Skip to content

AI SEO

What is AI SEO? A working definition, and what it actually changes

AI SEO is the practice of getting your business cited inside AI-generated answers. It overlaps with SEO but optimises for retrieval and citation rather than ranking.

Mawkaii Updated 9 min read

Key takeaway

AI SEO (also called answer engine optimisation or GEO) is the practice of making a business likely to be named inside AI-generated answers from tools like ChatGPT, Perplexity, Gemini and Google AI Overviews. It overlaps substantially with traditional SEO, because answer engines retrieve heavily from pages that already rank. It differs in what it optimises for: passage-level extractability, consistent entity descriptions across the web, depth of structured data, and crawler access for models that do not execute JavaScript.

There is a lot of confident nonsense being written about this, in both directions. One camp says AI SEO is a fabricated discipline invented to sell retainers. The other says search is dead and everything you know is obsolete.

Both are wrong, and the truth is more useful than either.

The definition

AI SEO is the practice of making a business likely to be named inside AI-generated answers. You will also see it called answer engine optimisation (AEO) or generative engine optimisation (GEO). The terms are interchangeable in practice.

The target is not a position in a list. It is inclusion in a synthesised paragraph that names two or three companies and omits everyone else.

What actually changed

Classic search returns ten links and lets the person choose. Being fourth is worth something. Being tenth is worth a little.

An answer engine returns one response. If a model names three companies in your category and you are not one of them, you received nothing — not less traffic, nothing. The distribution of outcomes is far more brutal than a ranked list.

That is the change. Not that search died, but that a growing share of it resolves into a winner-take-most format.

Five things AI SEO optimises for

1. Retrieval, not ranking

Retrieval-augmented systems fetch and rank passages, not pages. A brilliant 2,000-word article that answers the question in paragraph fourteen may never have that paragraph retrieved, because the chunk containing it lacks the context to stand alone.

Practically: answer the question in the first two sentences under a question-shaped heading, then elaborate. Every section should make sense read in isolation.

2. Entities, not keywords

Models reason about things and their relationships. Your company is an entity. So are your services, locations and people. The model’s confidence in citing you depends heavily on how consistently those entities are described across the web.

If your site says you were founded in 2019, LinkedIn says 2020, and Crunchbase says 2018, you have introduced uncertainty. Uncertainty means the model names someone it is surer about.

A link passes authority. A mention passes corroboration — and corroboration is what a retrieval system is looking for when deciding whether a claim is safe to repeat.

Being accurately described on directories, review platforms, comparison pages and in press coverage matters even when those mentions carry no link at all.

4. Machine-readable facts

Structured data has been undervalued for a decade because its visible payoff was a rich snippet. Its real payoff now is that it hands a model unambiguous facts it does not have to infer from prose.

Complete, correctly nested Organization, Service, Product, FAQPage and Article markup is one of the cheapest interventions available, and one of the most commonly done badly.

5. Crawler access

Most AI crawlers do not execute JavaScript. If your content is client-rendered, a model may see an empty shell where your product page should be.

This is the failure we find most often, and the one clients are most surprised by. It is worth checking before spending anything on strategy.

What we do not know yet

Being honest about the limits is part of the discipline:

  • Attribution is genuinely hard. Much of the value shows up as branded search rather than AI referral traffic, which makes clean measurement difficult.
  • llms.txt adoption is unconfirmed. It costs almost nothing to publish and forces useful structural discipline, but nobody should claim it is a proven ranking factor. It is not.
  • The engines change monthly. A tactic that worked in March may be irrelevant by September. Any agency presenting a fixed playbook is selling you a snapshot.

Where to start

Run twenty prompts your buyers would plausibly type into ChatGPT and Perplexity. Save the answers. Count how many name you, and note who they name instead.

That single exercise takes an afternoon and will tell you more about your position than any audit you could buy. If the answer is uncomfortable, the work is worth doing. If you are already being cited consistently, spend the money elsewhere — we will tell you that too.

  • AI SEO
  • answer engines
  • GEO
  • structured data

We do this for a living.

If this is the sort of thinking you want on your side, the first conversation is free.