Search Headline Set Ad Creative

Search Headline Set

Shared quality bar: ../references/output-standard.md. All numbers cited here live in ../references/thresholds.md. This skill produces search assets for one ad group at a time, because pinning and duplication are only judgeable against a single keyword set.

Use this skill when

A Google Ads ad group needs its Responsive Search Ad headlines and descriptions written or rebuilt, or an existing asset set has weak Ad Strength, or nobody can say which headlines are pinned and why. Also use it when an ad group has fewer live headlines than coverage.rsa_headline_floor. Not for Meta or LinkedIn feed copy, which is paid-social-primary-text-ad-creative. Not for the opening seconds of a video, which is video-hook-lines-ad-creative. Not for rewriting one proven angle across funnel stages, which is funnel-stage-variants-ad-creative.

Required input

• Ad group name and its keyword list, with match types.
• The destination URL, and the text of its first heading and first paragraph.
• The offer as it exists today: price, trial, guarantee, delivery promise.
• Any claim the advertiser wants used, with its source.
• Legal or brand constraints: forbidden words, mandatory disclaimers.
• Existing asset set and Ad Strength, if one is live.

Analysis workflow

1. Group the keywords by the job the searcher is doing, not by match type. Each job becomes a claim family. A family with no keyword behind it does not get a headline.
2. Read the destination heading and first paragraph and record the claim, the offer and the call to action present there. Apply match.above_fold_rule and match.promise_pairs before writing a single line.
3. Draft headlines one claim family at a time, each within limits.rsa_headline_chars. Write past the count you need, then cut.
4. Run the duplication pass. Two headlines sharing their main noun and verb are one headline, per the note on coverage.rsa_headline_floor. Merge them and record the merge, so the set is not padded to look full.
5. Count distinct claims against coverage.rsa_headline_floor and total slots against limits.rsa_headline_slots. Report the gap as a number of missing claim families, not as a request for more words.
6. Write descriptions within limits.rsa_description_chars, each carrying one proof or one mechanism, never a restatement of a headline.
7. Decide pinning. Pin only what must appear on every impression: a legal line, a brand requirement, a price the page contradicts if absent. Everything else stays unpinned so the system can assemble.
8. Apply claims.evidence_rule and claims.superlative_rule to every line. Mark unsourced factual lines [claim: needs source]. Check claims.restricted_topics and route those to the platform policy page.

Decision rules

• Distinct claims below coverage.rsa_headline_floor and no further evidenced claim available: report the shortfall, do not invent a family to fill it.
• Pins in one position above limits.rsa_pin_ceiling: rewrite. Heavy pinning removes the combinations the format exists to test.
• A headline whose claim is absent from the destination above the fold: investigate the landing page, and hold the headline out of the shipped set.
• Two headlines inside one claim family with no wording difference that changes the promise: keep one, label the other a duplicate.
• Any superlative without a cited comparison: cut it, and say what evidence would let it back in.
• Any line this skill produced that carries [claim: needs source] is approval_needed, whatever else its verdict says, and the marker is never stripped to make the set look finished. A marked line is not shippable copy: it waits for the named source, or it is rewritten without the claim. A marked headline is reported outside the shipped set, not inside it with a footnote.

Output format

No verdict line. Return a headline table with columns: headline, character count, claim family, pinned position or unpinned, evidence label, decision label. Then a description table with the same evidence and decision columns. Then a short duplication list naming each merged pair. Then What this could not see, Missing data, Approval gates.

This skill writes assets rather than reading a test or a time series, so it issues no high confidence label and does not borrow the comparison family's conditions, which belong to a test nobody has run on this copy yet. Where a row carries a confidence label it is about the evidence behind a claim and never about the writing: a line whose source was produced and checked is ship, an unsourced one is approval_needed. test.min_runtime and test.volume_floor start to apply the day the set goes live, not while it is being written.

Practical example

Illustrative and fictional. Advertiser "Bramble PM", ad group "project management software for agencies", 14 keywords. The advertiser supplies 11 draft headlines, asks for 15, and asks for a claim that the tool has the fastest onboarding in the category.

Duplication pass: "Set Up In Minutes", "Fast Setup For Teams" and "Quick Onboarding" share their noun and verb, so those three merge into one claim family. Three further pairs merge the same way. Eleven drafts minus two from the trio and one from each pair leaves 6 distinct claims, below coverage.rsa_headline_floor. Two evidenced families come off the destination page, taking the set to 8. The remaining slots up to limits.rsa_headline_slots are left empty rather than padded.

Shipped headlines, with character counts:

• Agency Project Management (25)
• Built For Agency Teams (22)
• Onboard A Client In A Day (25)
• See Every Project Deadline (26)
• Stop Chasing Timesheets (23)
• One Place For Client Work (25)
• Try Bramble PM Free (19)
• Start Free, No Card Needed (26)

Description 1 (89): "Brief, plan and invoice client work in one place, then give the client a live view of it."

Declined, and reported outside the shipped set:

• "Fastest Onboarding In The Category" (34). Refused under claims.superlative_rule: no comparison was supplied, so it is not written in any form, including a softened "onboarding that is hard to beat". It also runs past limits.rsa_headline_chars, which is the smaller of the two problems. Unlocked by a named benchmark against named competitors.
• "Cuts Admin By 6 Hours A Week" (28) [claim: needs source]. The 6 hour figure arrived in conversation only, so the line is approval_needed and is not one of the 8. It joins them the day somebody names the dashboard behind it.

Pinning: one legal line pinned to description 2, nothing else pinned.

Guardrails

• Never pad an asset set with reworded duplicates to reach a slot count.
• Never write a claim the destination page does not carry above the fold.
• Never invent a price, a customer count, a rating or a comparison.
• Never soften an unevidenced claim into a vaguer one to get it past review.
• Do not upload, edit or pause anything in the account. Recommend only.
• Report every line declined and the reason, including lines the advertiser asked for by name.