Technical SEO

A RAG system works in two coordinated parts: a retriever that searches a knowledge source for passages relevant to the query, and a generator that writes the answer using those passages as context. The retrieved text is inserted into the model's prompt, so the response is anchored to specific documents instead of produced from training weights alone.

Three pieces make it work: a document collection broken into chunks, an embedding model that turns those chunks and the query into vectors, and a vector index that returns the closest matches. Most production pipelines add a reranking step that scores the candidate passages before the best few reach the model. Change any piece and the answer changes, because the generator can only work with what the retriever hands it.

RAG sits underneath the AI answer surfaces marketers care about: Perplexity, Google AI Overviews, and Microsoft Copilot all use retrieval-driven architectures to ground their responses in live web sources. This is why AEO is largely a retrieval problem, and why AirOps helps teams structure and monitor content so those retrievers can find, parse, and cite it.

Measured as a daily score, SERP volatility captures the rate and size of ranking changes across a tracked keyword set. A high reading means positions are moving sharply and many domains trade places in the top ten. A low reading means the same pages hold their spots.

Volatility trackers follow a common method. They monitor a fixed sample of keywords, record positions each day, then measure how far they moved from the day before. The bigger the average movement across the sample, the higher the volatility score. Some indexes break the score down by industry, because a finance update can churn one category while others stay calm.

Volatility sits alongside algorithm tracking and rank tracking as an early-warning signal. It flags market movement, and it does not by itself explain why your own site dropped. The churn now reaches more than classic blue links, since AI Overviews and other answer features reshape what sits above the organic results. AirOps tracks movement across AI answers and classic rankings in one view, so you can see when a shift touches both surfaces.

Semantic search works by converting queries and documents into numerical vectors, then ranking results by how close their meanings sit in that vector space. It reads context, entities, and relationships, so a search for "apple nutrition" returns fruit facts while "apple earnings" returns the company.

Three components make this work. Natural language processing (NLP) parses grammar and intent. Embedding models place words and passages into a shared vector space where related meanings cluster together. A knowledge graph stores entities and the connections between them, so the engine understands that "espresso" relates to "coffee" and to "Italy."

Semantic search sits underneath modern AI search. Google has moved this way for over a decade, launching its Knowledge Graph in 2012 and RankBrain, its first deep-learning system in Search, in 2015. When ChatGPT or Google AI Overviews retrieve sources before generating an answer, they run semantic retrieval first, so the signals that help meaning-based ranking also shape which pages get cited. AirOps helps teams structure content and entities so retrieval systems can read and match them accurately.

In practice, schema markup translates the meaning of a page into a standardized vocabulary from Schema.org, so an answer engine can identify a FAQ, a product price, or an author as those exact things. Most implementations use JSON-LD, a script block placed in the page's code that maps visible content to types like FAQPage, Product, Organization, HowTo, and Author.

The markup itself is a set of key-value statements: each property names an entity or attribute and points to the matching text on the page. For it to work, the structured data must reflect what a reader actually sees; markup that describes content missing from the page violates search engine guidelines and gets ignored.

Schema markup sits alongside clean heading structure and answer-first writing as one signal answer engines use to decide what to cite. It does not replace quality content, and on its own it rarely moves citations; its value shows up when accurate markup pairs with a page that already answers the question. AirOps tracks pre- and post-refresh citation rates so you can see whether adding schema and structure actually improved how often AI engines cite a page.

Resources: See how to implement FAQ, Product, and Organization schema for answer engine optimization

Onsite SEO covers every optimization you make inside a page's own code and content, from the title tag and heading hierarchy to internal links, page speed, and structured data. It answers a simple question for each URL: can a crawler reach this page, understand what it covers, and trust it enough to rank and cite it?

Three layers have to hold up at once. The content layer signals relevance through the words, headings, and answers on the page. The technical layer controls whether a page can be crawled and indexed at all, covering site speed, mobile rendering, canonical tags, and clean URLs. The semantic layer, including schema markup and entity clarity, tells engines what each element means.

Onsite SEO sits between technical SEO and content strategy, borrowing from both. It is the layer you control fully, which is why teams using AirOps often start here when auditing why strong pages fail to rank or get cited by AI answer engines.

See how onsite structure feeds both traditional rankings and AI citations

In AI systems, a hallucination happens when a large language model fills a gap in its knowledge with statistically likely words in place of verified facts, producing an answer that sounds authoritative but has no grounding in reality. The model predicts the next token based on patterns in its training data, so it optimizes for plausibility, and truth is a separate question it never directly checks.

Hallucinations cluster into a few types. Factual errors state something untrue about the world. Faithfulness errors distort the source or prompt the model was given. Fabrications invent details outright: fake statistics, nonexistent studies, dead URLs, or product features that do not exist. For a brand, the most damaging version blends real, verifiable details with one wrong fact, which makes the error hard to spot.

This sits close to source grounding and citation accuracy, but a hallucination is a property of the generated text itself, while grounding describes whether that text traces back to a real source. AirOps monitors how engines describe your brand across ChatGPT, Perplexity, and Google AI Overviews, so you can catch a fabricated claim before it spreads.

Resources: how leading teams build a system of record that keeps brand facts accurate

Vector search retrieves information by measuring the distance between vectors, which are lists of numbers that encode the meaning of text. Each piece of content becomes a point in a high-dimensional space, and the search returns the points that sit closest to your query. Because the comparison happens over meaning, results can match a query that uses different wording from the source passage.

Three components make this work. An embedding model converts your text into vectors. Those vectors go into a vector index, which organizes them so the system can scan millions quickly. To rank results, the system applies a similarity metric such as cosine similarity or Euclidean distance.

Vector search sits underneath several concepts you already hear about. Semantic search is the user-facing experience that vector search often powers. Keyword search remains useful for exact matches like product codes, and many systems combine both. Retrieval-augmented generation (RAG), the architecture behind many AI answer engines, uses vector search in its retrieval stage to pull passages before a model answers (Lewis et al., 2020). AirOps builds content designed to be retrieved and cited in that stage.

A rich result is any Google Search listing that Google renders with enhanced features because valid structured data on the page told it what the content represents. It answers a search with more than a title and description, showing details like review stars, price and availability, event dates, or recipe cooking times directly in the listing.

Schema.org supplies the shared vocabulary for this. Structured data applies that vocabulary to specific elements on your page. Google then reads the markup and decides which enhanced format, if any, to display. Google accepts three markup formats for this work and recommends JSON-LD over the older Microdata and RDFa options.

Rich results sit next to other search enhancements like featured snippets and AI Overviews, but those are generated from Google's own extraction, while rich results depend on markup you supply. The same structured data that qualifies a page for rich results also helps answer engines parse it, which is why AirOps treats schema as a shared signal for both Google rich results and AI citations.

Resources: step-by-step guide to implementing schema markup for answer engine optimization

An internal link is a hyperlink that sends a visitor or a crawler from one page on your site to another page on the same site. Internal linking is the discipline of building those connections deliberately, so every important page is reachable and the anchor text tells search and answer engines what each linked page covers.

Each link has a source page where it sits and a destination it points to, with anchor text in between that describes the target. When those parts align, the link passes ranking signals, sometimes called link equity, and helps crawlers reach pages they would otherwise miss. Missing links leave pages stranded with no path in, which search systems read as low importance.

Internal linking sits alongside site architecture and topic clusters as part of how your site signals structure. External links and backlinks come from other domains, while internal links stay fully within your control. Platforms such as AirOps can map internal link coverage across a large library and flag pages that lack the connections they need.

Passage ranking is the retrieval step where a system compares each candidate section of a page to the query and scores those sections by how directly they answer it. The output is an ordering of passages within a single page, so one section can rank strong while another on the same page is ignored.

The mechanism depends on segmentation and semantic matching. A system splits the page into passages, often a heading and the paragraphs beneath it, converts each passage and the query into vectors, and measures their closeness in that vector space. Google introduced this capability in 2020 and said it would affect 7% of search queries across all languages once fully rolled out globally. AI answer engines apply the same passage-level logic when they assemble responses.

Passage ranking sits close to semantic search and retrieval-augmented generation (RAG), the process by which an AI model pulls external text into its answer. Semantic search finds meaning across pages; passage ranking decides which section inside a page wins. AirOps tracks passage-level performance so you can see which sections of your pages get cited across AI engines and where they lag.

Offsite SEO refers to every ranking and visibility signal generated outside your domain, from links and unlinked brand mentions to reviews, forum threads, and press coverage. Search engines and large language models read these external references as third-party confirmation that your brand is credible and relevant to a topic, which shapes how often you rank and whether an AI names you in an answer.

The work spans four broad components: link building from relevant, authoritative domains; digital PR and unlinked brand mentions; reputation and reviews on platforms like G2, Trustpilot, and Reddit; and local or entity citations that confirm consistent business details. Each component feeds the same underlying signal, which is external trust that engines cannot get from your own marketing copy.

Offsite SEO sits alongside onsite and technical SEO, and it now overlaps heavily with answer engine optimization, since AI systems lean on the same external sources to decide which brands to name. AirOps helps teams find, earn, and measure these third-party mentions so offsite work connects to visibility across both Google and AI search.

Resources: See how third-party sources drive most brand discovery in AI search

Index/no-index is a pair of indexing instructions that operate at the individual page level, telling a crawler like Googlebot whether to store a URL in the search index or drop it. The default state is index, so a page is eligible for search results unless you explicitly apply a noindex directive. You set noindex two ways: a robots meta tag in the page's HTML head, or an X-Robots-Tag in the HTTP response header, which also works for non-HTML files like PDFs.

The directive only takes effect if the crawler can reach the tag. If you block a URL in robots.txt, the bot never fetches the page, never sees the noindex, and the URL can still surface in results as a bare link. Indexing directives also stack with canonical tags, sitemaps, and internal links, all of which shape what search engines ultimately store.

Indexing directives sit alongside crawl directives and canonicalization in the technical SEO toolkit, but they answer a narrower question: should this specific page exist in the index at all. Because AI answer engines pull from the same crawled and indexed corpus, a noindex on a page also removes it from the pool that tools like AirOps track for citations.

Resources: how crawlability and indexation decide whether AI answer engines can cite a page

In the three-stage model of how search works, indexation is the middle step: crawling comes first, indexing follows, and serving results comes last. During indexation, a search engine reads a downloaded page and writes its text, title tags, alt attributes, images, and video into a structured index. Only pages stored there are eligible to be retrieved for a query.

Storage is not automatic. A crawled page can be dropped for low content quality, duplication, a noindex rule, or rendering problems that hide its content from the parser. The engine also evaluates language, topic, and duplication, then picks one canonical URL to represent any near-identical versions before writing to the index.

Crawling feeds indexation, and ranking draws from it, so a gap at this stage breaks everything downstream. Retrieval-based AI engines pull from indexed and accessible pages too, which is why AirOps tracks indexation as part of AI-search visibility.

Canonicalization sits early in the indexing workflow, and the canonical URL is the address Google selects as most representative from a cluster of duplicates. Google can only index that chosen version, so every other copy in the set stays out of results.

You send Google preference signals through the rel="canonical" tag, an HTML link element in the page head that names your preferred address. Google groups duplicate and near-duplicate pages into a cluster, then consolidates their ranking signals onto the single URL it marks canonical. Because indexing happens at the cluster level, only the canonical version becomes eligible to appear in search.

A canonical URL solves a different problem than a 301 redirect. The redirect moves visitors and crawlers to another address, while a canonical only expresses a preference. A noindex tag keeps a page out of results entirely, and an hreflang annotation maps language and region variants to the right audience. AirOps helps teams publish and refresh the canonical, citation-worthy version of every page.

The 301 in a 301 redirect is the HTTP status code a web server returns when a page has permanently moved, telling any client requesting the old URL to fetch the new one instead. This status code sits in the response header, so it works before the page loads, which is why crawlers and browsers honor it. Every 301 pairs one source URL with one destination URL.

Three parts make a 301 work: the old URL request, the 301 status code the server returns, and the Location header naming the destination. Search engines read this as a permanent instruction, drop the old URL from their index over time, and pass its ranking signals to the new URL. Google's Search Central documentation says a permanent redirect signals the target should become the canonical URL in search results.

A 301 differs from a canonical tag, which suggests a preferred URL without moving anyone, and from a 302, which keeps the old URL indexed. Use a 301 when the move is permanent. AirOps tracks whether the URLs an answer engine cites still resolve after a redirect, so migrations don't cost you citations.

Domain authority measures the overall ranking strength of a website by scoring its backlink profile against every other domain that Moz has indexed. Moz created the metric as the first score of its kind, and it remains a prediction of ranking potential. It does not report where you rank.

The score runs on a machine-learning model trained on real Google search results, fed by Moz's Link Explorer web index, a description Semrush corroborates. Its main inputs are the number of linking root domains, the quality of those links, and spam signals that flag manipulative link patterns. Because the model is calibrated across the whole index, your score can move when Moz refreshes that index, even in a week when you gained or lost no links.

Page authority applies the same modeling to a single URL, so a strong domain can still host weak pages. Ahrefs Domain Rating and Semrush Authority Score estimate similar ideas on their own indexes, and their numbers will not match Moz point for point. AirOps helps you build the authoritative content that earns the links and mentions these scores are trying to measure.

See which brand authority signals help you earn links and citations

Crawl budget combines two forces that Google balances for every site: the crawl capacity limit and crawl demand. The capacity limit is how many parallel connections your server can handle and how long Google holds them open. Crawl demand is how much Google wants to crawl your URLs, based on their popularity and how stale its copies are.

Every site starts with the same conservative default capacity, which rises only when Google has more demand and your server stays healthy, and falls when Google sees latency, 5xx errors, or 429 responses. That capacity is shared across all Google crawlers, so heavy fetching for one purpose reduces what is available for others. Perceived inventory is the demand factor you control most: the more URLs Google thinks exist, including duplicates and URL parameters, the more crawl it may spread thin.

Crawl budget sits upstream of indexation and ranking. A page that never gets crawled cannot be indexed, and a page that is not indexed cannot be retrieved or cited by AI answer engines that read the search index. AirOps helps teams keep the pages that matter crawlable and current so discovery work isn't wasted.

Search engines use a knowledge graph to store verified facts about entities and the ways those entities relate, so machines can interpret things instead of matching raw words. Google popularized the model in 2012 to power richer, more accurate search results.

Each entity becomes a node with a unique identifier, which keeps two people who share a name separate and traceable. Typed relationships, or edges, connect those nodes and label how they relate, such as "founded by" or "headquartered in." Attributes hang off each node to record properties like a founding date, category, or location.

This structure sits beneath features you already see, including knowledge panels, rich results, and the AI assistant answers that increasingly replace a results page. By May 2020, Google reported its Knowledge Graph held over 500 billion facts about 5 billion entities, a scale that lets it ground answers in confirmed data.