Adapted from Duane Forrester's essay on Duane Forrester Decodes. Aurora extends his framing for marketing teams running AI-search visibility programs.
The most common argument against investing in AI visibility goes roughly like this: AI referrals are still around 1% of publisher traffic, so optimizing for AI is optimizing for a rounding error. The argument is internally consistent and structurally wrong. It compares the ROI of one routing mechanism (search) to the visibility of a system that was never designed to route at all.
Search engines were built to crawl, index, rank, present options as a list, and let the user click. The design preserved liability by keeping the user as the active agent choosing the source. LLMs were built to produce the answer directly. Citations are grounding artifacts produced by a retrieval pipeline, not routing mechanisms. Asking how much traffic AI sent you is the wrong instrument applied to the wrong thing. The right question is how often you were the source the answer was built from.
The liability surface moved
The architecture that protected search engines from being responsible for what users clicked on does not protect LLMs the same way. When a user clicked a search result and got hurt, the engine could point to its ranked list and to user agency. When an LLM produces an answer in its own voice, the liability surface that the SERP was designed to offload sits with the model producing the output.
Cases are accumulating. Walters v. OpenAI was dismissed on summary judgment in May 2025, protected by disclaimers. Air Canada was held liable for its branded support agent because a customer could reasonably rely on an airline's own assistant. The New York Times v. OpenAI copyright case was allowed to proceed in March 2025. Anthropic settled with book authors in August 2025 for amounts reported in the billions. The legal vocabulary the courts are converging on is reasonable reliance — and the more specialized and authoritative a chatbot appears, the harder the disclaimer defense becomes. This matters for visibility because as the liability surface migrates, the platforms have a structural incentive to ground their answers in citable, trustworthy, attributable sources. Being one of those sources is no longer a brand-positioning soft outcome. It is the path of least legal resistance for the platforms themselves.
The denominator problem
The "AI is still only 1%" argument measures relative share of a shrinking pie. Organic news publisher traffic fell from 2.3 billion monthly visits in mid-2024 to under 1.7 billion by May 2025 — more than 600 million visits lost in a single year. Business Insider's search traffic dropped 55% between April 2022 and April 2025. HuffPost lost roughly half its search referrals. The New York Times saw search's share of its traffic decline from 44% to 37%. Zero-click searches climbed from 56% to 69% between May 2024 and May 2025. A Reuters Institute survey of 280 media leaders in late 2025 expects another 43% average decline over three years.
A stable percentage share of a shrinking pie is not stable. It is a loss. And the percentage is not even stable — it is shifting toward AI-mediated answers.
What the capex curves are telling you
The five largest US cloud/AI providers committed between $660B and $690B in 2026 capital expenditure, nearly doubling 2025 levels. Alphabet alone is guiding to $175–$185B (more than double 2025's $91B). Microsoft, Amazon, Meta, and Oracle are on similar curves. Bank of America estimates AI capex will reach 94% of operating cash flows in 2025–2026, up from 76% in 2024. That is not the shape of a defensive hedge. That is companies betting that the channel is the future even before the spreadsheet knows how to score it.
ChatGPT alone is around 900 million weekly active users, up from roughly 200 million eighteen months earlier. The full LLM category is past a billion weekly users. Behavior is shifting fast enough that adoption is settled. The measurement is what is unsettled.
A different measurement framework
AI visibility cannot be evaluated on a clicks-per-dollar basis. The unit of value is not the click. The unit is being the cited, grounded, trusted source inside the answer. Aurora is built around measurements that match the medium:
- Citation share per prompt, per engine. Of all the prompts in your campaign, in what fraction is your domain cited? Tracked over time, this is the only honest "did our content earn its place inside the answer" metric.
- Brand visibility (mentions and paraphrases). Even unlinked. The model is using your content; you want to see it whether or not it links you.
- Competitor exposure. When you do not appear, who does? The substitution pattern is half the diagnosis.
- Source intelligence per category. Which domains do the engines reach for in your category? That is the playing field. If you are absent from the playing field, you are losing for the right reason and can act on it.
- Engine drift. Per-engine deltas surface the moments when model upgrades changed the answer geometry. You want to see those weekly, not quarterly.
The practitioners who figure this out will not be the ones who finally found an ROI case that convinced the CFO. They will be the ones who looked at the capex curves, the behavioral curves, and the liability curves, and concluded that the channel is the future regardless of whether the spreadsheet knows how to score it yet.
The old ROI math is asking the wrong question. The new question is how often the answer is built from you.
