Glossary: SEO, advertising and AI
Short definitions of the terms that come up in reports and on calls. No jargon: explained the way we would explain them to a client.
SEO
10 termsCanonical page
The URL that search engines consider the primary one for a document when the same content is available at multiple URLs. It is specified by the rel="canonical" tag and determines which version of the page gets indexed.
ReadCrawl budget
How many website pages a search engine bot manages to crawl per visit. The limit is not infinite: if it is spent on junk URLs, important pages wait weeks for indexing.
ReadE-E-A-T
A set of signals Google uses to evaluate trust in content: experience, expertise, authoritativeness, and trustworthiness. Not a formula or a score, but what quality raters and algorithms look for.
ReadEPOS (Expertise, Usefulness, Originality, Substantiveness)
A framework of four criteria by which Yandex evaluates content: expertise, usefulness, originality, substantiveness. Described in Webmaster Help, it applies to both regular ranking and inclusion in Alice AI answers.
ReadKeyword cannibalization
Two website pages compete for the same query, and the search engine does not understand which one to show. As a result, neither grows: signals are split in half.
ReadPage recrawling
A request to a search engine bot to rescan a URL. It is submitted manually via Yandex.Webmaster or Google Search Console when a page is published or updated and you need search engines to discover it faster.
ReadScaled content abuse
Google's policy against the mass publication of pages created to manipulate search results rather than benefit the reader. The wording appeared in March 2024 and has since remained the main reason for penalties.
ReadSchema.org markup
A tag vocabulary a website uses to explain the meaning of its content to search engines: where the article is, where the author is, where the organization is, where the question and answer are. The machine stops guessing based on layout and gets direct statements about the page.
ReadSemantic core
A list of search queries targeted for website promotion. The work starts here: without a core, it is unclear which pages are needed at all and what to write on them.
ReadUser behavior signals
What people do on the website after clicking from search: how much they read, where they click, whether they return to the search results. Search engines consider this a hint of whether the page was useful.
ReadArtificial intelligence
5 termsContext window
How much text the model sees at once. Everything that didn't fit does not exist for it – hence the lost details in a long dialogue and forgotten pieces of a large document.
ReadFine-tuning
Fine-tuning a pre-trained model on your own data. It costs money and time, and is needed less often than it seems: more often the task is solved by a well-crafted prompt or document injection.
ReadLLM
Large language model – the technology behind ChatGPT, Claude and Gemini. It predicts the next piece of text based on everything read during training, rather than searching for a ready-made answer in a database.
ReadLLM hallucination
The model confidently outputs a fabricated fact: a non-existent quote, link, or condition. For business, this is a risk – the assistant might attribute prices and deadlines to you that do not exist.
ReadToken
The smallest piece of text processed by a language model: a part of a word, a punctuation mark, a space. Both the prompt size and the model's processing cost are calculated in tokens.
ReadGEO
3 termsAI search results
A search engine response that an AI model compiles from multiple sources and displays instead of or above the list of links. In Yandex, these are Алиса AI answers, in Google – AI Overviews and AI Search mode.
ReadGEO promotion
Working to ensure the website appears in the answers of generative AI models: Алиса AI, ChatGPT, Perplexity, Google AI Overviews. Not a position in a list of links, but a place among the sources from which the model compiles the answer.
ReadRAG (Retrieval-Augmented Generation)
Retrieval-Augmented Generation, an AI model workflow where it does not answer from memory, but first searches for documents and then compiles an answer from the findings. This exact mechanic underpins all GEO promotion.
ReadNeed a review, not a definition?
We will look at your site and explain in plain words what is happening there and what to fix. Free, within two business days.