AI readiness is not SEO, and treating it that way costs you

Most teams arriving at AI visibility bring an SEO mental model with them: find the ranking factors, optimise for them, watch position improve. It is a reasonable instinct and it is partly right. It is also the source of most wasted effort in this area, because the two systems reward different things and fail in different ways.

Search answers 'which page', retrieval answers 'what is true'

A search engine returns a ranked list of documents and lets the reader choose. Its job ends at the link. A system generating an answer has a harder problem: it must extract specific claims, decide whether they are trustworthy enough to repeat, and often attribute them. The unit of value shifts from the page to the claim.

That shift explains most of the divergence. A page can rank well while being nearly useless to an answer engine — long, hedged, structurally vague, with its actual claims buried in prose that cannot be lifted cleanly. Equally, a modest page that states something specific, attributes it, and dates it can be highly quotable while ranking nowhere in particular.

What carries over from SEO

The technical foundation transfers almost entirely. If a crawler cannot reach and parse your page, nothing downstream matters:

• Crawlability and clear robots directives
• Server-rendered content rather than content assembled only after JavaScript runs
• Clean status codes, canonical URLs, and a maintained sitemap
• Descriptive titles and headings that state what a page actually covers
• Structured data, particularly for organisation and article identity

If you have done this work already, you are further ahead than most. The technical layer is table stakes in both worlds.

What does not carry over

Three habits from search optimisation actively work against you here.

Volume as a strategy

Publishing many thin pages to cover keyword variations is a search tactic. For a system deciding whether to repeat a claim, thin pages are worse than no pages: they dilute the signal that your site knows something specific, and near-duplicate content across a site is itself a quality marker in the wrong direction.

Keyword-shaped writing

Text engineered around a phrase tends to be vague about everything else. Answer extraction rewards the opposite: a sentence that states one thing plainly, with a number, a date, or a named source attached.

Treating trust signals as compliance

An author byline added because a checklist demanded it, linking to nothing, contributes little. Authorship matters because it makes a claim accountable to someone. The signal and the substance are the same thing here, which is inconvenient but true.

The part nobody can sell you

This is worth saying plainly because the alternative framing — that some configuration unlocks AI traffic — leads teams to spend on the wrong things. The realistic goal is to be legible, identifiable, and credible enough that you are a reasonable thing to draw on. That is achievable, measurable, and mostly unglamorous.

Where to start

1. Confirm your important pages return real content in the served HTML, before JavaScript runs.
2. State your identity in machine-readable form with Organization structured data that matches your visible information.
3. Make sure About, contact, and policy pages exist and are reachable from your homepage.
4. Attribute and date anything time-sensitive.
5. Cut or consolidate thin pages rather than adding more.

None of that is exotic. It is the same discipline that made sites good before answer engines existed, applied with a clearer sense of who is reading.