How to Use AI to Compare Two Landing Page Drafts Against the Same Brief
Summary
- Use AI as a judge against the brief: define the brief as a scoring rubric, then evaluate each draft against the same criteria.
- Feed the model the brief, both drafts, and a strict output structure so you get comparable, decision-ready results.
- Ask for evidence-based notes: require quoted excerpts from each draft to justify every score and recommendation.
- Run review checks for compliance, clarity, message match, and conversion logic before you pick a winner or merge drafts.
- Finish with a concrete deliverable: a side-by-side scorecard, a recommended “best-of” merged draft, and a change list per draft.
If you have two landing page drafts written from the same brief, AI can help you compare them in a way that is faster and more consistent than “gut feel.” The key is to treat the brief as the source of truth, force the model to use the same rubric for both drafts, and require it to cite evidence from each draft (quotes) so you can verify the judgment.
This guide shows a practical workflow: what to paste in, what to ask for, what the output should look like, and the checks to run before you ship changes. All examples below are invented and labeled.
What you are trying to produce (the deliverable)
A useful AI-assisted comparison is not “Draft A is better.” It is a structured decision artifact you can share with stakeholders and act on. Aim for these outputs:
- Side-by-side scorecard against the brief (same criteria, same scale).
- Evidence notes with short quotes from each draft supporting each score.
- Winner by section (hero, benefits, social proof, FAQ, CTA) so you can merge intelligently.
- Risk and compliance flags (claims, guarantees, missing disclaimers, ambiguity).
- Action list: exact edits to make Draft A and Draft B meet the brief.
Step 1: Turn the brief into a scoring rubric (so AI cannot “move the goalposts”)
Before you paste anything into an AI tool, rewrite your brief into a rubric with explicit criteria and weights. This is what keeps the comparison fair.
Rubric template (copy/paste)
- Audience match (0-5): Does the language reflect the target persona’s context and sophistication?
- Value proposition clarity (0-5): Is the “what + who + outcome” clear in the first screen?
- Message consistency with brief (0-5): Does it stay within the promised positioning and scope?
- Proof and credibility (0-5): Are proof points present and believable (without overclaiming)?
- Conversion structure (0-5): Does it guide scanning, objections, and CTA progression?
- Voice and tone (0-5): Does it match the brand voice requirements?
- Compliance / claim safety (0-5): Are claims appropriately qualified and non-misleading?
- Readability and specificity (0-5): Concrete benefits, minimal fluff, clear nouns/verbs.
If your brief includes non-negotiables (required phrases, required sections, forbidden claims, legal disclaimers), list them separately as pass/fail gates. AI should fail a draft that violates a gate even if it “reads well.”
Step 2: Provide the source material in a controlled input format
AI comparisons go off-track when the model has to guess what matters. Use a consistent input block so the model can parse the brief and both drafts reliably.
Input structure (copy/paste)
- BRIEF: Target audience, offer, primary promise, differentiators, objections, tone, required sections, forbidden claims, CTA goal.
- RUBRIC: Criteria + weights + pass/fail gates.
- DRAFT A: Full text.
- DRAFT B: Full text.
- CONSTRAINTS: Output format, quote requirements, and “no rewriting until after scoring.”
Invented example input (labeled)
Invented example: Below is a fictional brief and two short draft excerpts to demonstrate the method. Replace with your real materials.
BRIEF (Invented)
Product: “Northwind Payroll” (B2B SaaS)
Audience: US-based HR managers at 50-300 employee companies switching from spreadsheets or legacy payroll
Primary promise: Run payroll in under 15 minutes with fewer errors
Differentiators: Guided setup, built-in compliance checks, human support during first 2 payroll runs
Objections: “Migration will be painful,” “Compliance risk,” “Support won’t respond”
Tone: Confident, plainspoken, no hype
Required sections: Hero, 3 benefit bullets, “How it works” (3 steps), proof block, FAQ (3 Qs), final CTA
Forbidden: “Guaranteed,” “100% error-free,” “always compliant”
CTA goal: Book a demo
RUBRIC (Invented)
Audience match (x1), Value prop clarity (x2), Message consistency (x2), Proof (x2), Conversion structure (x1), Voice (x1), Compliance safety (x2), Readability (x1)
Pass/fail gates: Must include “Book a demo” CTA; must include “How it works” 3 steps; must not use forbidden claims.
DRAFT A (Invented excerpt)
“Run payroll in minutes, without the headaches. Northwind Payroll helps HR teams pay employees fast and stay on top of filings. Get set up in a day with guided onboarding.”
Benefit bullets: “Fewer manual steps,” “Built-in checks,” “Support when you need it.”
CTA: “Book a demo.”
DRAFT B (Invented excerpt)
“The last payroll tool you’ll ever need. Northwind makes payroll 100% error-free and always compliant. Switch today and stop worrying.”
Benefit bullets: “Automated everything,” “Compliance handled,” “Instant setup.”
CTA: “See it in action.”
CONSTRAINTS (Invented)
1) Score first; do not rewrite until after scoring.
2) Every score must include 1-2 short quotes from the draft as evidence.
3) Call out any pass/fail gate failures explicitly.
4) Provide a final recommendation: choose A, choose B, or merge, with reasons tied to the brief.
Step 3: Use a comparison prompt that forces evidence and consistency
Your prompt should do three things: (1) lock the rubric, (2) require quotes as evidence, and (3) produce a standardized output so Draft A and Draft B are directly comparable.
Comparison prompt (copy/paste)
Prompt:
You are evaluating two landing page drafts against the same brief. Use ONLY the brief and rubric below as the standard. Do not invent product facts. Do not reward clever writing that violates the brief.
Inputs:
1) BRIEF: [paste]
2) RUBRIC (criteria, weights, pass/fail gates): [paste]
3) DRAFT A: [paste]
4) DRAFT B: [paste]
Output requirements:
A) Pass/Fail Gates: For each gate, mark Draft A and Draft B as Pass/Fail with a short quote as evidence.
B) Weighted Scorecard: For each rubric criterion, provide (i) score 0-5, (ii) weight, (iii) weighted score, (iv) 1-2 quotes as evidence, (v) a specific fix.
C) Section-by-Section Winner: Hero, benefits, how-it-works, proof, FAQ, CTA. Pick A/B/Tie and explain why using the brief.
D) Risks: List compliance/claim risks and ambiguity risks per draft.
E) Recommendation: Choose Draft A, Draft B, or Merge. If Merge, specify exactly which sections to take from which draft and what to rewrite.
F) Output in a clean table-like structure with consistent labels for A vs B.
Step 4: What a usable AI output should look like (example)
Invented example: This is a sample of the kind of output you want. Your real output will be longer and based on your full drafts.
A) Pass/Fail gates (invented example)
| Gate | Draft A | Draft B |
|---|---|---|
| Includes “Book a demo” CTA | Pass (“Book a demo.”) | Fail (“See it in action.”) |
| Includes “How it works” (3 steps) | Not shown in excerpt (check full draft) | Not shown in excerpt (check full draft) |
| No forbidden claims | Pass (no “guaranteed/100%/always”) | Fail (“100% error-free and always compliant.”) |
B) Weighted scorecard snippet (invented example)
| Criterion | Weight | Draft A (score + evidence + fix) | Draft B (score + evidence + fix) |
|---|---|---|---|
| Value proposition clarity | 2 |
4/5 Evidence: “Run payroll in minutes” / “pay employees fast” Fix: Specify the “under 15 minutes” promise from the brief and name the audience (“HR teams at 50-300 employees”). |
3/5 Evidence: “last payroll tool you’ll ever need” Fix: Replace vague superlative with the brief’s measurable promise; remove absolute claims. |
| Compliance / claim safety | 2 |
4/5 Evidence: “built-in checks” / “stay on top of filings” Fix: Add a clarifying line about what checks cover (without inventing specifics). |
0/5 Evidence: “100% error-free” / “always compliant” Fix: Remove forbidden absolutes; replace with qualified language aligned to the brief. |
C) Section-by-section winner (invented example)
- Hero: Draft A (closer to brief tone; fewer risky claims).
- Benefits: Tie (both need more specificity; Draft A is safer, Draft B is punchier).
- CTA: Draft A (matches “Book a demo” requirement).
Step 5: Ask AI for a “merge plan” that stays faithful to the brief
Once you have the scorecard, the next useful deliverable is a merge plan: which parts to keep, which to rewrite, and what to delete. This is where AI can save time without replacing your judgment.
Merge-plan prompt (copy/paste)
Prompt:
Using the brief as the source of truth and the evaluation you just produced, create a merged landing page draft. Constraints:
1) Keep only claims supported by the brief (do not invent proof).
2) Preserve the required sections and CTA wording.
3) Use the brand tone: confident, plainspoken, no hype.
4) Provide a change log: what you took from Draft A, what you took from Draft B, and what you rewrote.
Usable merged output structure
- Hero: Headline, subheadline, 3 bullets, primary CTA, secondary CTA (optional)
- How it works: Step 1, Step 2, Step 3
- Proof block: What proof exists (placeholders allowed), what proof is missing (flag)
- Objection handling: Migration, compliance, support
- FAQ: 3 questions aligned to objections
- Final CTA: Repeat “Book a demo” with a tight reassurance line
- Change log: A kept / B kept / rewritten / removed
Step 6: Review checks to run before you trust the comparison
AI can be consistent, but it can also miss context or reward surface-level polish. These checks help you validate the output and catch common failure modes.
1) Brief traceability check
- For every “recommendation” the model makes, ask: Which line of the brief does this map to?
- If it cannot map, treat it as optional editorial preference, not a requirement.
2) Evidence check (quotes are non-negotiable)
- Any score without a quote is easy for the model to hallucinate.
- Require short quotes for both drafts for each criterion, even if the quote is “missing” (e.g., “No proof section present”).
3) Claim safety check
- Scan for absolutes and implied guarantees (even if not on your forbidden list).
- Flag “instant,” “effortless,” “no risk,” “always,” “never,” “100%,” and unqualified outcomes.
- Confirm the draft does not imply certifications, partnerships, or results not in the brief.
4) Conversion logic check
- Does the first screen answer: What is it? Who is it for? Why now? What do I do next?
- Do headings help scanning, or are they clever but vague?
- Do CTAs match the brief’s goal (demo, trial, purchase, signup)?
5) Consistency check across sections
- Look for mismatched promises (hero says “15 minutes,” later says “in a day”).
- Ensure differentiators appear more than once: once in the hero/benefits, once near proof/objections.
Choosing an AI tool for this job (what matters, and what to avoid)
Because this workflow is about comparing two drafts against a fixed brief, the tool choice is less about “writing quality” and more about control and repeatability. When evaluating tools, look for:
- Structured outputs: Can it reliably follow a scorecard format with consistent labels?
- Long inputs: Can it handle your full brief plus two full drafts without truncation?
- Low creativity mode: Can you push it toward evaluation rather than rewriting?
- Easy iteration: Can you rerun the same rubric after edits to confirm improvement?
If you are comparing outputs from different models, keep the rubric and output format identical and run the same inputs through each model. That makes differences easier to interpret.
Common pitfalls (and how to prompt around them)
Pitfall: The model “picks a favorite” based on style
Fix: Add a constraint: “Do not reward cleverness; reward alignment to the brief. Penalize unsupported claims even if persuasive.”
Pitfall: It invents proof (testimonials, metrics, logos)
Fix: Add: “If proof is missing, say ‘missing’ and suggest what proof to request. Do not fabricate.”
Pitfall: It rewrites instead of evaluating
Fix: Force a two-pass workflow: “Pass 1: score only. Pass 2: merge draft only after scoring is complete.”
Pitfall: It ignores pass/fail requirements
Fix: Put gates first in the output and instruct: “If a gate fails, the recommendation must include a fix before selection.”
Frequently Asked Questions
FAQ 1: What should I paste into the AI to compare two landing page drafts fairly?
Answer: Paste (1) the brief, (2) a rubric derived from the brief (criteria, weights, and pass/fail gates), and (3) both drafts in full. Add constraints that require quoted evidence from each draft for every score and note.
Takeaway: The brief becomes the rubric, and the rubric forces a fair comparison.
FAQ 2: How do I stop AI from judging based on writing style instead of the brief?
Answer: Tell it to penalize anything that violates the brief even if it sounds persuasive, and require it to justify each score with direct quotes. Also separate evaluation from rewriting: “Score first, rewrite later.”
Takeaway: Evidence quotes and a two-pass workflow reduce style bias.
FAQ 3: What is the best output format for an AI landing page comparison?
Answer: Use a standardized scorecard: pass/fail gates first, then a weighted rubric table with scores, evidence quotes, and a specific fix per criterion, followed by a section-by-section winner and a final recommendation (A/B/merge).
Takeaway: A consistent structure makes the result decision-ready.
FAQ 4: How do I handle missing information in the brief when comparing drafts?
Answer: Add a rule: “If the brief does not specify it, mark as ‘not specified’ and do not assume.” Then ask the model to list clarifying questions (for you or stakeholders) and to score only what the brief actually defines.
Takeaway: Treat gaps as questions, not invitations to guess.
FAQ 5: Can AI help me merge the best parts of two drafts without inventing new claims?
Answer: Yes, if you constrain it: “Use only claims supported by the brief; if proof is missing, flag it as missing.” Ask for a change log that states what it took from Draft A, what it took from Draft B, and what it rewrote to match the brief and tone.
Takeaway: A merge plan plus a change log keeps the rewrite accountable.
FAQ 6: What review checks should I run after AI picks a “winner”?
Answer: Run (1) brief traceability (every recommendation maps to the brief), (2) evidence check (quotes support scores), (3) claim safety scan (no absolutes or invented proof), (4) conversion logic (first screen clarity + CTA alignment), and (5) consistency across sections (no conflicting promises).
Takeaway: Validate the reasoning, not just the conclusion.
FAQ 7: How do I compare two drafts if they have different page structures or sections?
Answer: Compare them on rubric criteria first (clarity, proof, compliance, conversion structure), then do a section mapping step: ask the model to label each paragraph by function (hero, benefits, objections, proof, CTA). That lets you compare “like with like” even when the order differs.
Takeaway: Evaluate by function, not by layout.
FAQ 8: Should I run the same comparison prompt multiple times?
Answer: It can help to rerun after you revise a draft, because you can see whether scores and gate checks improved. If repeated runs produce inconsistent judgments, tighten constraints (more gates, required quotes, stricter scoring definitions) and reduce ambiguity in the brief.
Takeaway: Re-running is most useful for before/after validation, not for “voting.”