ad API审查

Domain Knowledge

Character Limits — The Hard Constraints That Keep Failing

These are absolute. Count every character including spaces before outputting:

Critical: Count characters after DKI keyword substitution. If your longest keyword + surrounding text exceeds 30 chars in a headline, the default text fires — or worse, the ad breaks.

DKI — The Syntax That Must Be Exact

Format: {KeyWord:Default Text} — capitalization matters:

• {keyword:default} → all lowercase
• {Keyword:Default} → First word capitalized
• {KeyWord:Default} → Each Word Capitalized (most common for headlines)
• {KEYWORD:DEFAULT} → ALL CAPS (rarely appropriate)

DKI risks most people miss:

• Competitor names in keyword lists → your ad says "Buy [Competitor]" (policy violation)
• Misspelled keywords → misspelled ads
• Long-tail keywords exceeding character limits → default text fires unpredictably
• Grammatically awkward insertions: "Get {KeyWord:Services} Today" + keyword "plumber near me" → "Get Plumber Near Me Today"
• Always provide 2–3 static (non-DKI) headline variants as safety nets

Quality Score — The Three Levers

1. Expected CTR: Driven by headline specificity, numbers, strong CTAs. Fix: replace generic headlines with benefit-specific ones containing numbers.
2. Ad Relevance: Keyword must appear naturally in headline AND description. Fix: mirror the exact search intent language, not synonyms.
3. Landing Page Experience: Ad promise must match page content. Fix: if ad says "Free Trial," landing page must show free trial above the fold. Recommend specific landing page changes, not vague "improve your page."

Non-obvious: A 90% relevant ad with a mismatched landing page scores worse than a 70% relevant ad with a perfectly matched page. Landing page is the most underweighted factor in most people's thinking.

Search Intent → Copy Strategy Mapping

Mixed intent is real: "best CRM software" is commercial + transactional. Lead with comparison angle, include transactional CTA. Don't go so generic you convert nobody.

Urgency Psychology — What's Legitimate vs. Policy Violation

Legitimate urgency triggers:

• Real deadlines: "Offer Ends March 31" (must be true)
• Genuine scarcity: "Only 3 Spots Left" (must reflect real inventory)
• Value-based urgency: "Save $200 When You Start This Week"
• Loss aversion framing: "Don't Miss Your Q4 Window"

Policy violations (Google will reject):

• Fake countdown timers with no real deadline
• Perpetual "Last Chance" or "Final Sale" that never ends
• "Act Now Before It's Too Late" with no specific consequence
• Exclamation marks in headlines (Google disapproves multiple !!!)
• ALL CAPS words for fake urgency

Why each works (the psychology):

• Loss aversion: Losing $100 feels ~2x worse than gaining $100 (Kahneman). Frame savings as "Don't lose X" not just "Save X."
• FOMO: Social proof + scarcity combined ("Join 10,000+ Marketers — 50 Seats Left") outperforms either alone.
• Commitment bias: Low-barrier CTAs ("Start Free" vs "Buy Now") exploit that once started, people continue.

Headline Formulas That Actually Differentiate

Five genuinely distinct approaches (not surface rewording):

1. Problem-Solution: Name the pain → offer the fix. Best for: pain-aware audiences. "Stop Wasting Ad Spend | AI-Powered Optimization"
2. Authority/Social Proof: Credibility-first. Best for: skeptical/comparison shoppers. "Trusted by 50,000 Brands | #1 Rated Platform"
3. Specificity/Numbers: Concrete outcomes. Best for: analytical buyers. "Cut CPA 37% in 30 Days | See How"
4. Emotional/Aspirational: Identity and transformation. Best for: B2C, lifestyle. "Build the Business You Deserve"
5. Direct Offer: Lead with the deal. Best for: transactional, price-sensitive. "Plans from $29/mo | No Setup Fee"

For comparative analysis: Authority headlines typically show 15–25% higher CTR but lower conversion rates than direct offer headlines. Specificity headlines with real numbers outperform vague claims by 30–40% on CTR. These aren't arbitrary — they reflect that specificity filters intent (higher quality clicks) while broad emotional appeals cast wider nets.

Extension Copy — Distinct, Not Redundant

Sitelinks: Each must represent a genuinely different landing page destination. Bad: "Our Services" / "What We Offer" / "Our Solutions." Good: "Pricing Plans" / "Case Studies" / "Free Demo" / "Integration Docs."

Callouts: Must add value NOT already in headlines/descriptions. Think trust signals: "No Long-Term Contract" / "24/7 Live Support" / "SOC 2 Certified" / "Free Migration."

Structured snippets: Use Google-approved headers only: Amenities, Brands, Courses, Degree programs, Destinations, Featured hotels, Insurance coverage, Models, Neighborhoods, Service catalog, Shows, Styles, Types. Values must be specific items, not descriptions.

Common Mistakes — The Diagnostic Framework

Quick Start

Input: "Write RSA headlines for a project management SaaS targeting 'project management software for teams'"

Output (note: every headline is ≤30 chars, counted):

Pin 1 (keyword-rich):
H1: "Team Project Management Tool"    [30] ✓
H2: "Project Software for Teams"      [28] ✓

Pin 2 (benefit/differentiator):
H3: "Cut Project Delays by 40%"       [26] ✓
H4: "See All Projects in One View"    [29] ✓

Unpinned (variety):
H5:  "Trusted by 12,000+ Teams"       [24] ✓
H6:  "Start Free — No Card Needed"    [28] ✓
H7:  "Assign, Track & Deliver Fast"   [29] ✓
H8:  "Replace Spreadsheet Chaos"      [26] ✓
H9:  "Plans Starting at $9/User"      [26] ✓
H10: "4.8★ on G2 — See Reviews"       [24] ✓
H11: "#1 Rated by PCMag 2024"         [22] ✓
H12: "Onboard Your Team in 1 Day"     [27] ✓
H13: "Built-In Gantt & Kanban"        [23] ✓
H14: "Get Your Free 14-Day Trial"     [27] ✓
H15: "Integrates w/ Slack & Jira"     [27] ✓

Pinning rationale: H1/H2 pinned to Position 1 ensures keyword relevance for Quality Score. H3/H4 pinned to Position 2 ensures every combination leads with relevance and follows with a compelling reason. Remaining unpinned to maximize Google's optimization across combinations.

Character count verification: Every headline individually counted. Longest is 30 characters. ✓

Core Rules

1. ALWAYS count characters of every headline, description, and extension element before outputting. Display the count in brackets. This is the #1 failure mode — a single over-limit element invalidates the work.

2. When writing comparative analysis of ad formulas:

   • Ground CTR predictions in specific principles: "Specificity headlines with numbers typically achieve 15–25% higher CTR than generic equivalents because they pre-qualify clicks"
   • Never assign arbitrary percentages without stating the reasoning mechanism
   • Reference loss aversion, social proof, commitment bias, or specificity bias by name with brief explanation of mechanism

3. When using DKI:

   • List every keyword that will be inserted
   • Calculate max character count: longest keyword + surrounding static text
   • Flag any keyword that creates grammatical awkwardness, policy risk, or exceeds limits
   • Always include 2–3 non-DKI static headline alternatives

4. When writing urgency/CTA copy:

   • Every urgency claim must be verifiable or framed as conditional ("If you sign up this week" vs "Last chance ever")
   • Name the specific psychology principle powering each technique
   • Include at least one rejected example with explicit policy violation explanation

5. When writing descriptions, ensure each stands alone. Test: pair description D3 with headline H7 — does it make sense? Is there redundancy? Contradiction? Every combination must be coherent.

6. Prefer using 85–90 chars in descriptions over stopping at 50–60 chars. Short descriptions waste paid real estate. Fill the space with specific benefits, social proof, or objection-handling — not filler.

Decision Tree

• If task involves keyword insertion → check DKI syntax, verify every keyword fits within character limits after substitution, provide static fallbacks
• If task involves writing headlines → count every character, verify ≤30, ensure 5+ genuinely different angles (not surface rewording), specify pinning rationale
• If task involves extensions → verify sitelinks point to distinct pages, callouts don't repeat headline content, snippet headers are Google-approved
• If task involves intent matching → classify intent first, then write copy matching that stage, call out if intent is mixed and handle both
• If task involves fixing existing ads → diagnose the specific mistake by name, explain performance impact (CTR/QS/conversion), show before/after with the principle
• If task involves Quality Score → identify which of the 3 components is weakest, target fixes to that specific component, include landing page recommendations
• If task involves urgency → verify every claim is policy-compliant, name the psychology principle, include a rejected/non-compliant example for contrast
• If task asks for a "complete" ad build → deliver: keyword grouping → ad groups → full RSAs (15 headlines + 4 descriptions) → extensions (sitelinks, callouts, snippets) → DKI variants → pinning strategy → landing page recommendations → QS predictions with reasoning

Edge Cases

DKI with dangerous keywords

Trap: Client's keyword list includes competitor brand names, misspellings, or long-tail phrases. Agent uses DKI without auditing. Correct handling: Scan every keyword. Flag: competitor names (trademark policy), keywords >20 chars (leaving <10 for static text), misspellings, phrases that don't grammatically fit the template. Exclude those from DKI ad groups, serve them static ads instead.

Mixed search intent

Trap: "best accounting software free trial" — is this commercial (comparing "best") or transactional (wants "free trial")? Correct handling: Acknowledge mixed intent explicitly. Lead with commercial angle in H1 (comparison/authority), include transactional CTA in H2/description. Don't collapse to one intent.

Ad formula comparative analysis without data

Trap: Asked to predict CTR/conversion for different formulas, agent invents specific percentages (e.g., "this will get 4.2% CTR"). Correct handling: Provide relative comparisons grounded in principles: "Authority headlines typically outperform generic headlines by 15–25% on CTR based on social proof effects, but underperform direct-offer headlines on conversion rate because they attract broader audiences." Never claim exact CTR without test data.

Structured snippets with wrong headers

Trap: Using "Features" or "Benefits" as snippet headers — these aren't Google-approved. Correct handling: Only use: Amenities, Brands, Courses, Degree programs, Destinations, Featured hotels, Insurance coverage, Models, Neighborhoods, Service catalog, Shows, Styles, Types. If the business doesn't fit these, use "Types" or "Service catalog" as the most flexible options.

Callouts that duplicate headline content

Trap: Headline says "Free Shipping" and callout also says "Free Shipping." Correct handling: Callouts must add NEW information. If free shipping is in the headline, callout should cover something else: "No Restocking Fees" / "Easy 30-Day Returns."

Description character limit — the utilization trap

Trap: Writing descriptions of 40–50 characters that technically comply but waste 50% of available space. Correct handling: Target 75–90 characters. Fill with specific proof points, objection handlers, or secondary CTAs. "Get enterprise project management with real-time dashboards and Gantt charts. Start your free 14-day trial today." (90 chars) beats "Manage projects better. Try it free." (35 chars).

Anti-Patterns

• DON'T output ad copy without character counts in brackets. Instead, count and display [27] after every headline and description.
• DON'T write 15 headlines that are variations of the same message ("Great Software" / "Amazing Software" / "Excellent Software"). Instead, use 5+ distinct formula types: problem-solution, authority, specificity, emotional, direct offer.
• DON'T use vague CTAs like "Click Here" or "Learn More" as defaults. Instead, use conversion-specific verbs: "Start Free Trial" / "Get Your Quote" / "Book a Demo."
• DON'T claim urgency without a verifiable basis. Instead, frame conditionally: "Save 20% — Offer Ends [Date]" with a real date, or use value-based urgency: "Start saving this quarter."
• DON'T assign arbitrary CTR predictions like "this will get 3.7% CTR." Instead, provide relative performance predictions grounded in named principles: "Expect 20–30% CTR lift over generic copy based on specificity bias."
• DON'T recommend DKI without auditing the keyword list for overflow, policy risks, and grammar issues. Instead, list every keyword, calculate max character usage, and flag problems before writing the template.
• DON'T write sitelinks that point to conceptually identical pages. Instead, ensure each sitelink serves a distinct user need: pricing, proof, product details, getting started.
• DON'T diagnose ad problems as "could be better." Instead, name the specific mistake (keyword stuffing, feature dumping, intent mismatch), explain the performance consequence, and show the fix.
• DON'T skip landing page alignment when discussing Quality Score. Instead, specify what must appear on the landing page to match the ad's promise — "If the ad says 'Free 14-Day Trial,' the landing page needs a trial signup form above the fold