh cadence into your editorial calendar and treat it as maintenance, not a nice-to-have. Monitor across all major models, continuously. A monthly spot-check won't cut it. AI recommendations are volatile by nature, and changes in model training, web indexing, and competitor activity can shift your visibility quickly. The insight is in the trend, not the snapshot.

---

A new discipline, not a new trick

The temptation is to treat AI visibility as an extension of SEO, something you can bolt onto your existing program with a few tweaks. That underestimates what's changing.

SEO was built for a system that returns a list of options and lets the user choose. AI discovery returns a verdict. The model has already chosen for the user, or at least narrowed the field to a degree that makes the remaining choice feel effortless. Being excluded from that narrowed field is a different kind of loss than ranking on page two.

The skills that make a good SEO practitioner, understanding how systems evaluate content, thinking structurally about information architecture, measuring what matters, are exactly the skills this new discipline demands. The mechanics are different. The rigor is the same.

The people who figure this out first won't just do well. They'll set the terms for everyone who follows.

---

Renown is an AI visibility platform that tracks how AI models talk about your brand.