Glossary

Experience, Expertise, Authoritativeness and Trust — the frame Google raters use to judge page quality.

Not a ranking factor with a number behind it, but a checklist: who wrote this, where the figures come from, whether sources are linked, whether the piece is kept current.

Generative Engine Optimization — getting your material cited inside AI answers.

Unlike classic SEO, the fight is for a citation inside the answer, not a position in a list. Explicit definitions, figures and question-answer structure help.

The AI answer block above Google organic results.

It takes a share of clicks: the user reads the summary and never visits. Sources rotate per query, so presence there is tracked separately from rankings.

A plain-text file at the site root describing the project for AI crawlers.

Like robots.txt, but about meaning rather than restrictions. The standard is young and optional — and cheap to add.

The tag by which a page declares its primary address.

Needed when one piece is reachable at several URLs. Common mistake: the canonical points at an address that answers with a redirect, so the hint is ignored.

Google’s three field metrics: LCP, INP and CLS.

Largest paint, interaction responsiveness and layout stability, measured on real users rather than in a lab.

How many pages a bot is willing to fetch on your site over a period.

Matters from tens of thousands of pages up. Wasted by filters, sortings, parameter duplicates and redirect chains.

The collected and grouped list of queries a site targets.

The value is in grouping, not volume: which queries one page can cover and which need separate pages.

Splitting queries into groups by how much their result pages overlap.

If the top results for two queries share URLs, one page covers both. It settles the “one article or two” question without guessing.

The visible text of a link.

Tells the engine what the target page is about. Over-optimised commercial anchors are a classic cause of penalties.

The whole set of external links: who links, with what anchors, how fast.

Judged by composition, not count: donor topics, their own traffic, growth dynamics.

A site that links to yours.

Look at live traffic and topical fit rather than bought metrics: a high-rating domain with no traffic is usually a link farm.

A crawler that revisits pages almost immediately after publication.

Gives quick but temporary traffic to fresh material; positions usually settle after regular reindexing.

Faking user behaviour to lift rankings.

Works until the filter hits. Visible in analytics as spikes of search visits with identical depth, devices and time on page — worth checking even if you never bought it.

A machine-readable description of what is on the page.

Usually delivered as JSON-LD, a separate block in the code. Enables rich snippets and makes the page easier for AI to retell.

A large language model — a network trained to predict the next chunk of text.

Both the usefulness and the main risk follow from that: it continues text confidently even where it has no facts.

A fact invented by a model and delivered in the same tone as a true one.

Figures, dates, study names and links suffer most. The cure is checking the source, not rephrasing the prompt.

A numeric vector a text is turned into, so meanings can be compared arithmetically.

Closeness is the angle between vectors. Semantic search, “related posts” blocks and source selection for AI answers all rest on this.

Retrieval-Augmented Generation: the model fetches documents first and answers from them.

Cuts invention and allows citing a specific document — but the answer is only as good as the retrieval.

The instruction given to a model: role, task, constraints, output format.

A working prompt is longer than the intuitive one: it states acceptance criteria and what not to do.