GEO and SEO get filed under one heading, handed to one team and paid for out of one retainer, and that is why this work stalls in month two. Half of what sits under the AI visibility label is search engine optimisation, and your search team should carry on doing it. The other half decides the sentence a model writes about your category. It happens on pages you will never own, and it is bought the way media is bought.
That split decides who is accountable, which budget the money leaves and what you can honestly promise before anything runs. Brief it as one job and you hand a search supplier a target no search supplier can reach.
The overlap is real and it is worth saying out loud. Generative engine optimisation one quarter, answer engine optimisation the next: whatever the label, a real slice of this is ordinary work on your own site. Pages a crawler can read without running your JavaScript. Product facts stated plainly enough that a model can repeat them without guessing. A page for each thing you sell rather than one page covering everything. If your search team is any good they are doing most of it already, and they do not need a new discipline to keep doing it.
Search engine optimisation is not dead, and arguing about the name is a distraction from the harder question. The harder one is where the answer comes from when a buyer never types your name.
90% of the sources AI cites when answering are ones the brand does not control; only 10.15% of citations go to brand-owned domains. That is Foundation Marketing and AirOps, 5 May 2026, reading 57.2 million citations across 5.1 million AI responses. The sample's market is not stated, which is worth knowing before you repeat the figure in a meeting in Sydney or London.
Your own pages are a small share of what an answer gets assembled from. They count for most when a buyer already knows your name and asks about you directly. They count for least at the point that decides the sale, which is the question with no brand in it at all: someone describing a problem and asking what to do about it.
The site half is not a consolation prize. When someone asks a model about your brand by name, your own pages are often the source it reaches for, and anything wrong on them comes back with the model's confidence behind it. A retired product name, a support policy from two years ago: that is a site problem with a site fix, and the search team already owns it.
What no site change can reach is the other question. Nothing you publish on your own domain puts you inside an answer built from a magazine review and a comparison piece someone else wrote. The instinct at that point is to reach for authority, on the theory that a stronger domain gets cited more. That theory has now been measured.
~0 correlation between a domain's authority score and whether AI cites it (Spearman -0.073 to +0.026). Offline reputation no longer buys a seat in the answer. Surfer, 14 July 2026, across roughly 5 million unique citation source URLs from 20,000 prompts. That sample's market is not stated either.
Put the two readings side by side and the division of labour writes itself. Most of what the answer is built from sits on domains you do not own, and the lever a search retainer pulls hardest is not the one that decides which of those domains gets picked.
Inside a company the brief usually sits with the brand or media team. Inside an agency it is the planner who already buys partnerships, not the search lead. The work is commissioning: choose the topic buyers ask about, choose the titles a model reads on that topic, brief them, and let their editors write the piece. Branded content has worked like this for years and the mechanics do not change because part of the readership is now a machine. What changes is what you brief it against.
Here is the sentence a search supplier cannot say. A search retainer can recommend that you earn coverage on independent titles. It cannot commission one or hold an editor to a publishing window, because it holds no commercial relationship with the publisher and buying one is not what a search retainer is for. Recommending the work and being able to place it are two different businesses, and telling them apart is most of the decision about who to give this to.
It also has to stay genuinely editorial. The publisher's writers and editors decide what the piece says. We set the claim, the facts that have to appear and the standard we expect; they write it. That independence is why a model reads the page as evidence rather than as your marketing, and it is the same reason trust has always sat with publishers rather than with owned media.
Forcing it into the search line is how the second half quietly disappears. It is bought the way every other publisher partnership is bought: a topic, a set of unaffiliated titles, one window. That is a media line, on the media plan, with a start date and a finish date against it. Asking a search retainer to absorb it is asking a supplier to fund inventory it does not sell.
For a planner this is the easier half of the conversation, because a media line arrives with a reporting habit already attached. You brief it, you compare what comes back, you run it. Sending one brief to several publishers and comparing the responses is a workflow your team runs every week.
The trap in the funding conversation is the promise. Nobody controls the model, so nobody can sell you a mention, and a supplier who offers one is selling something they cannot know. What you can commit to before the money is signed off is a measured starting position, the questions you agree to be judged on, and the same read run again at the end. That is the discipline of deciding what success means before the campaign starts, one channel earlier.
The site work is continuous and belongs where it already lives. The publisher work is a campaign: one topic, several unaffiliated titles inside the same window, then the same read run again to see whether the answer moved on the questions you picked at the start. One of those jobs never finishes and the other has an end date, which is a second reason they do not belong on one line of the plan.
So the practical version, for the planner who has to put a number against a row this quarter. Leave the search line where it is and doing what it does well. Open a second line for the publisher half instead of asking the first one to stretch over ground it cannot reach. If you are not yet sure that second line is worth opening, find out where you stand first: our free AI Visibility Report shows which questions your buyers are asking, whether you get named on them, and which sources the answer is being written from. That source list is the closest thing there is to a brief for the half of this job that is not a search job.
10 September 2026 · 6 min read
Before any engine can quote your title, the page has to clear conditions that have nothing to do with the writing. The six that decide it, the two halves of a quotable page, and the crawler distinction most robots.txt files have never made.
10 September 2026 · 6 min read
OpenAI's own documentation says a sponsored placement sits below the answer and cannot change it. That makes the ad and the recommendation two different purchases, out of two different budgets.
9 September 2026 · 6 min read
The largest kind of source in an AI answer is often one nobody can buy. Here is the subtraction that shows what is actually left to commission in your category.
See how your brand appears in AI answers, and where trusted publisher content can strengthen its visibility.
