AI-first content strategy

What does an AI-first content strategy look like on a real site?

We are running one on attensira.com right now, so here is the honest starting point.

Our workspace reads 61 buyer questions on ChatGPT and Perplexity every day. One of them is "alternative to hiring a GEO agency", which is close to a description of what we sell. In the 30 days to 4 October 2026 it got 63 answers, read over the Attensira MCP, and none of them named Attensira. The ChatGPT answer at the top of this page shows what the engine recommends instead: build the capability in-house, and publish "sources AI systems can confidently reference", from comparison pages to original data and glossaries. Its three inline links went to Google's documentation and two agency roundups.

That is the gap an AI-first content strategy is meant to close, so the rebuild we are doing is the strategy, applied to ourselves:

The measurement that tells us whether it worked is the same question, read again: whether the next answers name us, per engine, with the sample size.

What is an AI-first content strategy?

It is a planning discipline, not a writing trick. It assumes the highest-volume reader of your site is a system that arrives for one fact, extracts a passage and leaves, and it plans for that:

• which pages exist, so each buyer question has exactly one page that owns it;
• where each fact lives, so a price or a policy is stated once and linked everywhere else;
• who may fetch what, decided per crawler rather than in one blanket rule;
• how pages stay true, because a stale copy stays retrievable long after it stops being right.

The reader of your prose is a selection mechanism. The reader of the resulting sentence is a person who may never see your page.

Why does access belong in the content plan?

Because the most consequential AI-first decision is often made in a config file by someone who was not in the content meeting. OpenAI documents separate agents with separate jobs:

• GPTBot

  Documented purpose

  content that may be used to train foundation models

  If you disallow it

  signals your content should not be used for training

• OAI-SearchBot

  Documented purpose

  surfacing websites in ChatGPT's search features

  If you disallow it

  not shown in ChatGPT search answers, though navigational links can still appear

• ChatGPT-User

  Documented purpose

  fetches a person asked for

  If you disallow it

  robots.txt rules may not apply

A blanket "block AI bots" rule answers all three the same way and usually gives up search to refuse training. Write each position deliberately in robots.txt.

What changes when you plan for extraction?

The unit of publication is the answerable section. Each section answers one question completely, under a heading phrased as that question, with the qualifier in the same sentence as the claim. A section that needs the one above it is not portable. See AI response optimization.

Depth beats multiplication. Retrieval picks passages by relevance to a question, not by page count. Two hundred templated near-duplicates compete with each other and multiply what you must keep true.

Being quoted is the outcome. In a zero-click answer the visit does not happen. Plan sections whose being quoted, with your name in the sentence, is worth having on its own.

Markup mirrors the page. Schema markup describes what the page visibly says. Google's guidance is explicit that structured data about information not visible to the user should not be added, even if accurate.

What are the objections to an AI-first content strategy?

"This is writing for robots." It is writing for people in units a machine can lift without breaking them. The test is whether a section still reads correctly when it is the only thing someone sees.

"Our content team cannot own robots.txt." Then the plan must name who does, because one rule can remove you from an engine's answers without an error anywhere.

"Deleting pages loses traffic." Overlapping pages split the same question between two URLs and keep stale copies alive. On a low-traffic site, one clear owner per question is worth more than two half-owners.

"We will see results next week." Engines re-read on their own schedules. Measure before and after, per engine, with the sample size, and expect weeks.

How does Attensira run this for you?

Attensira's crew runs this plan for a workspace the way we are running it on our own site: it maps the questions your buyers ask AI to the pages that should own them, writes and fixes those pages through content optimization, ships each change after your approval, and reads the answers every day to see whether the change moved them. That is the service of an SEO agency at the price of a tool, and pricing shows how the work is paid for in credits as it runs.