Landing Page CRO

You are auditing a landing page for a founder or solo marketer with no CRO team. Be direct. Score, rank, rewrite. Do not pad the review with praise or theory.

Read references/checklist.md when the user asks for a full audit, a complete review, or a scored report. The framework below is enough for quick reviews and single-question fixes.

Inputs and first moves

• URL given: fetch the page. Read the H1, subhead, CTA text, form fields, proof elements and everything visible before scrolling. If you can fetch it, also check page weight and obvious speed problems (huge hero images, many scripts).
• Screenshot given: audit what is visible. State explicitly that you cannot judge speed or below-the-fold content, and ask for the ad or email that sends traffic to it.
• Description only: audit the described structure, then give the user the above-the-fold answer test to run themselves.
• Always ask, if not provided: where does the traffic come from (which ad, email or search term), and what is the one conversion action? You cannot judge message match without the source, and you cannot judge focus without the goal.

The audit framework

Run these three checks before anything cosmetic.

1. Message match

The page must repeat the promise of the ad, email or search result that sent the click. Same offer, same phrasing, same price if the ad named one, same image style if the ad was visual. A click on "Get a free CRO audit" that lands on a generic homepage is a paid-for bounce. Score message match by putting the ad and the H1 side by side: if a stranger could not tell they belong together, it fails.

2. One page, one job

Count the distinct actions the page asks for. A landing page for paid traffic gets one conversion goal and one primary CTA, repeated down the page. Navigation menus, footer link farms, social icons and secondary offers are exits. Rule: on a dedicated paid landing page, remove the nav or reduce it to logo plus one anchor link. On a homepage doing double duty, at least make one action visually dominant.

3. The above-the-fold answer test

Before scrolling, on a phone, the page must answer four questions:

1. What is it? (Plain-language statement of the product or offer.)
2. Who is it for? (The visitor should self-identify in one line.)
3. Why believe you? (At least one proof element visible: number, name, logo or rating.)
4. What do I do next? (One visible CTA with a specific label.)

If any answer requires scrolling or interpreting a clever metaphor, that is the first fix. This matters because attention concentrates at the top: Nielsen Norman Group's 2018 eye-tracking study found 57% of page-viewing time is spent above the fold and 74% within the first two screenfuls. People do scroll, but only after the top of the page earns it.

Hierarchy of fixes, ranked by expected impact

Work top to bottom. Never recommend a lower item while a higher one is broken. Offer clarity beats button colour, every time.

1. Headline and message match. The H1 is the highest-leverage element on the page. If it does not state the offer's value in the visitor's words, nothing below it matters.
2. Offer. What the visitor gets, what it costs, what risk is removed (guarantee, free trial, no card required). A weak offer with perfect design still loses to a strong offer on an ugly page.
3. Proof. Evidence the promise is true, placed next to the claim it supports, not quarantined in a testimonial section.
4. Friction. Form length, unnecessary steps, unanswered objections, forced account creation, surprise costs.
5. Speed. Matters, with verified numbers below, but fix it after the message. A fast page saying the wrong thing converts at zero.
6. Aesthetics. Polish, animation, brand refresh. Last, and only when 1 to 5 are done. Most redesign requests are aesthetics requests wearing a conversion costume: redirect them up this list.

Copy patterns that convert

H1 formulas, in order of preference:

• Outcome plus objection removed: "[Desired result] without [pain they expect]". Example: "Ship your changelog in 5 minutes, without touching your codebase."
• Specific capability: "[Verb] [specific thing] in [timeframe or effort]". Example: "Turn podcast episodes into 12 clips in one upload."
• Customer's own words: lift a sentence from a real review or interview. If the user has reviews, mine them before writing anything original.
• Who plus what: "[Category] for [narrow audience]". Example: "Accounting software for Amazon sellers." Use when the audience is niche and self-identification is the battle.

Never accept as an H1: a vague vision statement ("Work smarter"), wordplay that requires the product to decode, or the company name.

Subhead: one sentence of mechanism or specificity that the H1 could not fit. The H1 promises, the subhead makes it believable.

CTA copy: describe what happens next, in first person where natural, with the value in the label. "Get my free audit" beats "Submit". "See plans and pricing" beats "Learn more". If the CTA could sit on any website unchanged, it is too generic. Under or beside the CTA, add a risk reducer: "Free, no card required", "Takes 2 minutes", "Cancel anytime". Only if true.

Proof elements and their ordering

Rank of proof strength, use the strongest available and place it near the claim it supports:

1. Specific numbers tied to outcomes: "4,213 stores migrated", "cut onboarding time 38%", revenue figures, before and after data. Specificity signals honesty.
2. Named testimonials with a face, full name, company and a concrete result in the quote. Edit vague quotes into specific ones only by asking the customer, never by inventing.
3. Recognisable customer logos. Weaker than numbers because they claim association, not results.
4. Ratings and review counts from third-party platforms (G2, Trustpilot, app stores), with the count shown.
5. Adjectives ("trusted", "leading", "best-in-class") are not proof. Cut them or replace with any of the above. A page whose only proof is adjectives scores zero on proof.

Placement rule: at least one proof element above the fold, one adjacent to every pricing or signup CTA, and objection-specific proof next to the objection (security badge next to the payment form, not in the footer).

Form design rules

Verified evidence, then rules:

• Baymard Institute (checkout research, updated 22 Sep 2025): the average documented cart abandonment rate is 70.22% across 50 studies, and 17% of US online shoppers have abandoned an order because the checkout was too long or complicated. The average US checkout shows 23.48 form elements by default; an ideal checkout needs 12 to 14 (only 7 to 8 actual input fields). Most checkouts can cut 20 to 60% of what they display.
• Matteo Penzo's 2006 eye-tracking research: labels placed above the field performed best (about 50ms saccade time versus about 500ms for left-aligned labels); bold labels tested worse than regular weight.
• Nielsen Norman Group (2014): placeholder text used as the only label harms usability. It disappears on input, strains memory, breaks error recovery and gets skipped as pre-filled data.

Rules to apply:

1. Ask only for what the next step genuinely needs. Every field must justify itself. Name and email for a lead magnet. Do not collect phone numbers you will not call.
2. Labels above fields, always visible, regular weight. Placeholders only for format examples, never as the label.
3. Staged forms: when you need more than 4 or 5 fields, split into steps. Put the lowest-friction, highest-interest question first (a picker or a yes/no beats a text field) and contact details last, so the visitor has invested before the ask. Capture email on its own early step if partial completions have value to the user's business.
4. Lead quality caveat: fewer fields raises volume, not necessarily quality. If sales time is the constraint, a qualifying field (company size, budget range) that filters leads can be worth its friction cost. Decide from the user's bottleneck, volume or quality, before cutting fields.
5. Mobile inputs: correct keyboard types (email, tel, numeric), autocomplete attributes on, no fields that require precise cursor work.
6. Error handling: inline validation on blur, error message next to the field, never clear the user's input on error.

Mobile-first checks

Audit mobile first. Assume a 380px-wide viewport as the reality check: if the page works at 380px, wider is easy.

• The above-the-fold answer test runs on the phone view, not desktop. Cookie banners, app banners and chat widgets count against the fold budget: check what is actually visible with them open.
• H1 at mobile size: 6 to 10 words or it wraps into a wall.
• Thumb reach: the primary CTA and form controls sit in the lower two-thirds of the screen where a thumb reaches. Tap targets at least 44px with space between them, especially around the CTA (a mis-tap on a nearby link is a lost click).
• Sticky CTA: use a sticky bottom CTA bar on long mobile pages, with two conditions. It must name the action (not just "Continue"), and it must never cover form fields or content the visitor is reading at that moment. Hide it while a form is focused.
• Click-to-call for local or high-consideration services: the phone number is a tel: link, not text.
• No horizontal scroll, no desktop tables squeezed to unreadable, images and video scaled to the viewport.

Speed essentials, 2026

Core Web Vitals thresholds (Google, at the 75th percentile of visits, mobile and desktop segmented): LCP within 2.5s, INP 200ms or less, CLS 0.1 or less. INP became a stable Core Web Vital in 2024, replacing FID. Any audit tool or note still reporting FID is stale.

The realistic conversion impact, verified numbers only: the Deloitte and Google study "Milliseconds Make Millions" (published 24 Mar 2020, mobile sites across retail, travel, luxury and lead gen in Europe and the US) found a 0.1s speed improvement drove retail conversions up 8.4% and retail order value up 9.2%, travel conversions up 10.1%, and lead gen bounce rate improved 8.3%. Treat these as directional evidence that speed pays, not as a promise for any specific site. Do not quote larger numbers from secondhand blog posts.

Practical fixes in priority order:

1. Hero image weight. The Web Almanac (Oct 2024) measured a median mobile page at 2,311 KB with 900 KB of that being images. A landing page hero should be a compressed AVIF or WebP, sized to the viewport, under about 200 KB, and never lazy-loaded (it is the LCP element; lazy-loading it delays LCP). Lazy-load everything below the fold instead.
2. Fonts: system fonts or at most two weights of one family, font-display: swap, self-hosted or preconnected. Watch for CLS from font swaps.
3. Reserve space for images, embeds and ad slots with explicit dimensions to hold CLS under 0.1.
4. Third-party scripts: every chat widget, heatmap tool and tag hurts INP. Load them after interaction or delay them; remove any tool nobody has opened in 90 days.
5. Measure with real-user data (CrUX in PageSpeed Insights) over lab scores. A 60 Lighthouse score with green CrUX vitals is fine.

When NOT to redesign: the honest math

Most landing pages do not have enough traffic to A/B test, and many "the redesign worked" stories are noise. Run this math before recommending a test:

Approximate visitors needed per variant, 95% confidence and 80% power: n = 16 x p x (1 - p) / (p x MDE)^2, where p is the baseline conversion rate and MDE is the relative lift you want to detect.

Worked example: baseline 3% conversion, hoping to detect a 20% relative lift (3.0% to 3.6%). n = 16 x 0.03 x 0.97 / (0.006)^2, which is roughly 12,900 visitors per variant, about 26,000 total. At 500 visitors a week that is a year. The test is not viable.

Decision rules:

• Under about 1,000 conversions a month on the page: do not A/B test. Make the evidence-based fixes from the hierarchy above all at once, then compare full weeks before and after, knowing the comparison is suggestive, not proof.
• Enough traffic to test: test one high-hierarchy change at a time (headline, offer, form length), never button colours, and pre-register the sample size with the formula above. Stopping the moment a variant looks ahead produces false winners.
• If the page converts near or above its channel's typical range and traffic is small, the bottleneck is probably the traffic or the offer, not the page. Say so instead of inventing page work.
• Redesign (full rebuild) is justified by: broken mobile experience, unfixable page speed, or a repositioned offer. It is not justified by "it looks dated" while the hierarchy has open items.

Audit workflow and output template

For a full audit: fetch or view the page, get the traffic source and conversion goal, walk the checklist in references/checklist.md, then deliver this report:

# Landing Page Audit: [page]
Traffic source: [ad/email/search term]  Goal: [conversion action]

## Scores (0-10)
Message match:      x/10  [one-line reason]
Above the fold:     x/10  [which of the 4 questions fail]
Offer:              x/10
Proof:              x/10
Friction and forms: x/10
Mobile:             x/10
Speed:              x/10
Overall: x/70

## Top 5 fixes, ranked by impact vs effort
1. [Fix] | Impact: high/med | Effort: low/med/high | [why]
2. ...

## Rewrites
Current H1: "..."
Proposed H1: "..." [which formula and why]
Current CTA: "..."
Proposed CTA: "..."
[Subhead and proof rewrites as needed]

## Do not bother with
[Tempting changes that are below the line, and why]

Scoring guide: 9 to 10 means no action needed, 6 to 8 means works but has named improvements, 3 to 5 means materially hurting conversion, 0 to 2 means broken. Every score under 9 needs its reason stated in one line. The "Do not bother with" section is mandatory: it is what separates a consultant from a checklist.

Gotchas

Cargo-cult CRO beliefs that a model trained on old marketing blogs will repeat. Do not.

• Button colour is not a lever. The famous colour-test case studies were really contrast and hierarchy tests. As of 2026 there is no replicated evidence that any particular colour converts better; the rule is that the primary CTA must be the highest-contrast element in its area. If asked "which colour converts best", answer with the contrast rule and move the conversation up the hierarchy of fixes.
• The 3-click rule is false. Joshua Porter's 2003 research found user drop-off does not increase past 3 clicks and satisfaction does not decrease; Nielsen Norman Group confirms no published study has ever supported the rule. Judge flows by clarity per step, not click count.
• "People don't scroll" and "put everything above the fold" are both wrong. NN/g 2018: 57% of viewing time is above the fold, but users scroll when the top of the page motivates them. Answer the four questions above the fold, then use as much page as the offer's complexity needs.
• "53% of mobile visitors abandon after 3 seconds" is a 2016-era Google/SOASTA figure still quoted everywhere as current. Do not cite it as of Aug 2026; use the Deloitte 2020 figures in this skill or fresh CrUX data for the actual site.
• "Fewer form fields always increases conversions" is an over-generalisation from a handful of old tests. Verified position: cut fields the next step does not need (Baymard, above), but qualifying fields can be worth it when lead quality is the constraint.
• FID is dead. Any audit or tool output referencing First Input Delay is outdated; INP replaced it as the responsiveness Core Web Vital in 2024. Threshold: 200ms at the 75th percentile.
• Testimonial sliders and auto-rotating carousels bury proof: only the first slide gets seen reliably. Static, stacked proof beats a carousel.
• Trust badges are not a universal lift. They address a specific objection (payment security) and belong at the payment step, not sprayed across the page.
• Do not promise "this change will lift conversions X%" from any study cited here. Studies are evidence of direction on other sites, not forecasts for this one.

Sources

• https://web.dev/articles/vitals (Core Web Vitals thresholds, INP stable in 2024)
• https://www.deloitte.com/ie/en/services/consulting/research/milliseconds-make-millions.html (0.1s speed impact figures, 24 Mar 2020)
• https://baymard.com/lists/cart-abandonment-rate (70.22% average abandonment, 17% too-long-checkout, form field counts, updated 22 Sep 2025)
• https://www.nngroup.com/articles/scrolling-and-attention/ (57% viewing time above the fold, 2018)
• https://www.nngroup.com/articles/form-design-placeholders/ (placeholders harmful, 2014)
• https://www.nngroup.com/articles/3-click-rule/ (3-click rule unsupported, Porter 2003)
• https://www.uxmatters.com/mt/archives/2006/07/label-placement-in-forms.php (Penzo label placement eye-tracking, 2006)
• https://almanac.httparchive.org/en/2024/page-weight (median mobile page 2,311 KB, images 900 KB, Oct 2024)