Glossary

Your audit report uses technical terms from web performance, UX research, and content analysis. Here's what they all mean in plain language.

Audit Modules

Accessibility & Usability: Single bucket for the dashboard's accessibility-and-usability evidence. Runs axe-core and IBM Equal Access for WCAG 2.1 AA / Section 508 violations; the Nielsen-10 heuristics scanner for usability friction (visibility of system status, error prevention, recognition over recall, help and documentation, etc.); content-level checks for new-window links, bold-as-heading misuse, all-caps blocks, and missing video captions; and a Tier-3 keyboard pass for focus traps and non-sequential tab order. The slug stays 'heuristic-audit' for backward-compat with existing tooltip deep-links.
AI Authenticity: This module analyzes sentence patterns, vocabulary choices, and writing style to identify markers commonly associated with AI-generated content. AI-sounding text can reduce trust and credibility with visitors.
AI SEO Readiness: Evaluates whether your content uses Schema markup, Q&A formatting, and semantic HTML that helps AI-powered search engines understand and surface your content in conversational results.
Brand Consistency: We compare the colours used across your pages and assets the way the human eye actually perceives them, not just as raw hex values. Inconsistent branding reads as careless — visitors who can't tell two pages of yours apart from each other lose confidence that you'll deliver consistently.
Call-to-Action (CTA) Analysis: Identifies and evaluates call-to-action elements based on copy urgency, visual prominence, and placement. Weak CTAs are one of the most common reasons for low conversion rates.
Clarity & Jargon: We score how accessible your writing is to a general audience and flag jargon that loses visitors. Complex language increases bounce rates. The 0-100 score reflects how each block of copy reads on its own and what specifically pulls it down — long sentences, niche terms, passive voice — so you know where to focus.
Cognitive Load: Combines visual clutter, information density, and layout complexity into a single score reflecting the mental effort required to process your page. Lower cognitive load means visitors can focus on what matters. The 0-100 score is (1 − the cognitive load index) × 100, where the index is a weighted blend of six signals of mental effort: visual clutter, choice overload, text density, semantic load, sentence complexity, and interaction effort.
Color Palette: Pulls the most-used colours from your page so you can sanity-check what visitors actually see against your stated brand palette. Surfaces sneaky third-party-widget colours, ad-hoc accents added over time, and palettes that drift across pages.
Lexical Diversity: Vocabulary richness across the page, measured in a way that fairly compares short and long pages. Higher variety keeps readers engaged and reads as more authoritative; repetitive copy reads as filler and hurts both attention and ranking signals.
Message Hierarchy: Looks at how the page sets up and delivers its main promise. Where the value statement sits (above or below where mobile visitors stop scrolling), whether the headline and supporting line agree, whether the CTA extends the headline's promise instead of reading like a generic "Submit". When any link in this chain breaks, visitors leave before the message lands.
Page Speed: Uses Google Lighthouse to measure Core Web Vitals and other performance metrics. Page speed directly impacts user experience, bounce rates, and search engine rankings.
Pricing Transparency: Detects psychological pricing tactics and evaluates how clearly pricing information is presented. Hidden or confusing pricing is one of the top reasons visitors abandon purchase decisions.
Semantic Coherence: Analyzes how well ideas connect logically using semantic similarity between adjacent sentences and paragraphs. Poor coherence means visitors get confused by disconnected or jumpy content. Each sentence becomes a vector; the 0-100 score starts at 100 and drops each time neighbouring sentences make a sharp topic jump.
Sentiment Analysis: Uses natural language processing to determine the emotional tone of your website copy. The right sentiment depends on context: consumer brands often benefit from positive tone, while B2B may prefer neutral or authoritative.
Syntactic Complexity: Analyzes grammatical dependencies between words to measure sentence complexity. Web users prefer short, simple sentences that can be quickly scanned. The 0-100 score is based on mean dependency distance (how far apart linked words sit): it is highest for short, direct sentences and falls as that distance grows, with an extra penalty when sentence length swings widely.
Technical SEO: Classic-crawler SEO checks distinct from the AI SEO Readiness card. Pulls Lighthouse's SEO category audits (meta-description, document-title, http-status-code, link-text, crawlable-anchors, is-crawlable, robots-txt, image-alt, hreflang, canonical, tap-targets, viewport, structured-data, font-size, plugins) and emits a finding per failing audit. Use this card to fix the SEO basics Google's crawler needs; use AI SEO Readiness to fix what LLM-based crawlers (ChatGPT, Perplexity, Gemini) need.
Text Overview: Reads the basic numbers about your page text (length, sentence structure, reading grade, grammar issue count, active-voice mix, repetition) and frames them against what works for your industry. A SaaS landing page reading at grade 14 is bad; a long-form media essay at grade 14 is fine. The card surfaces both the raw signals and a per-vertical interpretation so you know which fix actually applies to your kind of site.
Trust Evidence: Distinct from the Trust Signals module (which checks for privacy / security / contact existence). Trust Evidence checks the harder-to-fake signals: is there a money-back guarantee near the CTA, do the testimonials carry real names and roles, is there evidence the site is still active, are the trust claims backed by specific numbers instead of "thousands of customers", does the team page show real human faces. For regulated verticals (Finance, Healthcare), checks for the certifications regulators expect.
Trust Signals: Scans your page for credibility indicators that help visitors trust your business: SSL certificates, privacy policies, testimonials, reviews, security badges, and visible contact information.
Value Proposition: Analyzes the balance between benefit-focused language and business jargon. A strong value proposition clearly communicates unique benefits with simple language and prominent placement.
Visual Attention (Saliency): Uses AI models to predict visual attention patterns, showing where users are most likely to look. This helps ensure your most important content (CTAs, value props) is in the visual hot spots. The 0-100 score starts at 100 and loses points when key elements fall outside the predicted hotspots or when clutter spreads attention thin.

Findings

Accessible button name: Buttons that show only an icon (a magnifying glass, a hamburger menu, a heart) need a written name behind the scenes so screen-readers can announce "Search" or "Open menu". Without it, blind visitors hear "button" with no indication of what it triggers. The fix is one line of HTML (`aria-label`) per icon button.
Active voice — e-commerce: "You'll get free shipping" reads better than "Free shipping is included". Active voice puts the customer (or you) at the centre of each sentence; passive voice abstracts everyone out. For transactional sites, active is much warmer.
Audit stopped: accessibility blocker: When a page has too many basic accessibility problems (no headings, no labels, no readable contrast), the deeper checks (CTA flow, message hierarchy) would produce findings that don't apply to most of your visitors. The honest thing is to flag the accessibility blocker first. Fix the basics, then re-run and you'll get the full report.
Audit stopped: design fundamentals: Some pages have basic design issues — no clear focal point, no visual hierarchy, layout that breaks on common screens — that affect every downstream check. Rather than report findings that all roll up to "fix the design", we flag the design blocker. Address the fundamentals and re-run for the full report.
Audit stopped: navigation problem: When the page we audit has no clear way to reach your other pages — no header navigation, no footer links — we can't assess the experience your visitors actually get. Fix the navigation so visitors and search engines can find their way around, then re-run.
Audit stopped: page too slow: When a page is very slow — server response over 5 seconds, or scripts that never finish — we can't run the user-facing checks against a realistic snapshot. Speed is the blocker; fix that first. Common causes: a slow database query, an oversized hero image, or a third-party script blocking the page from finishing.
Authority signals: When ChatGPT, Perplexity, or Google's AI search recommend a business, they prefer pages that demonstrate who runs the business and why they're qualified to talk about the topic. A short "About" section with a named founder, a real photo, and a sentence about your experience moves you up in those AI answers. It also reassures human visitors that they're not buying from a faceless reseller.
Bold text used as a heading: If you make a paragraph bold and bigger to act as a section title, it looks right visually but isn't a heading underneath. Screen-readers and search engines don't see a section break — they see one long block of body text. Search engines also rely on real heading tags (H2, H3) to understand your page structure and rank you for the right queries.
Colour contrast: Low-contrast text (light grey on white, blue on dark blue) is hard for most people in bright sunlight and impossible for visitors with low vision. The WCAG accessibility standard sets a minimum contrast ratio of 4.5:1 for body text. Hitting this benchmark is also the difference between someone reading your value proposition and giving up.
CTA does not match the headline: Think of the headline as opening a question and the CTA as the answer. "Start your 14-day free trial" + a button labelled "Start free trial" closes the loop. "Start your 14-day free trial" + a button labelled "Submit" leaves the visitor wondering whether the button does what the headline implied. Generic buttons break the bridge between the promise and the next step.
Descriptive link text: When a screen-reader user pulls up a list of all links on a page, they should be able to scan and find the one they want. "Click here" is meaningless out of context; "View shipping rates" is useful. Search engines also use link text as a strong signal for what the linked page is about — vague links waste that.
Feature instead of benefit: Features describe the product. Benefits describe the customer's life after using it. Visitors don't buy features; they buy what those features do for them. The "so what?" test catches this: read each line of copy and ask "so what does this mean for me?" — the next sentence should answer. When it doesn't, you've listed a feature without explaining why anyone should care.
Generic claim: Superlatives without proof actively hurt credibility. Visitors have seen them too many times to believe them. Replace them with something concrete and specific: not "world-class support" but "average reply time: 4 minutes". Not "industry-leading" but "used by 12,000 Shopify stores". Specificity is what makes claims believable.
Grammar and spelling issues: Typos and grammar slips read as carelessness — visitors who notice often assume the product or service will be equally careless. For finance, healthcare, or professional services, the bar is near-zero. For lifestyle blogs or community content, more leeway is normal. We flag pages whose error rate sits above the typical range for your detected business category.
Heading order: Headings (H1, H2, H3...) work like a table of contents for both screen-readers and search engines. Jumping from H1 to H3 without an H2 in between confuses the outline — assistive tech can't build a proper navigation menu, and Google can't tell which sections are subsections of what. The fix is using heading levels in order, not for visual size (use CSS for size).
Headline and subhead disagree: When the H1 and the supporting line under it point in different directions, readers expend effort guessing which is the real claim. A coherent hero amplifies its own promise: the subhead extends the headline with proof or specificity, not a different topic. Example that works: H1 'Inventory management for growing brands' + subhead 'Cut stockouts by 40% with real-time SKU tracking'.
Heavy scripts: A heavy script is one that holds the phone's main thread for noticeably long after the page arrives. While that script runs, taps and swipes are queued — the page looks ready but doesn't respond. Common culprits: chat widgets, A/B-testing tools, analytics that pull in too many partner scripts.
Image alt text: Alt text is the short caption that loads when an image fails, that screen-readers read aloud to blind visitors, and that Google uses to understand what each image shows. A product photo missing alt text means a screen-reader user hears "image" instead of "hand-poured soy candle in amber jar". It also means Google Images can't surface your photo for the right searches.
Image description too long: Alt text should describe an image briefly and specifically. "Woman smiling while holding a coffee cup" is fine; a paragraph of marketing copy isn't. Long alt text turns into a wall of words a screen-reader user has to sit through before they can move on. Save the detailed explanation for the surrounding paragraph; keep alt text to one short sentence.
Jargon spikes right before the CTA: Decision moments cost mental effort. Visitors making a commitment have limited cognitive bandwidth left — jargon at this exact point forces them to re-translate at the worst possible moment, and many bail. Edit the paragraph immediately before your primary CTA: replace every acronym with its expansion, every domain term with its plain-language equivalent. The reader has just decided to act; don't make them re-translate.
JavaScript health overview: Every site ships JavaScript — code that runs in the visitor's browser. The card rolls up how much code arrived, how much actually did anything, how long the browser spent running it, and which vendors (your own scripts, third-party tags, ads, analytics) accounted for the bill. When this number is high, slow phones — exactly the ones middle-income visitors use — feel your site as sluggish.
Long tasks: Visitors don't see long tasks directly — they feel them. When the browser is busy for longer than about a tenth of a second, taps and scrolls don't register. The page looks frozen. Long tasks usually come from one heavy script doing too much at once; breaking the work into smaller pieces (your developer's job) keeps the page responsive.
Main message below the fold: On a typical phone, the "fold" is the first ~700 pixels of the page. If your main message — the one sentence that tells visitors what you do — sits below that, half of visitors leave before they ever see it. Hero images, video banners, and slow-loading sliders push your message down. Move the headline above the image, or shrink the image until both fit together.
Missing sector certification: Third-party-validated credentials beat self-declared claims by a large margin — and in regulated sectors, regulators often *require* them to be surfaced. If you hold the certification, surface it prominently: footer + about page + a dedicated "Security & Compliance" section. If you don't, that's the trust ceiling for this vertical — getting certified opens a door that copy alone can't open. In Healthcare (HIPAA) and Finance, this is a regulatory matter, not just a marketing one.
No clear value proposition: The value proposition is the one sentence your homepage hero has to nail: who you serve, what you do, what changes for them. Without it, visitors leave within the first 2 seconds — measured across hundreds of landing-page eye-tracking studies. A clear one names the audience ("founders"), the offer ("automated invoicing"), and the outcome ("so you get paid faster").
No customer or partner logos: A row of logos under "Trusted by" or "As seen in" lets visitors confirm in two seconds that you serve businesses like theirs. For B2B sites this is one of the highest-leverage trust signals. Even for small businesses, a few partner-brand or platform logos (Stripe, Shopify) signal you're set up properly.
No customer testimonials: Visitors trust other customers more than they trust your marketing copy. A handful of specific testimonials — with a name, a photo, and a concrete outcome — does more for conversion than another paragraph about you. Even better: link to an external review platform (Trustpilot, Google Reviews, G2) so visitors can verify the reviews are real.
No privacy policy link: A visible privacy policy link signals to visitors (and to Google, and to the EU's GDPR rules) that you handle personal data responsibly. When it's missing or buried, careful shoppers — exactly the ones who actually buy — pause at checkout. Place a link in the footer; it's where 90%+ of visitors look first. If you don't have a policy yet, free generators take 10 minutes.
No risk reversal: Visitors hesitate to commit when they can't see how to back out — even when the underlying decision is low-stakes. Add a single line near your primary CTA: "30-day money-back guarantee", "Cancel anytime — no questions asked", "Free 14-day trial, no credit card required". It's typically one sentence to add and is one of the most effective single edits you can make on a commitment page.
No security badges: When visitors hand over a credit card, even a small visible signal (a Stripe logo, an SSL padlock note, a "Secure checkout" line) tells them you've thought about safety. It's about reassurance, not technical security — your site might be perfectly safe already, but visitors can't tell unless you say so.
No specific contact info: Real-world existence is one of the top trust drivers a visitor checks before committing. An answered phone number, a postal address in the footer, and a named human email (sarah@yourdomain rather than info@) are worth far more than the average trust badge. For Healthcare and Local Business especially, visitors expect to see all three before they commit.
No subheadings for scanners: Most web visitors scan rather than read top-to-bottom. Without subheadings (H2s), scanners have nothing to scan — they see a wall of text and leave. A scanner should be able to read only the H2s and still get the argument. Break the body into 3-5 sections, each with a descriptive H2 (not 'Section 1', not 'About'). The H2 is the message; the body underneath is proof.
No visible return policy: Return-policy visibility is the second thing visitors look for after price. Even a single line near the buy button ("Free returns within 30 days") lifts cart conversion measurably. Hide the legalese on a linked page; surface the headline where the visitor is deciding to buy. A restrictive policy is still better than invisible — visitors who can't find a policy assume the worst.
No visible team or founder faces: A real photo on a page significantly increases trust ratings, even on product-focused sites. Add a team or about section with three or four real headshots and short bios. Real photos beat stock photos by a large margin. If the team is one person, surface the founder with name, photo, and a one-line bio. For B2B Services and Healthcare especially, visitors want to see the humans before they commit.
noscript fallback: A noscript element holds a message or content shown when JavaScript is turned off or fails to load. Its presence is worth noting because if core content or navigation only works with JavaScript, some visitors and some search crawlers may see an empty or broken page.
Pronoun balance — marketing pages: Visitors arrive thinking about themselves and their problems. Pages that lead with "we" — "we are a leading provider of..." — position the brand as the hero. Pages that lead with "you" — "you're tired of spreadsheets that break..." — position the visitor as the hero, which is the framing that converts. Aim for a 2:1 you-to-we ratio at minimum.
Reading level — B2B services: Buyers of professional services (legal, accounting, consulting, complex software) expect technical depth. Reading too simply can read as light or unprofessional. The bar is higher than e-commerce, but you still lose people above grade 14.
Reading level — e-commerce: Online shoppers scan, they don't read. Product copy that reads like a magazine article slows the purchase decision. Short sentences, concrete nouns, no jargon. Amazon product descriptions read around grade 6; that's the bar.
Reading level — local business: Visitors to local-business sites (a plumber, a dentist, a café) are looking for trust signals and basic info. Corporate or stiff language reads as inauthentic. Conversational tone with short sentences works best.
Reading level — media / editorial: Readers who arrive on an article expect to read — they came for the content. Higher reading levels are fine if the topic warrants it. The trap is making everything else on the site (navigation, calls to action) match the article's register; those still need to be simple.
Reading level — SaaS sites: B2B SaaS buyers include marketers, founders, and ops leads — not just developers. Plain language outperforms technical jargon for landing-page conversion across every study we have seen. Stripe's homepage reads at around grade 8; the Harvard Business Review at grade 9. Above grade 12 visitors stop trying.
Redundant alt text: When an image's alternative text duplicates text right next to it (the link it sits in, a caption, or a neighbouring image's alt), a screen reader announces the same words twice in a row. Either make the alt say something the surrounding text doesn't, or mark the image as decorative with an empty alt when the adjacent text already conveys it.
Redundant links: When neighbouring links share a destination, screen-reader users hear the same target announced twice in a row, which is slower and confusing. Combining them into one link, or giving each a distinct purpose, makes the page quicker to navigate.
Render-blocking resources: When your page links to a script or stylesheet in a blocking way, the browser stops drawing the page until that file arrives. On a slow phone connection, this turns into a blank white screen for several seconds. The fix is asking your developer (or platform) to load these files asynchronously, or to inline the small critical pieces directly into the page.
Sentence length — e-commerce: Shoppers scan. Long sentences force them to slow down and read carefully, which most won't do. Aim for an average of 10-16 words; reserve longer sentences for the occasional one-line emphasis.
Site looks dormant: Visitors checking whether you're still in business need to see a sign of life. A date stamp on a key page ("Updated 2026"), a single recent blog entry, or a linked changelog tells them you're still here. Even a 200-word "Q1 2026 update" is enough — recency signals are cheaper to produce than they look, and updating content frequently is one of the top 10 web-credibility drivers.
Superlatives without substantiation: Substantiated superlatives are fine — "#1 in G2 2025 enterprise category, 4.8/5 across 1,200 reviews" reads as evidence. Unsubstantiated superlatives — "world-class", "best-in-class", "industry-leading" with no proof — actively *reduce* credibility because every site says them. Either back the claim with a specific fact in the same sentence, or drop the superlative. In Finance and Healthcare there's a regulatory layer too: unsubstantiated "best" claims draw FTC scrutiny in the US.
Testimonials without attribution: "John D., happy customer" reads as fabricated. "Sarah Lin, Head of Operations, Acme Inc." reads as real. Attribution — name, role, company, ideally a photo — is what makes a testimonial believable. For higher leverage, link the same quote to an external review platform (Trustpilot, G2, Google Reviews) so visitors can verify it themselves.
Too many choices: When visitors face 8 menu options, 4 CTAs, and 12 product cards on the same screen, they freeze rather than pick. Researchers call this Hick's Law: decision time grows with the number of options. The fix is one primary action per page ("Start free trial") with secondary actions either de-emphasised or moved to a dedicated page. Counter-intuitively: fewer options usually sells more.
Unused JavaScript: Modern websites often bundle features together: a script that runs the chat widget might also include a video player nobody opens. That unused code still has to be downloaded, parsed, and held in memory — on a 4G connection it adds real seconds before the page is interactive. Code-splitting (loading only what each page needs) is the standard fix; your developer or platform should handle it.
Vague link text: Almost every page has one. "Click here" forces visitors to read the surrounding sentence before they know where the link goes; for screen-reader users, who often scan a list of all links, it's useless. Rewrite each as the destination's actual name: "Read our shipping policy" instead of "Read more about shipping".
Vague trust claims with no numbers: Precise numbers feel more credible than round numbers, even when the underlying claim is the same. "4,213 customers across 30 countries" beats "thousands of customers" on every conversion test. Pick three vague claims on the page and replace each with the specific version. "Trusted by leading brands" → "Trusted by Acme, Beta Inc., Gamma Corp, and 47 others". Writing specific copy is the work of earning trust.
Value proposition takes too long to arrive: The first sentence of body copy is the most-read sentence on the page. When that sentence is throat-clearing context ("In a world where…", "For the past decade…") instead of the value statement, you spend visitor attention on words that don't tell them why to stay. Cut the intro and move the value statement to the first sentence — context is for *after* the visitor decides to keep reading.
Visual clutter: Visual clutter is when nothing on the page is clearly the most important thing. Visitors scan, can't find a focal point, and bounce. The fix is hierarchy: one clear headline (largest), one clear primary action (high-contrast button), supporting content (smaller, lower-contrast). White space isn't waste — it's how you tell visitors what matters.
Weak value proposition: "The best way to do X" is a slogan, not a value proposition. A strong one specifies: the audience ("for freelance designers"), the offer ("a one-click invoice tool"), and the outcome ("paid 60% faster"). All three matter — slogans without specifics get ignored.
Word count — e-commerce: Too few words leaves visitors with unanswered questions; too many delays the purchase. The sweet spot covers benefits, key specs, shipping, returns, and trust signals — usually 200-500 words. Most of it can sit below an above-the-fold image and CTA.
Word count — SaaS: A SaaS buyer wants to understand what you do, who it's for, what makes you different, and how to start. That conversation fits in 300-700 words for most products. Longer is fine for complex enterprise tools; shorter is fine for clearly-defined utilities.

Performance Metrics

Core Web Vitals: Core Web Vitals are LCP (loading speed), INP (responsiveness, replacing FID), and CLS (visual stability). Google uses 75th-percentile field data from the Chrome User Experience Report (CrUX) and rolls these into search ranking. To "pass" Core Web Vitals, all three must hit the "good" threshold for 75% of real visitors over the past 28 days.
Cumulative Layout Shift (CLS): CLS quantifies how much visible content moves around unexpectedly during page load. A score below 0.1 is considered good. High CLS causes users to accidentally click the wrong thing or lose their reading position.
DOM Size: DOM Size counts the total number of HTML elements on a page. A large DOM increases memory usage, makes style calculations slower, and can cause layout reflows. Under 1,500 elements is ideal.
Domain authority (OpenPageRank): A 0–10 estimate of how much link equity your domain has accumulated. It is the closest free replacement for Google's retired PageRank toolbar, which the SEO industry used for 16 years to gauge a domain's overall link weight. A higher score means more sites link to yours, which generally correlates with stronger ranking potential in competitive search queries. The number adds context that the Lighthouse SEO score cannot: Lighthouse only checks on-page basics like meta tags and crawlability, so a perfect Lighthouse score on a low-authority domain can still struggle to rank for competitive keywords. Powered by OpenPageRank (https://www.domcop.com/openpagerank/what-is-openpagerank).
First Contentful Paint (FCP): FCP marks the point when the browser renders the first piece of content from the DOM, giving the user visual feedback that the page is loading. Under 1.8 seconds is good. A fast FCP reassures users that something is happening.
First Input Delay (FID): FID captures the time from when a user first interacts with the page (e.g., clicking a link) to when the browser begins processing that interaction. Under 100ms is considered good. FID is being replaced by INP as the primary responsiveness metric.
Interaction to Next Paint (INP): INP measures the delay between a user interaction (click, tap, key press) and the next visual update on screen. Under 200ms feels instant. Slow INP makes your site feel laggy and unresponsive.
Largest Contentful Paint (LCP): LCP measures the time from when a user starts loading the page until the largest image or text block is rendered within the viewport. A good LCP is under 2.5 seconds. It directly impacts how fast your site feels and affects search rankings.
Lighthouse Accessibility score: Lighthouse runs a subset of the axe-core ruleset (color contrast, ARIA roles, alt text, form labels, landmarks, etc.) and reports the proportion that pass. A perfect score does not guarantee WCAG conformance — automated tools can only catch around a third of accessibility issues. Manual testing with a screen reader and keyboard is still required.
Lighthouse Performance score: Computed by Google Lighthouse from a weighted blend of FCP, Speed Index, LCP, TBT, and CLS measured in a controlled lab run. Above 90 is good, 50–89 is needs improvement, below 50 is poor. Lab-based, so it doesn't always match what real users experience — pair with Core Web Vitals field data for a complete view.
Lighthouse SEO score: Checks fundamentals like having a `<title>`, meta description, valid `lang` attribute, descriptive link text, mobile-friendly viewport, and a robots-friendly meta-robots tag. It does not measure ranking factors like backlinks, content quality, or topical authority — those live in tools like Search Console, Ahrefs, or Semrush.
Speed Index: Speed Index measures how quickly content is visually displayed during page load. It captures the overall perceived loading experience by looking at visual progress over time. Under 3.4 seconds is considered good.
Time to First Byte (TTFB): TTFB measures the time from when a user navigates to a URL until the first byte of data is received by the browser. It reflects server processing time and network latency. Under 600ms is good.
Total Blocking Time (TBT): TBT measures the total time between First Contentful Paint and Time to Interactive where the main thread was blocked long enough to prevent input responsiveness. Under 200ms is good. High TBT means the page feels sluggish.

Technical Concepts

Accessibility score: This score blends the accessibility checks we run (two WCAG engines plus a live keyboard test) into one number you can track audit over audit. It is reported separately from the card's usability score and is usually lower, because it counts every WCAG issue on the page, not only the ones that also hurt conversion.
Adjacent Coherence: How smoothly ideas flow from one paragraph to the next.
Alt text: Alt text should describe the image's purpose in context, not its appearance. Decorative images use alt="" so screen readers skip them. Functional images (icons that act as links/buttons) describe the action ("Search"), not the picture ("magnifying glass"). Missing or generic alt text is one of the most common accessibility failures.
ARIA: ARIA (Accessible Rich Internet Applications) lets you communicate semantics that the browser cannot infer — for example, that a `<div>` acts as a button, or that a region is currently expanded. The first rule of ARIA is "don't use ARIA": prefer native HTML elements (button, label, nav, main) and only reach for ARIA when no native element fits.
Cognitive Load Index: Combines multiple visual analysis factors into a single number representing the mental effort required to process a page. Lower is better — aim for pages that feel clean and focused rather than overwhelming.
Color Contrast: WCAG requires a contrast ratio of at least 4.5:1 for normal text and 3:1 for large text (18px+ or 14px+ bold). Low contrast makes text hard to read, especially for people with low vision or in bright environments like outdoor mobile use.
Color Shift (ΔE): Delta E (ΔE) is a standard measure of perceptual color difference. A ΔE under 1 is imperceptible; under 3 is barely noticeable; over 5 is clearly different. Used here to check if your brand colors are being used consistently.
Error Prevention: Nielsen's fifth usability heuristic. Using HTML attributes like "required," "pattern," and "maxlength" catches input errors before submission. Inline validation that checks as the user types is even better — it prevents the frustration of submitting a form only to see errors.
Global Coherence: How well the overall message holds together across the entire page.
Heading Hierarchy: Screen readers let users jump between headings to scan page structure. If you skip from H2 to H4, that navigation breaks. Headings should reflect content structure, not visual styling — use CSS for size instead of picking heading levels for their appearance.
Incomplete audit: The Accessibility & Usability audit runs as a three-tier cascade. Tier 1 checks content structure and usability heuristics. Tier 2 checks page speed and baseline accessibility (contrast, labels, alt text). Tier 3 drives the page in a real browser to test keyboard navigation, focus traps, and dynamic states. Tier 3 is the most expensive stage, so it runs only when Tiers 1 and 2 pass. An audit is "incomplete" when Tier 3 did not run. The usual reasons: (1) the page failed an earlier gate, so the audit stopped early to surface the foundational fixes first (a deliberate fail-fast design, not a malfunction); (2) a stage hit a technical error, such as the headless browser timing out, the site blocking automated access, or a page script crashing. The findings that did run are still valid. Fixing the foundational issues and re-running usually unlocks the full audit.
Information density: Information density comes from the Shannon entropy of the text, normalised against the rough maximum for English. A low value means the writing is padded or repetitive; a very high value means it is dense and effortful to read. Most readable web copy sits in the middle of the 0-1 range.
Keyboard focus order: People who cannot use a mouse move through a page with the Tab key, and the focus order is the path they take. When it matches the visual layout (logo, navigation, main call to action, then the rest) the page is predictable; when it jumps around, keyboard and screen-reader users lose their place.
Landmark regions: Landmarks are roles like banner, navigation, main, and contentinfo attached to the major regions of a page. Screen readers list them so a visitor can skip repeated navigation and reach your main content in one step, the way a sighted visitor scans past the header.
Predictability: Predictability is derived from the compression ratio of the text. Copy that compresses well is repetitive and template-like, so it scores high; copy with varied sentence shapes and word choices compresses less and scores low. Very predictable copy reads as generic and is easy for visitors to skim past.
Reading Ease: A well-established readability score based on sentence length and how complex the words are. A score of 60-70 is the sweet spot for the web — understandable by a 13-15 year old. Higher means simpler; lower means visitors have to work harder to follow you.
Repetition: Repetition measures how often the same words and phrases recur across the analysed copy. A small amount reinforces key messages and aids recall. High values signal copy that leans on the same wording instead of varying it, which makes a page feel monotonous and can dilute the message.
Revenue impact: Revenue impact multiplies a finding's priority score by an assumed traffic volume, an average order value, and a conversion-elasticity factor (the share of conversions that friction typically costs). When your real traffic and order value are not connected to the audit, standard assumptions are used, so the dollar figure is best read as a way to compare issues against each other rather than an exact prediction. A higher number means fixing that issue is likely to return more.
Semantic Load: How much meaning and information the page tries to communicate at once.
System Feedback: Nielsen's first usability heuristic: "Visibility of System Status." When users click a button, they need confirmation the system received their action. Without feedback (spinners, progress bars, success messages), users assume the system is frozen and click again or leave.
User Control & Freedom: Nielsen's third usability heuristic. Every modal should have a close button, every multi-step form should have a back button, and every action should be undoable when possible. Without escape hatches, users feel trapped and abandon the page.
Visual Clutter: Visual clutter is the volume of competing elements (icons, badges, animations, popups) on a single screen. The fewer focal points there are, the easier visitors find what they came for. The page-level finding under the same name flags specific instances; this is the broader concept.
Visual Complexity: How visually busy the page is — more elements, colors, and movement means higher complexity.
Vocabulary richness (HD-D): HD-D (Hypergeometric Distribution D) computes, for each word in the text, the probability it appears at least once in a random 42-word sample. Averaging those probabilities gives a length-robust score that mirrors how surprising / fresh the wording feels to a reader. Most well-edited pages score 0.7-0.9; below 0.6 suggests heavy repetition.
Vocabulary variety (MATTR): MATTR (Moving-Average Type-Token Ratio) slides a fixed-size window through the text and averages the ratio of unique words to total words inside each window. The sliding window makes the score comparable across short and long pages — a plain type-token ratio drops as texts get longer just because more chances to repeat words exist. Higher values mean the writer reaches for a wider word stock; lower values indicate the same vocabulary cycling.
WCAG: WCAG is published by the W3C and structured around four principles: content must be Perceivable, Operable, Understandable, and Robust (POUR). It defines three conformance levels: A, AA, and AAA. Most laws and contracts reference WCAG 2.1 or 2.2 at level AA.
WCAG AA: WCAG (Web Content Accessibility Guidelines) defines three levels: A (minimum), AA (recommended), and AAA (enhanced). Level AA covers the most impactful requirements including color contrast, keyboard navigation, text alternatives, and form labels. It is the legal standard in many jurisdictions.
Words before repetition (MTLD): MTLD (Measure of Textual Lexical Diversity) counts how many words the writer can string together before vocabulary "exhausts" — defined as the running type-token ratio falling below ~0.72. The score is the average run length. Typical marketing copy lands between 60 and 120; technical or jargon-heavy writing often runs lower because the same domain terms recur.

Insights & Attribution

95% confidence interval: The CI is the range of plausible values for the true uplift, computed by re-running the synthetic-control fit against a placebo set (sites that did not get an audit) and reading off the 2.5th and 97.5th percentiles. If the interval includes zero, we cannot rule out "no real change" — that is reflected in the confidence tag. If the interval is entirely positive (or entirely negative) the direction of the change is statistically meaningful at the 95% level.
AI Overview exposure: We bucket each unique query in your last 90-day window into "AI-Overview-eligible" or "not", based on a rule-based heuristic: question prefixes (how, what, why, is, are, do, can, ...) and comparison shapes (" vs ", " or ", "best", "top", "review", "compare"). Sums of impressions and clicks on each bucket give the AIO impression share and the AIO vs non-AIO CTR. Google does not expose AIO eligibility per query — this is a coarse heuristic — but it tracks the surface-level signal well enough to flag where SERP-feature compression is most likely to bite.
Attribution coefficient: When you accumulate enough audits, we refit a model per audited metric (e.g., trust signals, clarity, page speed) that estimates how strongly each module's recommendations are associated with subsequent organic-search movement. Coefficients update as more audits land — they are descriptive of *your* historical pattern, not a universal claim, and are clamped to a sensible range so a single noisy audit does not dominate.
Average position (GSC): Average position is the impression-weighted mean rank across every query × URL × day in the window. It is not the position of any single result — a URL that ranked #1 on Monday and #10 on Tuesday averages to 5.5. Use position alongside clicks: a stable position with falling clicks usually means SERP features are taking the click, not that your ranking dropped.
Brand authority: We split your Search Console queries into "brand" (the query contains your domain's brand term) and "non-brand" buckets and report each bucket's share of clicks, impressions, and average position. A healthy site usually shows both: brand traffic confirms recall, non-brand traffic confirms reach. Sudden swings in the mix are often the leading indicator that something changed in the SERP.
Causal impact: Causal impact answers "did the audit move the needle?" rather than "did the metric change?" — those are not the same question. We fit a synthetic-control baseline to your pre-audit history, project it forward through the post-audit window, and report the gap between observed and projected as the audit's contribution. Confidence tags and the 95% CI describe how seriously to take the headline.
Click compression (YoY): We compare the most recent N-day window against the same window exactly one year ago, sourced from your daily Search Console totals. Impressions, clicks, and CTR deltas are reported as percentages of the prior window; the position delta is reported in absolute rank units (positive = worse — moved further from #1). The interesting story is when position is flat or improved and clicks are still down — SERP-feature compression eating clicks even though Google ranked you the same.
Confidence tags: Each metric carries a tag describing how much you should trust the headline number. "High" means the synthetic control fits your pre-audit data closely and the post-audit gap is large relative to noise. "Medium" means a meaningful gap with more uncertainty. "Low confidence" means the gap is small or the donor fit is loose. "Insufficient data" means we do not yet have enough post-audit history to compute a stable estimate. "Confounded" means we detected a parallel event (algorithm update, seasonality break, etc.) that makes attribution to your audit unsafe.
CTR compression vs non-AIO: Computed as 1 − (AIO CTR / non-AIO CTR), clamped to [0, 1]. A compression of 60% means clicks on AIO-eligible queries land at 40% of the rate they do on the rest of your queries. When this is high, no amount of on-page optimisation will lift those queries to non-AIO CTR — the SERP itself is the bottleneck. Knowing the ceiling helps decide where investment in featured-snippet content, FAQ schema, and entity markup is worth the work.
Donor pool (synthetic control): To estimate what would have happened without the audit, we build a "synthetic" version of your site from a weighted blend of similar sites that did not run an audit in the same period. The donor weights add up to 1.0 and reflect how closely each donor site tracked yours during the pre-audit window. When no donor pool is available, we fall back to a trend-and-seasonality baseline using only your own pre-audit history.
Headline derivation: Each insights card emits a structured "derivation" describing the question it answers, the answer in plain language, a severity (info / warn / critical) and a confidence tag. The page-level summary picks the top three, ranked by severity weighted by confidence, so a high-confidence "warn" outranks a low-confidence "critical." Cards that do not yet have data emit a derivation flagging that — they never silently disappear.
Organic clicks (GSC): Pulled directly from Google Search Console. Counts clicks on any URL on your verified property. Clicks are deduplicated within a session — multiple clicks from the same user on the same query within a short period collapse to one. This is the most concrete "did people actually visit" number Google publishes.
Organic CTR (GSC): CTR is clicks ÷ impressions for the time window, expressed as a percentage. There is no universal "good" CTR — it depends heavily on average position and SERP features. Compare your CTR over time and against your own non-AIO baseline rather than against industry benchmarks; the latter conflate query mix and brand strength.
Organic impressions (GSC): Pulled directly from Google Search Console. An impression is counted when a result link appears in a SERP and the user could plausibly see it (Google handles the visibility logic — pagination, AI Overview citations, etc.). Use impressions as the demand signal: if impressions are flat and clicks fall, the issue is CTR (often SERP-feature compression), not interest in your topic.
Pre / post windows: We compare two equal-length time windows: a pre-audit window ending the day before your audit ran, and a post-audit window starting the day after. Both windows are at least 28 days long when enough data is available, so day-of-week effects wash out and the trend is stable. The dates for the windows used in this report are shown in the headline impact card above.
Top movers: For each audit we rank queries by absolute change in clicks and impressions between the pre- and post-audit windows, surfacing the biggest gainers and losers. Combined with the causal-impact headline, this tells you not just whether the audit moved traffic, but which specific search queries drove the change.

Severity Levels

Critical: A serious issue that likely drives visitors away or prevents conversions. Fix first.
Critical findings — what to do: A critical finding either prevents visitors from completing a key action (broken navigation, contrast so low text is unreadable) or actively damages trust at a decision moment (no privacy policy on a checkout page). These should be addressed before any cosmetic work. Most don't require a developer — content, links, and visible trust elements are usually under your direct control.
High: A significant issue that noticeably hurts user experience. Fix soon.
High-impact findings — what to do: High-impact findings reduce how well your page performs without making it unusable. Examples: a weak value proposition, generic testimonials, slow images that delay the first paint. They're the next priority after criticals — usually a week or two of focused content/design work.
Low: A minor improvement opportunity — nice to fix but not urgent.
Medium: A moderate issue that could be improved for a better experience.

Score Bands

Critical (0–39): Significant problems that are likely hurting your business. Prioritize fixing these.
Critical score — what it means: A critical score means visitors are actively bouncing because of this module's issues. The findings list shows specifically why. Fix these first — the rest of the audit's improvements compound on top of these foundations.
Good (70–100): This area of your site is performing well. Maintain it and focus efforts elsewhere.
Good score — what it means: A good score doesn't mean perfect — it means you're in the top third for sites in your category. Keep an eye on it as you ship changes; the same content that was good a year ago can drift below the benchmark if the rest of the industry moves up.
Needs Attention (40–69): Room for improvement — addressing these issues will noticeably improve user experience.
Needs improvement — what it means: "Needs improvement" is the warning band. Visitors are leaking, but it's a slow leak — you won't see a single dramatic number drop. The work usually pays back within a quarter once you address the findings.

Glossary

Audit Modules

Findings

Performance Metrics

Technical Concepts

Insights & Attribution

Severity Levels

Score Bands

Related reading

SEO vs CRO: Which Should You Fix First?

SEO Conversion Rate Optimization Checklist

10 Reasons Your Conversion Rate Is Low