Read this first — SEO for engineers

This course assumes you can build software and read a measurement, and that you have never worked in SEO. The lessons teach one unknown term at a time; everything else — the working vocabulary every study and every dashboard throws around — is defined once, here. Read it once before episode 1, then return by term when prose surprises you. Definitions of the course's own concepts (AI Overview, citation, chunk) live in the 📖 SEO Glossary; this page carries the industry's supporting vocabulary.

Each term gets a plain definition, then an In this course line: how the number is used in the lessons and where it is commonly misread. If a lesson's sentence still does not parse, that missing context belongs on this page; flag it and it gets written.

Click arithmetic

Impression
One rendering of your page on a results page. Google showed it, clicked or not. An impression is an event on Google's side; your site sees nothing of it.

In this course: most studies report clicks per thousand or per million impressions. Read the denominator before reading any rate.

Click-through rate (CTR)
Clicks divided by impressions: the share of people who see a result and click it. A rate, not a count: "CTR fell 78%" says the share shrank, nothing about the absolute number of clicks.

In this course: always paired with who is in the denominator. Two studies with different denominators produce two different levels; only ratios drawn from the same population compare directly.

Keyword impression
One appearance of a result for one tracked keyword: "someone searched k, the results page rendered, your link was on it". A keyword-focused study (Ahrefs) samples a fixed keyword list and measures the position-1 result's CTR across those impressions.

In this course: the denominator that makes matched-across-time comparisons possible. Do not confuse it with page-type samples.

Position 1
The first organic (unpaid, un-promoted) result on a results page. "Position-1 CTR" measures that one slot; "all-position" averages fold positions 1 through 10 together and read much lower than Position-1 CTR on the same searches.

In this course: check which before comparing any two CTR numbers: the quiz explanations hinge on knowing which denominator a figure averages over.

Intent (informational / transactional)
Why a person searched: an informational query seeks facts or how-to knowledge ("how big is the ai search market"); a transactional query seeks an action (a purchase, a signup). Studies sample by intent because collapse and recovery differ between them.

In this course: the modifier "informational keywords" marks Ahrefs' denominator: informational queries only. Read every study's intent mix before borrowing its rate.

Denominator vs level
A rate has a numerator (clicks) and a denominator (impressions of some defined population). The denominator defines the level; two rates with different denominators are two different populations and cannot be quoted as one trend.

In this course: the 78% (same keywords, two dates) and the 52% (two page types, one date) are both true and both different. Ratios of per-impression rates sharing an inflated denominator largely cancel; absolute rates read low.

Counterfactual forecast
An estimate of what a metric would have reached on an unchanged timeline, used as the comparison baseline when history itself changed (here: the AI Overview rollout). Ahrefs' 58% headline compares the measured 1.6% against a forecast of about 3.7%, not against the raw 7.3%.

In this course: whenever a study headline says "loss", ask loss against what: the observed past or the modeled no-event path.

People and channels

Conversion / conversion rate
A conversion is the visitor doing what the site wants (a lead, a signup, a sale). The conversion rate is conversions divided by visits. Sites define it themselves, so conversion numbers compare only within one site's own definition.

In this course: the measure that separates "traffic" from "audience". AI-referred visitors convert at multiples of search visitors in two studies, both correlational.

Referrer / referrer segmentation
The browser sends the destination server an HTTP header (historically misspelled Referer) carrying the URL the visitor came from. Analytics groups visits by it: a visit from chatgpt.com or perplexity.ai is an AI referral (on desktop web; in-app mobile webviews often strip the referrer, landing those visits as direct traffic). The event happens on your server, not Google's; no search-side Google tool (Search Console) sees it.

In this course: the attribution instrument for AI-originated clicks from the standalone assistants: each arrives with its own host in the referrer. Clicks out of Google's own AI surfaces (AI Overviews, AI Mode) carry plain Google referrers, the same as organic, so referrer logs cannot split them. Search Console cannot see post-click behavior.

Funnel
The visitor's path from arrival toward the site's desired action (visit → lead → signup → sale); conversion numbers attach to one stage of one funnel, so two funnels' rates do not compare directly.

In this course: why "a different funnel with different selection" explains the signups-vs-leads spread between the conversion studies.

Organic search
Search-engine visits that arrive through unpaid results. "Organic CTR" is the click-through rate of those unpaid results; paid placements sit outside every number in this course.

In this course: what the episodes measure and defend. If a figure says only "CTR", organic is implied throughout.

Behavioral panel
A research method in which consented users run instrumented software and their real browsing is recorded, versus asking people to recall their behavior. Pew's episode-1 numbers (900 US adults, 68,879 searches) come from panel measurement, not surveys.

In this course: why the panel numbers are session-level truth: what people did, not what they say they do.

Search-engine instruments

Search Console
Google's reporting console for site owners: which queries showed your pages, how many impressions and results-page clicks, indexing coverage. Everything in it is measured on Google's side of the click. A property is a site-level registration: one domain (a domain property) or one URL prefix. Settings inherit down the chain, so a parent property's state can decide for your site.

In this course: two reports matter: the main Search results report (clicks live here) and the Generative AI performance report (AI-surface impressions only, no click column). For clicks the standalone assistants carry, attribution is a referrer job; Google's own AI surfaces stay merged in the main report.

Bing AI Performance report
Bing Webmaster Tools' report (public preview February 2026) showing page-level citations, grounding queries, and citation share across Copilot and Bing-grounded AI surfaces: the non-Google counterpart to Search Console's Generative AI performance report.

In this course: the second search-ecosystem citation instrument. Register your site for Bing to see the Copilot side; Google's surfaces stay on Search Console.

Canonicals
A canonical is each page's declared self-identity: a <link rel="canonical"> in the HTML head pointing at the one absolute URL that represents the page. It tells crawlers which copy of a page counts when variants (query strings, http vs https, trailing-slash forms) could serve the same content. It is the deduplication signal: without it, a retrieval system cites a duplicate.

In this course: the floor's rule is one absolute canonical per page with a consistent trailing slash, so no duplicate-page ambiguity exists to mis-cite.

Structured data / JSON-LD
Machine-readable metadata embedded in a page (JSON-LD is the JSON-in-script encoding Google prefers) declaring what the page is (an Article, a Course, breadcrumbs). Rich results are Google's enhanced listings (breadcrumb trails, FAQ boxes, review stars); this markup makes a page eligible for them, on top of ranking. It is not ranking fuel.

In this course: part of the static-site floor. FAQ rich results are retired (May 2026); write the prose, skip that markup.

llms.txt
A proposed, robots.txt-shaped site file that tells AI crawlers how to read the site; no measured effect. Google's public guide explicitly dismisses it.

In this course: the exemplar of a "second front door" tactic. The lesson's rule: AI features are rooted in core ranking; no second algorithm means no second door. The site itself keeps an llms.txt as a developer convenience bundle (a plain-text course file you can paste into your coding assistant), not an AI-discovery signal; the difference between a convenience artifact and a tactic is the point.

Impression over-counting window
Google's own admission that Search Console over-counted impressions (not clicks) from May 13, 2025 through April 27, 2026. Absolute per-impression rates inside the window read low; ratios across segments sharing the inflated denominator largely cancel.

In this course: why Seer's citation-premium ratio stands while its in-window levels are floors, not facts.

Sources & further reading

  • Get started with Search Console Google Search Central · official guide · verified 2026-11-13

    The official map of the reports referenced in the search-engine-instruments section; what Google itself considers each report to measure.

  • AI Performance - Bing Webmaster Tools Microsoft Bing Webmaster Team · official documentation · verified 2026-11-13

    The non-Google citation instrument: page-level citations, grounding queries, and citation share across Copilot and Bing-grounded surfaces.

  • Referer header MDN Web Docs · reference · verified 2026-11-12

    The exact mechanism behind referrer segmentation, on your side of the click.

  • Google users are less likely to click on links when an AI summary appears in the results Pew Research Center · research summary · verified 2026-11-12

    The behavioral-panel study behind episode 1's session-level numbers, and a live example of how this vocabulary reads in the wild.

See also