AI SEO: what it is, what works, and how to measure it

Search volume for AI SEO passed 8,100 a month in the United States while the category still has no settled definition. That gap is the problem. Half the pages ranking for it mean "using AI tools to do SEO faster" and the other half mean "getting your brand into AI answers". Those are opposite jobs. This page covers the second one, says plainly where the first one fits, and gives the measurement framework for both.

AI SEO means two different things, and the difference matters

The term collapsed two disciplines into one phrase, which is why buyers keep signing contracts that do not deliver what they expected.

AI SEO as production. Using large language models to draft briefs, cluster keywords, generate meta descriptions, build internal link maps and write first drafts. This is traditional SEO with the labour cost pushed down. The output still competes on a Google results page and is still judged by rankings, clicks and sessions.

AI SEO as visibility. Getting a brand named and cited inside ChatGPT, Perplexity, Gemini, Claude and Google AI Mode answers. There is no results page and no position, so rankings and sessions cannot measure it. The unit of success is a citation, and the input that produces one is coverage across sources the engine already trusts.

Both are legitimate. They need different budgets, different reporting and different people. Buying the first while expecting the second is the most common failure in the category right now. If your reporting still leads with keyword positions, you bought production. If it leads with share of answer, you bought visibility.

How AI SEO differs from traditional SEO

The overlap is real: crawlability, structured data and genuinely useful content serve both. The divergence is in what you optimise and what you count.

What actually works, measured

Most AI SEO advice is asserted rather than measured. The figures below come from a corpus of 39,948 citation events across 72 audited crypto brands, April to August 2026, and the full method is published on the methodology page.

Coverage predicts citation, and almost nothing else does. Citation volume tracks the number of pages that mention a brand at approximately 1.48 citations per page. The coefficient of variation on that ratio is 9.5 percent across payment processors, exchanges, DeFi, DePIN and gambling. Doubling the pages that mention a brand roughly doubles its citations. No other variable in 40,000 events moves the multiplier by more than about 16 percent.

Domain authority does not predict citation. Sites holding fewer than 35 referring domains are being named by AI engines in competitive categories today, while sites with over 3,000 referring domains are not. The backlink graph and the citation graph are not the same graph.

Being the subject of a page beats being listed on one. Subject-of-page treatment returns 6.22 citations per brand, against 3.78 for being ranked in a third-party list and 3.39 for absence. The subject-of-page interval does not overlap either of the others.

Most citations are not yours to control directly. 92.22 percent land off the brand's own domain. Owned publishing matters, but the majority of the surface is third-party, which is why earned coverage moves AI visibility more than a blog alone.

The AI SEO checklist

In order of leverage, based on the evidence above rather than on convention.

• 1. Confirm the engines can reach you. Allow GPTBot, OAI-SearchBot, ChatGPT-User, PerplexityBot, Perplexity-User, ClaudeBot, anthropic-ai, Google-Extended and CCBot in robots.txt. Server-render anything that matters. A JavaScript-only page is invisible to most AI crawlers, and this failure is silent.
• 2. Fix the entity layer. One consistent name, one consistent description, and a sameAs graph pointing at profiles that already exist. Directory and reference profiles are disproportionately effective here because engines lean on them to resolve who you are.
• 3. Own the page about yourself. A definitive, well-structured page for your own category question, because subject-of-page is the strongest treatment measured.
• 4. Expand coverage. Get mentioned on more pages that engines already read. This is the lever, and it is mostly earned coverage, comparison pages, directories and communities rather than more posts on your own blog.
• 5. Structure for extraction. A direct answer in the first paragraph, question-shaped headings, tables for comparisons, and FAQ markup that matches the visible text. This does not create citations on its own, but it makes an already-trusted page easier to lift from.
• 6. Measure on a fixed prompt set. Same prompts, same engines, weekly. Changing the questions between runs makes the movement unreadable.

How to measure AI SEO

Three metrics, and no vanity substitutes.

Mention rate. The share of prompts in your fixed set where the brand is named at all. This is the first thing to move and the easiest to read.

Citation rate. The share where your own domain is cited as a source. Lower than mention rate for almost everyone, and the harder of the two to shift.

Share of answer. Your mentions as a proportion of all brands named across the set. The only one of the three that tells you whether you are winning or merely present.

One caution that applies to every tool in this category: AI answers are non-deterministic. The same prompt can return different brands on consecutive runs. Single-run scores are noise. Sample repeatedly, hold the prompt set constant, and read the trend rather than the number. This is also why two tools report different scores for the same brand on the same day.

Tools

The tooling market splits the same way the term does. Production tools help you make content faster. Visibility tools tell you whether engines name you. A full comparison of the visibility side, scored on published criteria, is in the AI visibility tools breakdown, and the crypto-specific set is covered in crypto AEO tools compared.

What no tool does: close the coverage gap. Every product in the category measures whether you were named. None create the pages that cause the naming. Budget for the measurement and the coverage separately, because the evidence says the second one is what moves the metric.

AI SEO for crypto and Web3

Crypto is the category where this shift arrived first, because crypto buyers were early to ask an AI before opening an account, bridging funds or connecting a wallet. It is also the category where generic tooling breaks down hardest: a prompt set adapted from B2B SaaS will not surface how someone chooses an exchange, and a generic taxonomy will group an L2 with a tax tool.

Three things differ in practice. Engines apply heavier caution to financial and YMYL queries, so trust signals such as audits, proof of reserves and independent security coverage carry more weight. The source population is unusual: crypto answers lean on a mix of category media, aggregators, community platforms and video that does not resemble a typical B2B citation profile, covered in where AI gets crypto answers. And competitor sets need to be modelled per vertical, because exchange buyers, DeFi users and wallet buyers ask structurally different questions.

AI SEO, GEO, AEO and LLM SEO: the same work

Four names, one discipline, and the naming is unsettled because the category is roughly two years old. AEO (answer engine optimization) frames it around the answer. GEO (generative engine optimization) frames it around the generative system producing that answer. LLM SEO frames it around the model. AI SEO is the broadest and also the most ambiguous, since it is the one term that also means production.

Practically: if a vendor uses these interchangeably, that is normal and not a red flag. If a vendor cannot tell you which metric they will report, that is the red flag. The deeper comparisons are in AEO vs SEO and AEO vs GEO.

AI SEO, answered