I opened my Berlin talk with a disclaimer, so I’ll repeat it here: I am not an SEO expert. I run marketing operations at Surfer, which mostly means I own the budget, the headcount, and the deeply unpopular question of what we stop doing this quarter.
So this isn’t an SEO lecture. It’s what our data did to my resource planning.
The attention economy is quietly being outsourced
For two decades, marketing ran on one assumption: attention is scarce, so grab it and hold it long enough for someone to reach for a credit card. The hero video, the retargeting, the hook in the first three seconds — all of it exists to win that fight.
Now people ask a model to research, compare and shortlist for them, and show up only after the shortlist exists. Our job shifts from persuading humans to feeding agents accurate data across platforms, so ChatGPT or Claude can do the first round of selling for us. The new internet won’t only be searched. It will be delegated.
And here’s the part that reorganized my thinking: that shortlist is not assembled from your website alone. We’ve analyzed over 20 million LLM scrapes, and 82% of the brand mentions feeding AI answers come from third-party pages. Four out of five signals shaping your AI presence are written by people who don’t attend your brand workshops — and those mentions genuinely move recommendations. Across 289,105 URLs, we found a moderately strong correlation between how many cited sources mention your brand and how strongly the model recommends you.
Models have no independent opinion of you. They inherit opinions from whoever wrote about you the most.
My favorite example is one where I look bad
I asked ChatGPT how much Surfer costs. It told me $89 to start, plus an extra fee for AI credits.
That is a lie. A confident, beautifully formatted, well-sourced lie — because almost 60 referral pages were still describing our old pricing, and sheer volume beat our own website. Somewhere out there, a buyer got that answer, did the math, and quietly chose someone else. No dashboard will ever show me that person.
So: whose job was that? Not SEO’s, not PR’s, not product marketing’s. It lived in the gap between all three, which is the most expensive real estate in any marketing team.
Same question, different flavor, all the way down: when our influencer manager Tamara got us a more prominent placement in Marketer’s Milk — our second-biggest core source — our visibility score spiked. She did that for free, on a Tuesday. Wix, and HubSpot now evaluate influencers on LLM visibility alongside the usual criteria. LinkedIn’s marketing team runs a standing cross-functional squad against one visibility backlog.
AI visibility is a team sport. Community, social, partnerships, PPC and product marketing all generate signals that compound. But a team sport still needs a captain.
The captain is you, and the job title is strategic enablement
SEO and content people are the only ones in the building holding retrieval data. That makes your job less “produce forty articles” and more: work out which sources actually matter in your niche, brief the people who can influence them, point budget at what moves.
Finding those sources is unglamorous and effective. Run a prompt that matters to your business in the same model, day after day. Patterns appear fast — some sources show up every time, some flicker, some are hallucinated. In our testing, around the fifteenth repetition roughly a quarter of them reveal themselves as the stable core of the retrieval set. Those are the pages worth pitching, fixing, or writing for.
One warning, because this one is expensive to get wrong: track scraped answers, not API responses. We compared 1,000 prompts both ways and found only 24% overlap in brands and 4% in sources. A different door makes for a different search engine. APIs monitor a place your customers never visit.
And please stop chasing Reddit or Wikipedia because a generic ranking told you to. Reddit had a real moment inside ChatGPT and has faded; in Perplexity it spiked early and then nearly vanished; Google-backed surfaces love YouTube and Reddit for reasons that have more to do with ownership than merit. If those platforms aren’t recurring sources in your ecosystem, you’re buying prestige, not leverage.
Then write like a model is paying for every word it reads (because it does)
This is a token economy, and reading your page costs the model money. It will not go spelunking into paragraph nine when a competitor put the answer in their first sentence.
We scored content against 20 AI ranking factors, and three keep winning across every model: search intent alignment, early query confirmation, and early query answer. Match the intent, confirm you’re answering the question, then answer it — in that order, immediately. 38% of all AI citations come from the first 100 words, and pages that confirm and answer early get cited at 45% versus 23% for pages that don’t.
It’s also why our PPC landing pages and our product changelog get scraped constantly. Perfect intent match, instant answer, zero floral language. Whoever writes your changelog is doing AI visibility work and has no idea.
If you’d rather catch that before publishing than three months after, Surfer now has writing guidelines and topical map for both SEO and GEO, along with a scrape-based AI tracker — giving you everything you need to both own the existing narrative and become the source yourself. Try for free ->
No, you cannot cancel SEO
Someone in your company is about to propose cutting SEO budget “because we’re doing AI visibility now.” Two numbers for that meeting: the more queries and fanouts you rank for, the higher your odds of being pulled into AI Overviews, and 95% of cited sources rank in the top 10 for at least one keyword.
Fun twist, though: ranking matters, domain authority doesn’t. Across roughly 5 million citation sources, authority metrics correlated with AI citations at approximately zero. Your 15-year-old backlink profile is not a moat here. Which, if you’re the underdog in your category, is the best news in this article.
AI has no opinions. It has sources. The only real question left is whether your team is writing them, or letting sixty strangers with outdated pricing do it for you.
Want to know what the models are saying about you? Get a free AI visibility audit at surferseo.com, and help yourself to the rest of our research — we publish the full datasets on what gets cited, what doesn’t, and how each model behaves differently.
And if you want to argue with me about Reddit, need the model-specific factor data, or my arguments help you win a budget fight — find me on LinkedIn and send me a DM. Especially about the budget fight.
