AI-first content strategy
Planning a content estate on the assumption that its main reader is a machine that extracts one fact and leaves — what that changes about structure, access policy and maintenance.
Karl-Gustav Kallasmaa, Founder & CEOLast updated An AI-first content strategy plans a content estate on the assumption that its highest-volume reader is a machine that arrives for one fact, extracts a passage, and leaves. It is a planning discipline rather than a writing technique: it decides what gets published, where each fact lives, who is allowed to fetch it, and how it is kept current.
The assumption it rests on
That most reads are partial and most readers are not people. A retrieval system fetches a page, splits it, and passes a fragment into a model's context window alongside fragments from other sites. The reader of your prose is a selection mechanism; the reader of the resulting sentence is a human who will never see your page.
This inverts the usual funnel logic. A traditional estate is designed so a visitor lands, is oriented, and is moved toward an action. An AI-first estate is designed so that any single section can be removed from its context and still be correct, attributable and useful — because that is the normal case, not the edge case.
It is not a licence to write for machines. Google's structured data guidance states the constraint directly: "Don't create blank or empty pages just to hold structured data, and don't add structured data about information that is not visible to the user, even if the information is accurate." Machine-readability is a formatting of what the human page says, never a substitute for it.
Access is a strategy decision, not an ops chore
The most consequential AI-first decision is usually made in a config file by someone who was not in the content meeting.
OpenAI documents three separate agents with three separate purposes:
Those are three different policy questions wearing one label. A blanket "block AI bots" rule answers all of them the same way and typically costs the search surface in order to refuse the training surface. An AI-first strategy states each position deliberately in robots.txt, and knows which of the three each rule affects.
What changes in the plan
The unit of publication becomes the answerable section. Structure a page as a set of sections that each answer one question completely, with the qualifier in the same sentence as the claim. A section that requires the section above it is not portable.
One fact gets one home. Every fact — a price, a limit, a supported version — should have exactly one page that states its current value, with other pages linking rather than restating. Restatement is how a superseded number stays retrievable for years. Retrieval will happily return the stale copy; nothing marks it as outdated.
Freshness becomes an obligation, not a signal. Every duplicate of a fact is a future contradiction. Maintenance load is therefore a function of how many places a fact is written, which makes consolidation a strategic act rather than tidying.
Depth beats multiplication. Retrieval picks chunks by relevance to a query, not by how many pages a domain has. Two hundred near-identical templated pages compete with one another for the same retrieval slot and multiply the surface that must be kept true.
Attribution becomes the goal of the page. In a zero-click answer the visit does not happen. If the section is only valuable as bait for a click, it has no value in the read that actually occurred. Plan sections whose being quoted with a source link is itself the outcome.
Failure modes
- Blanket bot blocking. One rule that refuses training also refuses search, and the loss appears months later as absence with no error anywhere.
- The scattered fact. The same limit stated in a blog post, a changelog and a docs page; two are wrong after the next release, all three remain fetchable.
- Markup without matter. Schema asserting things the page does not visibly say — explicitly against Google's stated guidance, and useless to a retrieval system that reads the body text.
- Gated depth. The substantive material behind a form. An agent cannot fill in the form, so the estate's best content is the part no model can see.
- Volume as strategy. Publishing more thin pages to raise the odds of citation, which raises maintenance cost and dilutes the pages that could have been authoritative.
- Preamble-heavy writing. Sections that open with context and reach the claim in paragraph three lose to sections whose first sentence is the answer.
What it does not mean
It does not mean a separate machine-facing site, a new file format, or keyword-shaped prose. Keyword stuffing is actively counterproductive in this setting: it degrades the clarity that makes a passage extractable. The whole discipline reduces to writing fewer, truer, more self-contained sections, and being deliberate about who is allowed to read them.
Frequently asked questions
Is this writing for machines?
No. It is writing for humans in units a machine can lift without breaking.
What is the single highest-leverage decision?
Crawler access policy, stated per agent rather than in one blanket rule.
Why does duplicating a fact hurt?
Because every copy is a future contradiction that stays retrievable after it stops being true.
Does publishing more pages help?
Not by itself. Retrieval selects on relevance, and near-duplicates compete with each other.
Terms related to AI-first content strategy
The practice of getting a source reproduced inside an AI-generated answer, introduced as a named paradigm in a 2023 research paper.
The standardised file that tells crawlers what they may fetch, now the main place where AI training and AI retrieval access are decided separately.
The Schema.org vocabulary and the JSON-LD you embed with it, what it is genuinely good for, and the limits of what it can make happen.
The two separate stages that decide whether a page can be retrieved at all, and the reason a serving rule on a blocked page is never read.