--- name: edholofy/ad-copy-google-ads source: https://app.decimal.ai/s/edholofy-ad-copy-google-ads@10/SKILL.md source_sha256: f422a6483c19 --- ## Domain Knowledge ### Character Limits — The Hard Constraints - **Headlines: 30 characters** (including spaces, punctuation, em dashes, pipes — every character counts) - **Descriptions: 90 characters** - **Sitelink text: 25 characters** - **Sitelink descriptions: 35 characters** - **Callout extensions: 25 characters** - **Structured snippet values: 25 characters** - DKI default text + `{KeyWord:}` wrapper must fit within the headline's 30-char limit. The *longest keyword* in the ad group determines whether DKI is safe. **Critical: LLMs consistently miscount characters by 1-3.** Never estimate. Count every character deterministically using a reliable method. "Expert cosmetic dentistry" = 25 chars, not 28. Recount before finalizing. ### Google-Approved Structured Snippet Headers (Exhaustive List) Only these are valid: **Amenities, Brands, Courses, Degree programs, Destinations, Featured hotels, Insurance coverage, Models, Neighborhoods, Service catalog, Shows, Styles, Types.** Anything else (e.g., "Features," "Integrations," "Coverage Options") will be rejected by Google. ### Quality Score — The Three Levers 1. **Expected CTR**: Driven by headline relevance to query + compelling language. Specific numbers and benefits > generic claims. 2. **Ad Relevance**: Keyword must appear naturally in headlines/descriptions. Not keyword stuffing — semantic alignment. 3. **Landing Page Experience**: Ad promise must match landing page content. If the ad says "Free Trial," the landing page must show a free trial above the fold. ### DKI Syntax and Capitalization - `{KeyWord:Default Text}` — Title Case each word - `{Keyword:Default Text}` — Sentence case - `{keyword:default text}` — lowercase - `{KEYWORD:DEFAULT TEXT}` — ALL CAPS - Default text appears when the keyword is too long or would create policy violations. ### Search Intent → Copy Strategy Mapping | Intent | User Mindset | Headline Approach | CTA Style | |---|---|---|---| | Informational | Learning | Educate, guide | "Learn More," "Read the Guide" | | Commercial | Comparing | Differentiate, prove | "Compare Plans," "See Why We're #1" | | Transactional | Buying | Sell, convert | "Buy Now," "Start Free Trial" | | Navigational | Finding specific brand | Confirm identity, redirect | "Official Site," "Log In Here" | ### Non-Obvious Insights - **Pinning headlines reduces Google's optimization ability.** Only pin when brand compliance or messaging control outweighs performance. Pin position 1 for brand terms, position 2 for CTAs if required. - **Callouts should never repeat what's already in a typical headline.** "Free Shipping" as a callout is wasted if it also appears in headline rotation. - **Each RSA description must work independently with ANY headline combination.** Descriptions that depend on a specific headline for context will produce incoherent ads 70%+ of the time. - **Feature dumping kills CTR.** One specific benefit with a number ("Save 3 Hours/Week") beats listing five features. - **Legitimate urgency converts; fake urgency violates policy.** "Offer Ends March 15" is valid. "Last Chance!" running perpetually is a policy violation. - **CTR predictions should be grounded:** authority-based headlines (awards, stats) typically outperform emotional appeals for B2B; emotional and specificity approaches outperform for B2C impulse purchases. ## Quick Start **Task: Write a complete RSA for a SaaS project management tool targeting "project management software"** **Step 1 — Count characters BEFORE writing:** Plan each headline to land at 25-29 chars. Never rely on estimation. **Step 2 — Write 15 headlines with variety:** ``` Pin 1: "Project Management Software" (28 chars) — keyword match Pin 2: "Start Your Free Trial Today" (27 chars) — CTA : "Trusted by 10,000+ Teams" (24 chars) — social proof : "Cut Project Time by 40%" (23 chars) — specific benefit : "All Tasks in One Dashboard" (26 chars) — feature/benefit : "No Credit Card Required" (23 chars) — objection handler : "Award-Winning PM Platform" (25 chars) — authority : "Plans Starting at $9/Mo" (22 chars) — price anchor : "See Results in Week One" (23 chars) — speed : "Built for Remote Teams" (22 chars) — audience : "Gantt Charts & Kanban" (21 chars) — feature specific : "Replace Spreadsheet Chaos" (25 chars) — pain point : "Enterprise-Grade Security" (25 chars) — trust : "Integrate With 200+ Apps" (24 chars) — ecosystem : "5-Star Rated on G2 & Capterra" (29 chars) — social proof ``` **Step 3 — Write 4 descriptions (each independent, each under 90 chars):** ``` "Manage tasks, timelines & teams in one intuitive platform. Start free — no card needed." (88 chars) "Join 10,000+ teams who cut project delivery time by 40%. See the difference in one week." (89 chars) "Drag-and-drop Gantt charts, Kanban boards & real-time reporting. Plans from $9/month." (86 chars) "Rated 4.8/5 on G2. Integrates with Slack, Jira & 200+ tools. Try it free for 14 days." (87 chars) ``` **Step 4 — Verify every character count.** Recount each headline and description character by character. ## Core Rules 1. **ALWAYS count characters deterministically for every headline, description, and extension before including it in output.** LLM estimation fails consistently by 1-3 characters. Count spaces, punctuation, and special characters. Report the count alongside each line. 2. **ALWAYS produce a complete response for every task, even under uncertainty.** If the prompt is ambiguous, state assumptions and deliver full output. Zero output is the worst possible outcome. If a subtask is unclear, produce a best-effort version with a caveat rather than skipping it. 3. **ALWAYS validate structured snippet headers against the approved list** before including them: ``` IF header NOT IN [Amenities, Brands, Courses, Degree programs, Destinations, Featured hotels, Insurance coverage, Models, Neighborhoods, Service catalog, Shows, Styles, Types]: REJECT header SELECT closest valid alternative from approved list NOTE the substitution ``` 4. **ALWAYS validate DKI safety with this process:** ``` FOR each keyword in ad group: insertion = apply_capitalization(keyword, dki_format) full_headline = template.replace(dki_placeholder, insertion) char_count = count_characters(full_headline) IF char_count > 30: FLAG as overflow, default text will show IF insertion creates grammatical/semantic issues: FLAG PROVIDE at least 2 static non-DKI headline alternatives ``` 5. **Prefer specific, quantified claims over generic value propositions when the advertiser has real data.** "Save 3 Hours/Week" beats "Save Time." "Trusted by 10,000+ Teams" beats "Trusted by Many." If no data exists, use specificity of mechanism ("Drag-and-Drop Gantt Charts") rather than vague benefits ("Easy to Use"). 6. **ALWAYS differentiate ad groups by search intent, not just keywords.** Each ad group must target a distinct user need with messaging tailored to that need. If two ad groups could share the same ad copy, they should be merged. 7. **ALWAYS separate CTAs from urgency triggers.** CTA = action verb + specific outcome ("Start Free Trial," "Book a Demo"). Urgency = time/scarcity context ("Offer Ends March 15," "Only 3 Spots Left"). Combine them, but never confuse one for the other. ## Decision Tree **If writing headlines:** - If brand campaign → Pin brand name to position 1, use official brand terms - If non-brand campaign → Pin highest-intent keyword headline to position 1 - If CTA is required → Pin CTA headline to position 2 - If unsure about pinning → Don't pin; let Google optimize **If choosing DKI vs. static headlines:** - If all keywords in ad group are ≤20 chars → DKI is safe (leaves room for template text) - If any keyword > 22 chars → Test carefully; likely need static fallback - If keywords include competitor names, misspellings, or sensitive terms → Use static only - If ad group has >15 keywords with varied phrasing → DKI adds more risk than value; use static **If diagnosing Quality Score issues:** - If CTR is low but relevance is high → Headlines lack compelling hooks; add numbers, urgency, or social proof - If relevance is low → Keyword doesn't appear in headlines/descriptions; restructure ad groups for tighter theme - If landing page score is low → Check that ad promises match landing page H1, CTA, and above-fold content **If classifying search intent:** - If query contains "how to," "what is," "guide" → Informational - If query contains "best," "vs," "review," "top" → Commercial investigation - If query contains "buy," "price," "discount," "near me," "free trial" → Transactional - If query contains a brand name + generic term → Navigational - If query is ambiguous → Write copy that serves the most commercially valuable intent, with a secondary description covering the alternative intent **If writing extensions:** - If sitelinks → Each must link to a genuinely different page (not /features and /all-features) - If callouts → Each must add a benefit NOT already in typical headlines - If structured snippets → Select header from approved list first, then write values **If writing urgency/CTAs:** - If there's a real deadline → Use it explicitly: "Ends Dec 31" - If there's genuine scarcity → State it: "Only 50 Seats Available" - If no real urgency exists → Use value-based urgency: "Start Saving Today" (not fake countdowns) - If the CTA is "Click Here," "Submit," or "Learn More" for a transactional query → Replace with specific action: "Get Your Free Quote," "Start Building Now" ## Edge Cases ### Character Count Miscounts **Trap:** The agent writes "Best Noise Canceling Earbuds" and reports "30 chars." Actual count: 28. **Correct handling:** Count each character: B-e-s-t-[space]-N-o-i-s-e-[space]-C-a-n-c-e-l-i-n-g-[space]-E-a-r-b-u-d-s = 28. Always recount. Report the count next to each line item. ### DKI Overflow **Trap:** Template is `{KeyWord:Running Shoes}` (total 30 chars with longest keyword "lightweight waterproof running shoes"). DKI will overflow, but the agent doesn't flag it. **Correct handling:** Calculate: "Lightweight Waterproof Running Shoes" = 36 chars. This exceeds 30. Flag it. Note the default "Running Shoes" will display instead (which works at 13 chars in the template context). Provide 2 static alternatives. ### Non-Approved Structured Snippet Headers **Trap:** Agent uses "Features" or "Integrations" as a snippet header. **Correct handling:** Neither exists in Google's approved list. Map "Features" → "Types" or "Service catalog." Map "Integrations" → "Brands" (if listing integration partner brands). Always state the mapping rationale. ### Descriptions That Depend on Specific Headlines **Trap:** Headline: "Our Award-Winning Software" / Description: "It also comes with free support." ("It" has no clear referent if paired with a different headline like "Start Your Free Trial.") **Correct handling:** Each description must be self-contained. Rewrite: "Award-winning software with free 24/7 support. Try it risk-free today." ### Mixed-Intent Keywords **Trap:** "CRM software" could be informational (what is it?) or transactional (buy it). Agent writes purely transactional copy and alienates researchers. **Correct handling:** Pin a keyword-rich headline to P1 for relevance. Include one description that educates ("See How CRM Centralizes Your Sales Pipeline") and one that converts ("Start Your 14-Day Free Trial — No Card Needed"). ### Perpetual Urgency Violations **Trap:** Ad copy says "Last Chance — Sale Ends Soon!" but runs indefinitely with no actual deadline. **Correct handling:** This violates Google's misrepresentation policy. Replace with value-based urgency: "Start Saving on Your Plan Today" or tie to a real event: "Black Friday Pricing — Ends Nov 30." ### Ad Group Overlap **Trap:** Two ad groups — "project management tool" and "project management software" — with nearly identical copy. **Correct handling:** These should be one ad group. If kept separate, differentiate messaging: "tool" group emphasizes ease/simplicity, "software" group emphasizes enterprise features/integrations. ## Anti-Patterns **DON'T estimate character counts.** Instead, count every character deterministically and report the count inline (e.g., `"Start Free Trial Today" (22 chars)`). **DON'T return empty or partial responses when a task is complex.** Instead, state assumptions, break the task into parts, and deliver complete output for each part. Partial output with caveats > no output. **DON'T invent structured snippet headers.** Instead, reference the 13 approved headers and select the closest match. If none fit, use "Types" or "Service catalog" as the most flexible defaults. **DON'T write descriptions that only make sense with one specific headline.** Instead, read each description in isolation and ask: "Does this make complete sense paired with ANY of my 15 headlines?" **DON'T keyword-stuff headlines.** Instead, include the primary keyword naturally in 2-3 headlines maximum. Use the remaining 12-13 headlines for benefits, proof, CTAs, and differentiation. **DON'T use generic CTAs ("Click Here," "Learn More," "Submit") for transactional intent.** Instead, match the CTA to the conversion action: "Get Your Free Quote," "Book a Demo," "Start Your Trial." **DON'T list features without connecting them to outcomes.** Instead of "Gantt Charts, Kanban Boards, Reports" write "Plan Projects Visually With Drag-and-Drop Gantt Charts." **DON'T create fake urgency.** Instead, use legitimate deadlines, real scarcity, or value-based motivation. Fake countdown timers and perpetual "last chance" claims violate Google policy and erode trust. **DON'T provide DKI templates without testing every keyword in the ad group against the character limit.** Instead, list each keyword, compute the resulting headline length, and flag any that overflow or produce awkward copy.