Write an AI brand content strategy from the answers
How to decide what a brand should publish and fix so that AI names it and describes it correctly. Written for content and SEO teams and for the agencies that deliver for them.
Content for AI is a strategy and not a tactic
Almost every page a brand owns was written to rank for one query. One hero page carries the argument and the facts are stated three different ways across the rest of the site. That worked when a search engine sent a person to the page. A model does not send anyone anywhere. It samples everything it can find about the brand and repeats what it finds most often and most consistently. So the brand that is named is rarely the one with the best page. It is the one whose plain facts appear the same way everywhere the model looks and at every step of the journey.
That changes what a content strategy is. It is no longer a list of keywords with a page for each. It is a decision about what the brand should be known for and a map of which pages it needs at each step and a standard every page has to meet to be read by a model. I call those the three layers and I write every plan in that order.
Start from the pages the models read
The charts on this page are from my research. The studies are on the Research page.
Before writing anything I read the pages the answers cite for the category. Most of them are not the brand's. They are comparison sites and review sites and a rival's pages and the brand's own page comes fourth or fifth if it comes at all. The chart shows that for one energy supplier at each step of the journey. The plan has to cover those pages as well as the brand's own because a page the brand does not own can still be corrected or bettered.
Whose pages the answers draw on
At Learn the answers are built from guides on sites the brand does not own. At Execute the brand's own switching page is read most. The plan starts where the brand's share is lowest and the choice is made and that is almost always Shortlist.
The three layers of the plan
The topics where the brand is in the answer today and the ones the plan adds.
Where the brand has a page the models read and where it is absent.
The checks a cited page passes and a brand's own page does not yet.
- Known for. Decide what the brand should be named for and read what the models name it for today. The gap between the two is the first layer. A supplier that wants to be known for service and is named for price has a positioning problem before it has a page problem and no amount of publishing fixes the wrong subject.
- Present at. For each topic and each step of the journey work out how much of what the models read is the brand's own. Draw it as a board. The low cells are the briefs and the order of the plan is the order of the steps where the brand is most missing.
- Ready to be read. Judge the brand's pages and the pages cited instead of them against one list. Structured facts stated once. Prices and terms given. Proof in numbers and reviews. Headings in the words of the question. A date this year. Then check that the brand's own pages agree with each other because a model that finds three versions of a price repeats one of them.
The board
One supplier across six topics and four steps. Each cell is the share of the pages the answers cite that are the supplier's own. The supplier owns Execute on every topic because its switching pages are read. It is almost absent at Learn and Shortlist and that is where the plan starts.
What the cited pages have that the brand's do not
The same six subjects judged on the same list for the supplier's pages and for the pages cited instead of them. The cited pages state prices and carry structured facts and the supplier's pages carry prose. That difference is the brief for every page on the list.
The plan I write
The plan is a list of briefs and not a deck. Each brief names the question it answers and how often the brand is named for it today and the page it has to beat and the facts it carries from the brand's own fact sheet and what the page needs to be read. The briefs are ordered by the step where the brand is most missing. The structured data and the outlines can be drafted with AI from the fact sheet and the team or its agency writes the pages. For the pages the brand does not own the brief is a correction request with the facts attached or a better page of the brand's own for the models to find instead.
Measure after every change
A plan that cannot show its result is a cost. I compare every page behind the answers with the week before and log each change and call it trivial or material and then follow the share of answers that recommend the brand after it. A move inside the range is not a result. A new model version is not the team's work. The plan is reordered each month from what moved.
Recommendation share after page changes
One supplier over eight weekly readings. Four material changes landed in weeks three and four and the share of answers that recommend it rose from a third to two in five over the month that followed. The range is drawn so that the move can be seen to clear it.
Try this in your own team
Take the Shortlist question for your category and ask four AI models with web search on and collect every page they cite. Open the three cited most and your own page on the same subject side by side and list what they state that yours does not. That takes a morning and it is the first brief of the plan.
This guide was first published on 15 October 2026.
Questions people ask me
What is an AI brand content strategy?
A plan for what a brand publishes and fixes so that AI models name it and describe it correctly. It says what the brand should be known for and which pages it needs at each step of the user's journey and what each page needs for a model to read it. It covers the pages the brand does not own as well as its own and it is measured by whether the answers moved.
How is content for AI answers different from SEO content?
SEO content is written to rank one page for one query. A model samples every page it can find and repeats what it finds most often. Content for AI states the same plain facts on every page and carries proof and fits any index and it exists at every step of the journey and not only where the traffic was.
Does a brand need to publish more to be named by AI?
Usually not more. Usually the same facts stated consistently everywhere it already publishes and a few pages at the steps where it is missing. Volume without consistency gives a model three versions of a price to choose from.
Can one good page change what ChatGPT says about a brand?
Rarely. A model does not read the best page once. It samples everything and repeats what it finds most often so one page is outweighed by the pages that disagree with it. A plan changes many pages a little and the earned pages too and that is what moves an answer.
How does a team know a content change moved the AI answer?
Compare every page behind the answers with the week before and log each change and call it trivial or material. Follow the share of answers that recommend the brand after each change with its range so that a move inside the range is never called a result.
My details are on the contact page if you disagree with any of this or want to compare notes.