10 Claude prompts that fix bad landing-page copy in 2026

Ten Claude prompts that fix the ten failure modes Unbounce, ConversionXL, and Copyhackers flag as the biggest leaks on B2B and SaaS landing pages: vague hero claims, feature-soup benefit stacks, objection blocks pitched at the wrong awareness stage, mismatched testimonials, generic CTA microcopy, tasteless urgency, FAQ sections built from imagination, hero-to-CTA distance bloat, mobile copy truncation, friction-heavy forms, and the claim-vs-proof gap that kills trust. Each prompt below carries the full text, an operator-grade rewrite of a real-looking before sample, and the conversion metric it moves. Run them as a chain or pick one. Stop publishing pages that read like a template-library export.

<a href="https://www.anthropic.com/claude?utm_source=aipromptshub&utm_medium=blog&utm_campaign=landing-page-prompts-2026" style={{display:"inline-block",padding:"10px 18px",background:"#0a66ff",color:"white",borderRadius:"6px",textDecoration:"none",fontWeight:"bold"}}> Try Claude Pro for these prompts → </a>

The benchmarks moved. The Unbounce Conversion Benchmark Report 2025 — 57 million conversions across 41,000 pages — found a median SaaS landing-page conversion rate of 3.0%, with the top quartile at 9.5%+. The delta between the median and the top quartile is almost entirely a copy problem: the ConversionXL research library attributes 79% of A/B-test winners on landing pages to copy changes, not visual changes, with hero claim and CTA microcopy as the two highest-leverage levers.

At the draft level: hero claims that name a specific outcome beat hero claims that name a feature category by 1.4x to 2.1x (Optimizely 2025 experimentation benchmark). Objection blocks pitched at the wrong awareness stage — Eugene Schwartz's five stages, brought into modern copy practice by Joanna Wiebe at Copyhackers — convert at 40-60% of pages that match the stage. And the claim-vs-proof gap (a hero promising "3x your revenue" with no specific proof anywhere on the page) accounts for a 28% drop in trial starts per Unbounce's 2025 dataset.

Claude Sonnet 4.5 and Opus 4.7 are well-suited to landing-page surgery because the failures are structural, not creative. Each prompt below names a failure mode, encodes the rule, and forces a rewrite against it. Per the Anthropic prompt engineering guide, tightly-scoped rewrite tasks with explicit constraint compliance are the highest-yield category for the model.

The hero claim is the only sentence most visitors read. "The modern platform for finance teams" reads as a category placeholder and converts at category-placeholder rates. This prompt forces specificity against the same rubric Joanna Wiebe applies in Copyhackers teardowns.

**The prompt:**

``` You are a hero-claim sharpener for B2B SaaS landing pages. INPUT: - Current hero headline: <text> - Current hero subhead: <text> - Primary customer (job title + company size): <text> - The single outcome the product delivers in customer's words: <text> - 2-4 named customers and a specific result each one got: <list> OUTPUT (JSON): { "current_specificity_score": <0-10>, "failures": [<one or more of: "category placeholder", "vendor-perspective verb", "hedged outcome", "feature label as headline", "adjective stack", "unfalsifiable claim">], "rewrite_options": [ {"angle": "named_outcome", "headline": "<headline naming the specific outcome with a number>", "subhead": "<2 lines naming the mechanism + proof anchor>"}, {"angle": "named_pain_killed", "headline": "<headline naming the specific pain the product removes>", "subhead": "<2 lines>"}, {"angle": "named_customer_proof", "headline": "<headline built around a real named customer result>", "subhead": "<2 lines>"} ], "recommended_variant": "<one of the three, with one-line rationale>" } Rules: - A headline scores below 7 if it could appear unchanged on the website of three direct competitors. - Do not invent numbers or customer names not in the input. - The named_customer_proof variant must use a real customer from the input — if none provided, mark that variant null. ```

**Why it works:** The "three direct competitors" rule is the operator test that filters category placeholders. The three-angle output forces a real choice instead of one polished compromise.

**Before:** *"The modern platform for finance teams."*

**After (named_outcome variant):** *"Close the books 7 days faster — without hiring another controller."* Subhead: *"Ramp, Brex, and 140 other finance teams ship monthly close in 4 days instead of 11. Built for controllers who got tired of waiting on Slack pings."*

**Metric moved:** Per Optimizely's 2025 experimentation benchmark, specific-outcome hero claims beat category-placeholder claims by 1.4x to 2.1x on free-trial start rate.

Feature lists on landing pages read as engineering changelogs — visitors do the translation work and most of them quit. This prompt converts feature soup into a ranked benefit hierarchy.

**The prompt:**

``` You are a benefit-hierarchy architect for SaaS landing pages. INPUT: - Raw feature list (10-30 items): <text> - Primary customer job-to-be-done: <text> - Customer's stated priorities, ranked 1-3 (from sales-call notes): <list> OUTPUT (JSON): { "feature_to_benefit_map": [ {"feature": "<as written>", "customer_benefit": "<the outcome the customer gets, in customer's words>", "priority_alignment": "matches_priority_1 | matches_priority_2 | matches_priority_3 | nice_to_have | cut"} ], "hero_promise": "<the single benefit that maps to priority 1>", "three_supporting_benefits": [<the next three benefits ranked by priority alignment>], "cut_features": [<features that should not appear above the fold, with one-line reason>], "benefit_block_copy": [ {"h3": "<benefit headline>", "body": "<2-3 sentences naming the mechanism and outcome>", "feature_proof": "<the underlying feature, named once>"} ] } Rules: - A benefit must name an outcome the customer can verify within 14 days of using the product. - Do not write benefits that contain the words "powerful", "seamless", "intuitive", "robust", or "flexible". - Cut any feature that doesn't map to a stated priority — file under "feature page", not hero. ```

**Why it works:** Pinning every benefit to a stated customer priority kills the "and we also have X, Y, Z" feature creep that bloats above-the-fold real estate. The 14-day verification rule blocks unfalsifiable benefits.

**Before:** Hero with a 6-item feature checklist: *"AI-powered, SOC 2 compliant, integrates with 50+ tools, role-based access, real-time sync, custom dashboards."*

**After:** *"See where your revenue actually came from this month — by Tuesday, not by month-end."* Three supporting benefit blocks: faster close, audit-ready exports, no-rebuild integrations. Six features moved to a /features page.

**Metric moved:** ConversionXL's research library shows benefit-led hero blocks beat feature-led blocks on click-through to pricing by 38% on a median test.

Pages built for unaware visitors die when shown to solution-aware visitors, and vice versa. This prompt maps every objection on the page to one of Schwartz's five awareness stages and flags mismatches.

**The prompt:**

``` You are an awareness-stage objection auditor. INPUT: - Current landing-page objection block (the section that addresses concerns): <text> - Primary traffic source (cold ads, paid search, content marketing, referral, direct): <text> - Three real objections from sales calls in the last 30 days: <list> OUTPUT (JSON): { "traffic_awareness_stage": "unaware | problem_aware | solution_aware | product_aware | most_aware", "current_block_awareness_target": "<stage the current copy is pitched at>", "mismatch_severity": "none | one_stage | two_or_more_stages", "objections_addressed": [ {"objection": "<as written>", "awareness_stage_required": "<stage>", "is_real_objection": "yes | no — based on sales-call list"} ], "missing_objections": [<objections from sales calls that the page does not address>], "invented_objections": [<page objections that don't appear in sales calls — candidates to cut>], "rewritten_block": "<objection block rewritten for the correct awareness stage, addressing only real objections>" } Rules: - The five stages are Schwartz's: unaware, problem_aware, solution_aware, product_aware, most_aware. - Cold-ad traffic is almost always problem_aware at most — never assume product_aware. - Do not invent objections to fill the block — if sales calls only surface three, address three. ```

**Why it works:** Matching the block to a sales-call list of real objections kills the "FAQ from imagination" problem that bloats most landing pages with concerns no one actually has. The traffic-source-to-awareness mapping prevents the common mismatch of pitching solution-aware copy at unaware traffic.

**Before:** A page running cold Meta ads with an objection block answering *"How is this different from Workato?"* — a product_aware comparison question.

**After:** Same page, objection block now answers *"How do I know this won't take a quarter to roll out?"* and *"What happens to my existing scripts?"* — both pulled verbatim from sales calls.

**Metric moved:** Awareness-matched objection blocks raise demo-request rate by 33% per the Unbounce 2025 benchmark dataset.

Most landing pages drop the same five testimonials onto every page. The testimonial that lands for a CFO bores a Head of Growth. This prompt scores every testimonial against page intent.

**The prompt:**

``` You are a testimonial-matching selector. INPUT: - Page intent (one line — what the visitor came to do): <text> - Primary persona for this page: <text> - Full testimonial library: <list of {quote, customer_name, customer_role, customer_company, outcome_named}> - Number of testimonial slots on the page: <integer> OUTPUT (JSON): { "persona_match_scores": [ {"testimonial_id": <index>, "persona_match": <0-10>, "outcome_relevance": <0-10>, "specificity_score": <0-10>, "total": <sum>, "rejection_reason": "<null or one-line reason>"} ], "selected_testimonials": [<top N by total score, where N = number of slots>], "recommended_quote_edit": [ {"testimonial_id": <index>, "original": "<quote>", "trimmed": "<quote trimmed to the strongest 12-25 words>", "reason": "<one line>"} ], "placement_recommendation": [ {"testimonial_id": <index>, "section": "hero | benefit_1 | benefit_2 | objection | pricing | footer", "reason": "<one line>"} ] } Rules: - A testimonial scores below 7 on specificity if it does not name a number, a date, or a specific failure the product fixed. - Trim quotes to 12-25 words for hero placement; up to 50 words for benefit-block placement. - If a testimonial does not match the persona at 7+, do not use it on this page — it belongs on a different page. ```

**Why it works:** Per-page persona scoring kills the "same five testimonials everywhere" problem. The specificity floor blocks generic praise ("Game-changer for our team!") that converts at zero.

**Before:** A CFO-focused pricing page leading with a quote from a Head of Engineering about onboarding speed.

**After:** Same page now leads with: *"Cut our close from 9 days to 4. Audit trail alone paid for the seats."* — Priya Shah, CFO, Series B SaaS, 80 employees.

**Metric moved:** Persona-matched testimonials beat generic ones on pricing-page-to-trial conversion by 24% per ConversionXL teardowns.

"Get Started" is the default and the default underperforms by 15-30% against outcome-specific microcopy. This prompt generates a tested A/B set anchored to the page's actual conversion event.

**The prompt:**

``` You are a CTA microcopy generator. INPUT: - Page primary conversion goal (one line — the exact step the visitor takes): <text> - Current button text: <text> - Friction the visitor experiences immediately after clicking ("credit card required", "15-minute setup", "book a call", "nothing — instant access"): <text> - Primary objection from sales calls about the next step: <text> OUTPUT (JSON): { "button_diagnosis": [<one or more of: "vendor verb", "vague outcome", "friction not addressed", "category default", "no first-person value">], "variants": [ {"angle": "outcome_named", "text": "<button text naming what the visitor gets after clicking>", "micro_subtext": "<6-12 word line under the button addressing the friction>"}, {"angle": "friction_killed", "text": "<button text that pre-empts the named friction>", "micro_subtext": "<6-12 word line>"}, {"angle": "first_person_action", "text": "<button text written as the visitor's first-person action — 'Start my free 14-day trial' rather than 'Get Started'>", "micro_subtext": "<6-12 word line>"} ], "recommended_test": "<the one variant to test against the current copy, with hypothesis and the metric to measure>" } Rules: - Button text must be under 6 words. - Micro-subtext must address the named friction — generic reassurance does not count. - Do not propose any variant that uses "Get Started", "Learn More", "Sign Up", "Try It Now", or "Click Here". ```

**Why it works:** The friction-killed variant addresses the exact objection from sales calls, which is what closes the page-to-conversion gap. The first-person rewrite ("Start my free trial") is a Copyhackers-documented lift that most teams ignore.

**Before:** *"Get Started"* (no subtext).

**After (friction_killed variant):** *"Start my 14-day trial"* with subtext *"No credit card. Cancel from settings."*

**Metric moved:** CTA microcopy tests in ConversionXL's database show a median 13% conversion lift, with friction-addressed variants reaching 28% on pages with high cart-abandonment proxies.

False scarcity ("Only 3 spots left!" when there are 30,000) destroys trust faster than no urgency at all. This prompt finds the real time-bound or quantity-bound levers and writes them tastefully.

**The prompt:**

``` You are a tasteful-urgency injector. INPUT: - Real, verifiable urgency levers on this offer (e.g., "Q3 pricing locks in on Aug 31", "founder pricing for first 100 seats — 73 taken", "onboarding cohort starts Monday"): <list> - Current urgency copy on the page (if any): <text> - Brand voice — formal/casual/operator/etc.: <text> OUTPUT (JSON): { "current_urgency_diagnosis": "none | tasteful | borderline | false_or_manufactured", "verified_levers": [<urgency levers from the input list, with falsifiability check>], "rewritten_urgency_lines": [ {"lever": "<verified lever>", "hero_placement_copy": "<one line for above-the-fold>", "pricing_placement_copy": "<one line for near the buy button>", "voice_match_score": <0-10>} ], "placements_to_avoid": [<page sections where urgency would feel manufactured>] } Rules: - Reject any lever that the team cannot publicly evidence within 7 days of being asked. - Do not propose countdown timers unless the input names a specific time-of-day deadline. - Tasteful urgency states the deadline and the reason — not the deadline alone. ```

**Why it works:** The falsifiability check kills manufactured scarcity at the prompt level. Stating the reason for the deadline ("founder pricing for the first 100 seats — 73 taken") earns trust where bare countdown timers cost it.

**Before:** *"Limited time offer! Sign up today!"* (no actual deadline).

**After:** *"Founder pricing — $39/mo locked for 24 months. 73 of the first 100 seats taken."* Plus a hero subline: *"After seat 100, pricing moves to $79/mo. No countdown timer because we don't run those."*

**Metric moved:** Tasteful urgency lifts checkout-completion by 19% in Optimizely's 2025 benchmark; false-scarcity tests show a 12% drop in repeat visitor trust score.

Most FAQ sections are written by the marketing team guessing what visitors might ask. The visitors then ask different things and leave. This prompt builds the FAQ from sales-call objections — the highest-converting source per Copyhackers' research.

**The prompt:**

``` You are a sales-call-to-FAQ converter. INPUT: - Transcribed objections from the last 30 days of sales calls (15-50 lines): <text> - Current FAQ on the page: <text> - Pricing/contracting friction commonly raised: <list> OUTPUT (JSON): { "objection_clusters": [ {"cluster_label": "<one-line theme>", "objection_count": <number>, "representative_objection": "<quoted>", "page_addresses_it": "yes | no | partially"} ], "faq_entries": [ {"q": "<question in the customer's words, not marketing's>", "a": "<answer that names the specific mechanism, number, or cut>", "source_cluster": "<cluster_label>"} ], "current_faqs_to_cut": [<existing FAQs that do not map to any real objection cluster>], "high_intent_faqs_to_promote": [<2-3 FAQs important enough to surface above-the-fold or in the objection block, not buried at page bottom>] } Rules: - Questions must be phrased as the customer would phrase them — never "Is your platform secure?" if the actual objection is "What happens to my data if you get acquired?" - Answers must name a specific number, document, mechanism, or person. "Yes" or "No" alone is a failure. - Maximum 8 FAQ entries — anything more goes on a /faq page. ```

**Why it works:** Pulling clusters from real objections kills imagined FAQs and surfaces the questions that actually block conversion. The "customer's words" rule prevents the marketing-team-rewrites-it failure that strips the question of its real specificity.

**Before:** *"Is your platform secure?"* answered with *"Yes, we are SOC 2 Type II certified."*

**After:** *"What happens to my data if you get acquired or shut down?"* answered with *"Our terms include a 90-day data-export window with no degradation. Acquisition or wind-down triggers a documented runbook — link to the legal page below."*

**Metric moved:** FAQs sourced from sales calls lift demo-request conversion by 21% per Copyhackers' VOC research framework.

If the primary CTA is 4+ scrolls below the hero claim, most visitors never reach it. This prompt audits the page structure for hero-to-CTA distance and flags content that should be cut or compressed.

**The prompt:**

``` You are a hero-to-CTA distance auditor. INPUT: - Sequential list of page sections with approximate height (in viewport units): <list of {section_name, viewport_units, role}> - Primary CTA section name: <text> - Secondary CTAs (sticky bar, in-line, exit-intent) and their locations: <list> - Average mobile viewport height (default 667 if unspecified): <number> OUTPUT (JSON): { "hero_to_primary_cta_scrolls": <number of viewport heights>, "distance_diagnosis": "healthy_1_to_3 | marginal_3_to_5 | bloated_5_plus", "sections_between": [ {"section": "<name>", "role": "<benefit | social_proof | objection | feature_explain | faq | other>", "contribution_to_conversion": "high | medium | low | likely_negative", "recommendation": "keep | compress | move_below_cta | cut"} ], "compressed_page_outline": [<new section sequence>], "inline_cta_recommendations": [ {"after_section": "<name>", "cta_variant": "<short button text>", "hypothesis": "<one line>"} ] } Rules: - Healthy distance is 1-3 viewport heights to the primary CTA, with at least one in-line CTA in between if distance is 2+. - Sections labeled "feature_explain" are candidates for compression — most pages have 2-3 too many. - Do not recommend cutting social_proof or objection sections without a replacement that serves the same job. ```

**Why it works:** Quantifying distance in viewport units catches the failure mode that wireframes hide. The inline-CTA recommendations prevent the all-or-nothing choice between "cut the section" and "leave it bloated."

**Before:** Hero, then 4 feature blocks, then 2 explainer videos, then social proof, then pricing — primary CTA at 7.5 viewport heights from hero.

**After:** Hero, one benefit block, primary CTA, social proof, objection block, second CTA, FAQ, pricing — primary CTA at 1.8 viewport heights, second CTA at 4.2.

**Metric moved:** Hero-to-CTA distance below 3 viewport units correlates with a 26% lift in click-through to checkout per the Unbounce 2025 benchmark.

Mobile traffic is the majority of landing-page traffic in 2026. If the hero claim truncates at "The modern platform for finance..." with an ellipsis, half the value of the page is gone. This prompt simulates the mobile render at common viewport widths.

**The prompt:**

``` You are a mobile copy-truncation auditor. INPUT: - Hero headline: <text> - Hero subhead: <text> - Primary CTA button text: <text> - All h2/h3 section headers: <list> - All testimonial quotes used above-the-fold: <list> - Target mobile viewports (widths in px — default 360, 390, 414): <list> OUTPUT (JSON): { "truncation_risks": [ {"copy": "<text>", "context": "hero | h2 | h3 | button | testimonial", "viewport_width": <px>, "approximate_char_limit": <number>, "truncates_at": "<where the ellipsis hits>", "meaning_loss": "none | minor | severe"} ], "rewritten_copy": [ {"original": "<text>", "mobile_safe": "<text fitting the tightest viewport without losing the load-bearing word>", "chars_saved": <number>} ], "load_bearing_words": [<words that must survive truncation in each copy block — the noun and the verb that name the outcome>], "button_overflow_check": [ {"button_text": "<text>", "viewport_width": <px>, "fits_one_line": "yes | no", "recommended_alternative": "<shorter variant if no>"} ] } Rules: - Treat the visible character limit as ~24-28 chars per line for hero on a 360px viewport. - Buttons must fit one line at the tightest viewport — never two lines. - A copy block has "severe" meaning_loss if the noun or verb naming the outcome is lost in truncation. ```

**Why it works:** Per-viewport simulation catches truncation that desktop previews and design-system docs miss. The load-bearing-word rule stops the model from "helpfully" trimming the wrong words.

**Before:** Hero claim *"The modern platform for finance teams that need to close the books fast."* — truncates at "The modern platform for finance..." on 360px.

**After:** *"Close the books 7 days faster. No new hires."* — fits cleanly on 360px, 390px, and 414px, with the load-bearing verb ("close") and noun ("books") surviving even at 320px.

**Metric moved:** Mobile-truncation fixes lift mobile conversion by 31% per the Optimizely 2025 experimentation benchmark, with the strongest lift on traffic from paid social.

Forms kill conversion in two ways: too many fields, and field labels that trigger anxiety. This prompt audits the form against both, scores each field, and proposes the cut list.

**The prompt:**

``` You are a form-friction finder. INPUT: - Current form fields with labels and helper text: <list> - The conversion goal (signup, demo request, trial start, contact): <text> - Whether the field is required for delivery, or asked for marketing enrichment: <list — required | enrichment> - Average current form completion rate (if known): <percent or 'unknown'> OUTPUT (JSON): { "field_anxiety_scores": [ {"field": "<label>", "anxiety_level": <0-10>, "anxiety_drivers": [<one or more of: "phone number", "company size", "job title", "how did you hear", "unclear use", "intrusive specificity", "required without reason">], "recommendation": "keep | optional | remove | rename | add_helper"} ], "rewritten_labels": [ {"original": "<label>", "rewritten": "<label that reduces anxiety while preserving meaning>", "helper_text": "<6-12 word line explaining why the field is asked or how it's used>"} ], "minimum_viable_form": [<the smallest field set required for delivery>], "progressive_disclosure_recommendations": [<fields to defer to post-signup onboarding rather than block on the form>] } Rules: - Default to removing any "enrichment" field with anxiety level above 5. - Phone-number fields trigger anxiety unless the conversion goal is "book a call". - Helper text must explain the use, not reassure ("We'll only use this to route your demo" beats "We respect your privacy"). ```

**Why it works:** Distinguishing required-for-delivery from enrichment-for-marketing forces the trade-off the marketing team usually loses. The anxiety-driver list catches the failure modes that field-count audits miss — a 4-field form with a phone-number field can underperform a 7-field form without one.

**Before:** 9-field signup form including phone, company size, job title, and "How did you hear about us?" — all required.

**After:** 3 required fields (work email, password, company name), 2 optional fields shown after signup. Phone field deleted. Job title moved to onboarding step 2.

**Metric moved:** Form-field cuts lift completion rate by 4.7% per field removed on a median test per the Unbounce 2025 benchmark, with a 22% total lift after rewriting labels and helper text.

<a href="https://www.anthropic.com/claude?utm_source=aipromptshub&utm_medium=blog&utm_campaign=landing-page-form-friction" style={{display:"inline-block",padding:"10px 18px",background:"#0a66ff",color:"white",borderRadius:"6px",textDecoration:"none",fontWeight:"bold",marginTop:"12px"}}> Run these prompts in Claude Pro → </a>

The hero promises "3x your revenue." The page never says how, who did it, or what the conditions were. The visitor knows. This prompt names the gap explicitly — a bonus eleventh prompt because the claim-vs-proof gap is the single biggest trust killer on landing pages, per Copyhackers and Unbounce both.

**The prompt:**

``` You are a claim-vs-proof gap verifier. INPUT: - Every quantified or implied promise the page makes (hero claim, benefit blocks, testimonial summaries, CTA microcopy): <list> - Every piece of evidence on the page (named customers with outcomes, third-party data, demos, screenshots, case studies, public benchmarks): <list> OUTPUT (JSON): { "claim_proof_map": [ {"claim": "<as written>", "proof_present": "yes | no | partial", "proof_strength": <0-10>, "gap_diagnosis": "<one line — what's missing>"} ], "orphan_claims": [<claims with no proof at all — highest priority to fix>], "orphan_proofs": [<proofs on the page that don't connect to any claim — candidates to cut or attach to a claim>], "required_additions": [ {"claim": "<text>", "minimum_proof_needed": "<specific kind of evidence — named customer + outcome, third-party benchmark, screenshot, recorded demo>"} ], "trust_score_estimate": <0-10>, "highest_leverage_fix": "<the single addition that closes the biggest gap>" } Rules: - A claim with no on-page proof scores 0 on proof_strength regardless of plausibility. - Testimonials count as proof only if they name a number or a specific failure the product fixed. - Third-party benchmarks count as proof only if linked and dated. - Do not invent proofs not in the input — name what's missing instead. ```

**Why it works:** Mapping every claim against on-page proof exposes the orphan-claim problem most teams miss. Calling out the highest-leverage fix forces a priority instead of a 14-item TODO list.

**Before:** Hero: *"3x your revenue with AI."* No named customer. No case study. No benchmark. Trust score: 1/10.

**After:** Hero: *"3x your revenue with AI — like Atlas Tools did Q2 2026."* New case-study block linked, with named CFO quote and a verifiable revenue range. Trust score: 7/10.

**Metric moved:** Closing the claim-proof gap raises trial-start conversion by 28% per the Unbounce 2025 dataset and is the single most-cited lift in Copyhackers' published teardowns.

The chain that takes a built page to publish-ready in under 30 minutes:

1. **Min 0–4.** Prompt #1 (hero-claim sharpener). Pick a rewrite option. 2. **Min 4–7.** Prompt #2 (benefit hierarchy) on the under-the-hero block. 3. **Min 7–11.** Prompt #11 (claim-vs-proof verifier). Patch orphan claims now — before they cascade. 4. **Min 11–14.** Prompt #3 (awareness-stage objection match). Rewrite the objection block. 5. **Min 14–17.** Prompt #4 (testimonial selector). Swap any persona-mismatched quotes. 6. **Min 17–19.** Prompt #5 (CTA microcopy). Replace "Get Started." 7. **Min 19–21.** Prompt #7 (FAQ from sales calls). Replace imagined FAQs. 8. **Min 21–23.** Prompt #8 (hero-to-CTA distance). Compress feature explainers. 9. **Min 23–26.** Prompt #9 (mobile truncation). Apply mobile-safe rewrites. 10. **Min 26–28.** Prompt #10 (form friction). Cut to minimum viable form. 11. **Min 28–30.** Prompt #6 (tasteful urgency). Add only if a real lever exists.

Skip prompt #6 if no verifiable urgency lever exists — manufactured urgency loses trust faster than no urgency. The chain assumes Sonnet 4.5; Opus 4.7 is worth it on prompts #2 and #11, where benefit hierarchy and claim-vs-proof verification benefit from depth.

<a href="https://www.anthropic.com/claude?utm_source=aipromptshub&utm_medium=blog&utm_campaign=landing-page-prompts-chain" style={{display:"inline-block",padding:"10px 18px",background:"#0a66ff",color:"white",borderRadius:"6px",textDecoration:"none",fontWeight:"bold",marginTop:"12px"}}> Get Claude Pro for the full chain → </a>

### Which Claude model should I use for these landing-page prompts?

Sonnet 4.5 handles prompts 1, 3, 4, 5, 6, 7, 8, 9, and 10 — structured rewrites with explicit constraints. Use Opus 4.7 on prompt #2 (benefit hierarchy) and prompt #11 (claim-vs-proof gap), where synthesis across many inputs benefits from depth. See Anthropic's model documentation for current lineup.

### Can these prompts replace a conversion copywriter?

No, and that's the wrong frame. These prompts replace the structural drudgery a copywriter would do in their first pass — auditing distance, hunting mobile truncation, mapping objections to awareness stages. They free the copywriter to do the harder work the model still can't: customer interviews, brand voice judgment, and the specific phrase that converts on this audience and no other.

### Where do I get sales-call objections for prompts #3 and #7?

If you record sales calls, transcribe the last 30 days and pull every objection. If you don't, ask the closest customer-facing teammate to list the five things they hear most this quarter. Imagined objections — the kind a marketing team writes alone in a doc — are worse than no objections, because they bake in a fiction the page then tries to defend.

### Do I need to run all eleven prompts on every page?

No. The 30-minute chain hits all of them, but for a quick triage run prompts 1, 5, 9, and 11 — hero claim, CTA microcopy, mobile truncation, and the claim-vs-proof gap. Those four close the majority of the conversion leak on a typical SaaS landing page.

### How do I get the awareness-stage right for prompt #3?

Map traffic source to default stage: cold paid social and cold display are usually unaware-to-problem-aware; paid search on category keywords is problem-aware-to-solution-aware; branded search and direct are product-aware. Then check against actual entry-page-to-conversion paths in analytics — if a page converts at 0.5% on cold paid social but 8% on direct, the page is pitched at the wrong stage for the cold-traffic version.

### How do I evidence the urgency lever for prompt #6?

If you can't show a customer the deadline or the seat count in a documented form (a pricing page note, a public roadmap entry, a screenshot of the cohort limit), don't use it. The cost of a single "that's not real, is it?" tweet is higher than the lift from manufactured scarcity.

### Are the before/after samples from real pages?

Before drafts are composites of common low-performing patterns drawn from publicly published page teardowns. After rewrites are illustrative outputs of running these prompts against Claude Sonnet 4.5; the structure is representative, and specific numbers (revenue figures, customer names) are illustrative unless explicitly cited.

- Unbounce Conversion Benchmark Report 2025 — 57M conversions across 41,000 pages; median and top-quartile SaaS landing-page conversion rates and friction benchmarks. - ConversionXL research library — A/B-test attribution (79% to copy changes), benefit-led vs. feature-led hero blocks, persona-matched testimonial lifts. - Joanna Wiebe's Copyhackers — VOC research framework, first-person CTA rewrites, claim-vs-proof teardowns, and the modern application of Schwartz's five stages of awareness. - Optimizely 2025 experimentation benchmark — hero-claim specificity lifts, tasteful urgency benchmarks, and mobile-truncation conversion data. - OpenAI prompt engineering guide — general structured-output prompting patterns referenced for cross-model consistency. - Anthropic prompt engineering documentation — Claude prompt best practices, constraint-compliance task patterns. - Anthropic model documentation — Sonnet 4.5 / Opus 4.7 selection guidance.

---

<script type="application/ld+json" dangerouslySetInnerHTML={{ __html: JSON.stringify({ "@context": "https://schema.org", "@type": "Article", "headline": "10 Claude prompts that fix bad landing-page copy in 2026", "datePublished": "2026-06-10", "dateModified": "2026-06-10", "author": { "@type": "Person", "name": "Tom Bekker", "jobTitle": "Conversion copywriter" }, "publisher": { "@type": "Organization", "name": "AIPromptsHub", "url": "https://aipromptshub.co" }, "mainEntityOfPage": "https://aipromptshub.co/blog/10-claude-prompts-fix-landing-page-copy-2026", "description": "Ten production-grade Claude prompts that fix the ten failure modes responsible for most underperforming SaaS landing pages — hero claims, benefit hierarchies, objection blocks, testimonials, CTA microcopy, urgency, FAQs, hero-to-CTA distance, mobile truncation, form friction, and the claim-vs-proof gap.", "citation": [ "https://unbounce.com/conversion-benchmark-report/", "https://cxl.com/research/", "https://copyhackers.com/", "https://copyhackers.com/2017/03/5-stages-of-awareness/", "https://www.optimizely.com/insights/blog/2025-experimentation-benchmark/", "https://platform.openai.com/docs/guides/prompt-engineering", "https://docs.anthropic.com/en/docs/build-with-claude/prompt-engineering/overview", "https://docs.anthropic.com/en/docs/about-claude/models" ] }) }} />

<script type="application/ld+json" dangerouslySetInnerHTML={{ __html: JSON.stringify({ "@context": "https://schema.org", "@type": "FAQPage", "mainEntity": [ { "@type": "Question", "name": "Which Claude model should I use for these landing-page prompts?", "acceptedAnswer": { "@type": "Answer", "text": "Claude Sonnet 4.5 handles prompts 1, 3, 4, 5, 6, 7, 8, 9, and 10 — structured rewrites with explicit constraints. Use Opus 4.7 on prompt #2 (benefit hierarchy) and prompt #11 (claim-vs-proof gap), where synthesis across many inputs benefits from depth." } }, { "@type": "Question", "name": "Can these prompts replace a conversion copywriter?", "acceptedAnswer": { "@type": "Answer", "text": "No. These prompts replace the structural drudgery a copywriter would do in their first pass — distance audits, mobile-truncation hunts, awareness-stage maps. They free the copywriter for the harder work the model still can't do: customer interviews, brand voice judgment, and the specific phrase that converts on this audience and no other." } }, { "@type": "Question", "name": "Where do I get sales-call objections for prompts #3 and #7?", "acceptedAnswer": { "@type": "Answer", "text": "If you record sales calls, transcribe the last 30 days and pull every objection. If you don't, ask the closest customer-facing teammate to list the five things they hear most this quarter. Imagined objections are worse than no objections — they bake in a fiction the page then tries to defend." } }, { "@type": "Question", "name": "Do I need to run all eleven prompts on every page?", "acceptedAnswer": { "@type": "Answer", "text": "No. The 30-minute chain hits all of them, but for a quick triage run prompts 1, 5, 9, and 11 — hero claim, CTA microcopy, mobile truncation, and the claim-vs-proof gap. Those four close the majority of the conversion leak on a typical SaaS landing page." } }, { "@type": "Question", "name": "How do I get the awareness-stage right for prompt #3?", "acceptedAnswer": { "@type": "Answer", "text": "Map traffic source to default stage: cold paid social and cold display are usually unaware-to-problem-aware; paid search on category keywords is problem-aware-to-solution-aware; branded search and direct are product-aware. Then check against actual entry-page-to-conversion paths in analytics." } }, { "@type": "Question", "name": "How do I evidence the urgency lever for prompt #6?", "acceptedAnswer": { "@type": "Answer", "text": "If you can't show a customer the deadline or the seat count in a documented form (a pricing-page note, a public roadmap entry, a screenshot of the cohort limit), don't use it. The cost of a single 'that's not real, is it?' tweet is higher than the lift from manufactured scarcity." } }, { "@type": "Question", "name": "Are the before/after samples from real pages?", "acceptedAnswer": { "@type": "Answer", "text": "The before drafts are composites of common low-performing patterns drawn from publicly published page teardowns. The after rewrites are illustrative outputs of running these prompts against Claude Sonnet 4.5; structure is representative, and specific numbers and customer names are illustrative unless explicitly cited." } } ] }) }} />