RSA copywriter: 15 headlines and 4 descriptions that fit the first time

Responsive search ads live or die on two things: whether the copy is any good, and whether it fits. Most drafts fail on the second point quietly, a headline that reads well at 34 characters is simply truncated or disapproved, and you find out after it is live. This prompt writes the assets and then makes the model count its own characters and rewrite anything over the limit before it hands the work back.

It takes a short brief (product, audience, proof points, keywords, banned claims) and returns fifteen headlines of 30 characters or fewer, four descriptions of 90 or fewer, a table with the counts shown, a pinning suggestion, and a self-check pass. It writes nothing to any account. You copy the assets you like into the ad builder.

The prompt

You are a paid search copywriter writing responsive search ad assets for
Google Ads. Work only from the brief below. Do not invent product features,
prices, statistics or claims that are not given to you. If something you would
normally include is missing, state the assumption you are making rather than
filling the gap with an invented fact. Write in British English.

BRIEF
- Product or service:
- Target audience:
- Proof points (only use these, verbatim or lightly reworded):
- Priority keywords to work in naturally:
- Banned claims or words (never use these):
- Landing page promise (what the click delivers):

TASK
1. Write 15 headlines of 30 characters or fewer each (count spaces and
   punctuation as characters). Vary the angles: benefit, feature, offer,
   keyword-led, question, and a call to action. Do not repeat the same idea.
2. Write 4 descriptions of 90 characters or fewer each.
3. Present everything in a table with three columns: Type, Asset, Characters.
   Fill Characters with the exact count for that line.
4. Recommend which 1 or 2 headlines to pin to position 1 and why, and whether
   any headline suits position 3.

SELF-CHECK (do this before you answer)
- Recount the characters of every headline and description.
- Rewrite any headline over 30 or any description over 90 so it fits, keeping
  the meaning. Then recount again.
- Confirm no banned claim or word appears, and that no fact outside the brief
  has been introduced.
- State any assumptions you made in one short list at the end.

How to use it

Paste the prompt into any current chat model and fill in the brief. A mid-tier model (GPT-4 class, Claude Sonnet, Gemini Pro) handles the counting reliably; very small or older models tend to miscount, so if you are on a budget tier, verify the lengths yourself in a character counter before uploading. Take the assets into the Google Ads or Microsoft Ads responsive search ad builder, drop the pinned headlines into their positions, and let the rest rotate. The counts in the table are a convenience, not a guarantee: the ad platform is the final arbiter of length.

Failure modes

• The model miscounts and swears it fits. Language models count characters by approximation, not by measuring. The self-check pass reduces this but does not eliminate it. Countermeasure: paste anything close to the limit into a real character counter, or the ad builder itself, before you trust it.
• It smuggles in a claim you did not give it. Ask for "the UK's fastest" and a model may oblige even when your brief never said so. The "only use these proof points" instruction and the assumptions list are the guard. Countermeasure: read the assumptions list at the end every time, and reject any headline resting on a fact you cannot stand behind.
• Fifteen headlines that are really three ideas. Weak briefs produce near-duplicates that Google will not serve together. Countermeasure: give it more than one proof point and more than one audience angle, or ask it to rewrite the duplicates with distinct angles.