SEO, AEO and GEO: what marketers should prioritise

Every few years search marketing gains a new acronym and a new set of promises. AEO and GEO are the current ones. The question behind them is old and sensible: when someone asks a question, will your content be part of the answer, and will that lead anywhere useful?

Here is the short version. SEO is being found and chosen in search results. AEO (answer engine optimisation) is being the direct answer in features that respond to a question. GEO (generative engine optimisation) is being included or cited in AI-generated answers. The three overlap heavily, and people use the terms inconsistently. For Google’s AI features, Google’s own guidance says the foundation is ordinary SEO, with no special markup or files required. So prioritise useful original content, pages that can be crawled and indexed, clear authorship and sensible internal links, and measure the whole path from being mentioned to receiving an enquiry.

Three terms for one overlapping job

The terms are marketing labels, not technical standards. Different writers draw the boundaries differently. This is how I use them:

In May 2026 Google published a guide to optimising for generative AI features in Google Search. Its message, as widely reported at the time, is that for Google these practices are still SEO: the AI features draw on the same search index, so a page that is not technically sound or good enough to rank will not do well in AI answers either.

From discovery to enquiry: five different outcomes

The most useful thing I can offer on this subject is a distinction. Discussions of AI search often blur five outcomes that can each happen without the next:

1. Discovered and indexed. The search system can reach the page and has stored it.
2. Used in an answer. The page’s content informed an AI-generated answer.
3. Cited. The answer shows a link or names your site.
4. Visited. Someone clicks through.
5. Useful action. The visitor does something that matters to you, such as reading further, subscribing or enquiring.

A page can be cited and never visited, because the answer satisfied the question. It can be visited by the wrong people. It can bring one enquiry worth more than a thousand visits. If you report only the first three, you are measuring visibility, not value.

What to prioritise now

These practices are supported by Google’s documentation and by long-standing search fundamentals. None depends on a trick.

• Useful, original material. Worked examples, clear judgements, first-hand evidence you are allowed to share. An AI system summarising ten similar pages has no reason to cite the eleventh.
• Public, crawlable, indexable text. The answer should exist as text in the page, not only in an image, a PDF behind a form or content that appears only after interaction. Google’s documentation on AI features and your website says a page needs to be indexed and eligible to appear in Search with a snippet to be shown as a supporting link, and that there are no additional technical requirements.
• A clear answer early. Put a direct, self-contained answer near the top, then the reasoning. This is a readability choice that also happens to suit answer features, not a formula.
• Accurate authorship and dates. Who wrote it, what they know, when it was published and meaningfully updated. Structured data should describe what is visible on the page.
• Sensible internal links. Useful pages that nothing links to are hard to find, for people and for crawlers.

Claims to treat with care

A lot of AEO and GEO advice is hypothesis presented as fact. The table below separates what is documented from what is speculative. It concerns Google unless stated; other AI platforms run their own crawlers, controls and policies, so check their current official documentation before acting on any product-specific claim. Whether a crawler may use your content for training and whether your page can appear in an AI search product are separate questions with separate controls.

Two notes on this table. First, it will age: platforms change quickly, so date your own version and recheck it. Second, “low priority” is not “never”. This site publishes an llms.txt file; I treat it as a small courtesy for tools that use one, not as a ranking lever.

Measure it honestly

Measurement is where AEO and GEO advice tends to be thinnest. Build a simple sheet that follows the five outcomes:

Some practical limits:

• Search Console. Google launched generative AI performance reports in Search Console in June 2026 and completed the worldwide rollout by the end of August. At launch they showed impressions in AI features by page, country, device and date, not clicks or queries. Check the current documentation for what the report now includes.
• Observed citations. Checking whether an assistant cites you means asking it questions. Answers vary by person, place, time and wording, so a manual check is a sample, not a ranking. Use a fixed list of questions and record the date, so you compare like with like.
• Referrals. Some AI tools pass a referrer and some do not, so analytics will undercount AI-driven visits.
• Useful actions. This is the column that matters commercially, and the one that needs your CRM, not just your analytics. Connecting visits to enquiries and pipeline is its own subject: B2B marketing metrics: connect lead quality to pipeline.

A worked example: a new site’s backlog

This example is illustrative. The site and its problems are invented to show how to prioritise.

A fictional bilingual consultancy site has a backlog of eight discovery ideas from different advisers. Three problems stand out when someone reviews the site against the five outcomes:

• Its most useful guide is orphaned: no page links to it, and it is not in the navigation.
• A pricing explainer is not indexable: an old noindex tag was never removed.
• A popular article contains an unsupported claim: a statistic with no source that a reader or an answer engine may repeat.

The team chooses three actions: link the orphaned guide from the pages that discuss its topic, remove the stray noindex tag and request indexing, and either source the statistic or delete it. Each has a clear reason and a way to check it worked.

It defers the rest: adding FAQ markup to every page, rewriting every paragraph to a fixed length, creating an llms.txt file before anything else, and buying an “AI visibility” tool before it has a baseline. None of these is forbidden. None beats fixing a page nobody can find.

Where AI helps, and where it cannot

AI is useful for auditing your own pages. Ask it whether a page answers its main question in the first paragraph, which claims lack a source, which obvious follow-up questions are missing and which sections repeat each other. It is also useful for drafting candidate questions a reader might ask.

What it cannot tell you is whether a page will rank or be cited. An assistant confidently predicting its own citation behaviour is not evidence. The quality work before publishing matters more: AI content quality: an editorial checklist before publishing. For a step-by-step workflow once the priorities are clear, see how to get cited in AI search. And for deciding what deserves a page at all, start from B2B content strategy.