You don't need more traffic. You need to be the answer.

Something broke quietly in late 2024. Not a single algorithm update, not a penalty. The pipeline that carried people from a question to your website simply got shorter. A model summarised the answer. The click never happened.

Google AI Overviews now suppress click-through on top-ranked results by roughly 58%, according to data from Authoritas. ChatGPT handles approximately 2.5 billion prompts per day, and Search Engine Journal estimates around 65% of those are search-adjacent queries. People are not abandoning search. They are getting answers without ever landing anywhere. The site that ranked first still ranks first. It just stopped receiving the visit.

This is not SEO dying. It is the traffic model shifting. And the studios, consultancies, and product teams that understand the shift early will hold a structural advantage over the ones still optimising title tags.

What the model is actually doing

When ChatGPT or Gemini or Perplexity generates an answer, it is not crawling the web in real time and paraphrasing the top result. It is drawing on training data, retrieval-augmented context, and a set of signals that determine which sources it treats as authoritative.

Those signals are not keyword density. They are entity clarity, source consistency, and what you might call claim ownership. If your content defines a concept clearly, takes an explicit position, and is cited by other sources that the model already trusts, you become part of the answer. If your content exists only to rank for a phrase, it gets compressed into background noise.

This is what generative engine optimisation, GEO, actually means. Not a new set of tricks layered on top of old SEO. A different question: not "how do I rank for this keyword" but "how do I become the source a model reaches for when this topic comes up".

The gotcha here is that most content teams have not made this shift. They are still measuring impressions and positions. Both numbers can stay flat while actual model citations drop to zero.

Three things worth doing in the second half of 2026

None of these are quick fixes. Each one has a gotcha baked in.

Structure content around entity claims and source authority, not keyword volume. This means writing about things your studio or company genuinely owns: a method, a position, a definition you coined. A model needs to be able to map your content to a specific claim by a specific source. Generic category pages do not do that job. Original frameworks do.

The gotcha. this requires publishing positions you can defend, not safe content that hedges everything. Most organisations are not comfortable doing that. The discomfort is the point.

Make your brand citable. Original data, clear definitions, explicit stances. When we published Speed of Taste and the AI Tool Selection piece at Studio Hyra, we were not just writing for readers. We were placing stakes in the ground. A model trained on the web, or retrieving from it, needs something concrete to point to. "Studio Hyra believes X" is not citable. "Studio Hyra defines X as Y, because Z" is.

The gotcha. one good piece does not build a citation pattern. Models weight recency and consistency. You need a publishing cadence, not a content campaign.

Measure your AI presence. This is the part almost nobody is doing yet. Run prompts. Ask ChatGPT, Perplexity, and Gemini the questions your clients ask. See who gets cited. See how your studio or product is framed when it does appear. Then ask whether that framing matches how you want to be positioned.

The gotcha. there is no clean dashboard for this yet. You are doing manual prompt audits, tracking outputs in a spreadsheet, watching for patterns over weeks. It is unglamorous and it is the only way to actually know where you stand.

Storytelling is distribution now

Here is the part that tends to surprise people who came up through performance marketing: the studios and consultancies that will be most visible inside AI-generated answers are the ones that have been publishing strong original thinking for years.

Not because they gamed a system. Because they built a body of content with real claim density. Specific positions. Named frameworks. Data with attribution. The model does not know or care that they intended to rank. It just finds them credible.

This is why the editorial instinct and the distribution instinct have converged. A well-argued piece on why AI tool selection is a product decision, not a procurement one, does more for your AI visibility than ten pages of optimised service copy. The content has to be worth citing. If it is, the distribution follows.

At Studio Hyra, this is how we have always thought about publishing. Write something we actually believe. Make the argument specific enough that someone could disagree with it. Attribute the data. Take a position. That is not a content strategy. It is intellectual honesty. But it turns out intellectual honesty is also what makes you citable.

What to do this week

If this is new territory, start with the audit. Spend an hour with ChatGPT and Perplexity. Ask the ten questions your best clients asked you before they hired you. Write down who gets cited. Write down whether you appear. Write down how you are framed if you do.

That one hour will tell you more about your actual AI visibility than any ranking report. And it will show you exactly which gaps are worth filling first.

The traffic model has changed. The content model that responds to it is not complicated. It is just more demanding. You have to know something, say it clearly, and publish it somewhere a model can find it. That has always been the standard for good writing. Now it is also the standard for being found.